The RSX Conspiracy - RSX <-> G80 Connection ?

[wotter]

has anyone stopped to ask the obvious question. if ken went to IBM and asked for the most powerful cpu they can think of, would he not have gone to Nvidia with the same question?

But he did! 3 goals had to be achieved by Nvidia with RSX:
1. Built the most powerful GPU
2. Synthesize movie quality graphics
3. Take advantage of the bandwidth/power of Cell

[bluvi69]

Why does people keep relating DX10 to PS3? PS3 wont be using DX10 at all.

at least there still people understand common sense ,

PS3 will use openGL, if there is nextgen console that is benefit from dx9 or dx10 implementation that is only one console, and it is X360 with its xenos.

afterall it is msoft who invented directx variant, and it will benefit mostly to msoft xbox360 console.

if crysis team create 2 team for crysis nextgen console version
1st team : DX10 PC game into DX variant version of X360 (X360 xenos, is maybe we can called it DX9.5)
2nd team : DX10 PC game rewrite almost all engine into OpenGL equivalent implementation.

then which team will have better result and easy implementation?????

please dont make RSX like super GPU, it is almost 6 month old.

[archy141]

bluvi69, Hix

This DX10 compliance statement seems to be causing a little confusion.


You are stating If the RSX core was DX10 compliant i.e. G80 it would not serve any use to the PS3 as it does not utilise DX API.

Doesn't the fact that it is DX10 compliant just tell us it is uptodate & supports all of the latest bells & whistles straight out of hardware.

Couldn't the PS3 still take advantage of all these extra features thats are defined by Microsoft by just using a diferent API or even direct assembly if wanted to ?


Archy

[WASHIMUL]

THOSE SLIDES ARE MEANINGLESS

SONY was supposed to release the ps3 in spring 2006 according to that timeline

do u see the ps3 in the market?

NO

those slides are useless

[WASHIMUL]

Why does people keep relating DX10 to PS3? PS3 wont be using DX10 at all.

at least there still people understand common sense ,

PS3 will use openGL, if there is nextgen console that is benefit from dx9 or dx10 implementation that is only one console, and it is X360 with its xenos.

afterall it is msoft who invented directx variant, and it will benefit mostly to msoft xbox360 console.

if crysis team create 2 team for crysis nextgen console version
1st team : DX10 PC game into DX variant version of X360 (X360 xenos, is maybe we can called it DX9.5)
2nd team : DX10 PC game rewrite almost all engine into OpenGL equivalent implementation.

then which team will have better result and easy implementation?????

please dont make RSX like super GPU, it is almost 6 month old.

how do you know tht RSX is 6 month old when the devs of WARHAWK claimed that they received the FINAL version of RSX in JUNE

[archy141]

the devs of WARHAWK claimed that they received the FINAL version of RSX in JUNE


Do you have a link to reliable source that can confirm that statement ?

[Guilty Bystander]

I believe people on these forums have enough intellect to read & decide for themselves what is wrong & right. It’s not like I didn't make it clear from the starting & throughout the thread that all my reasoning is based on speculation. I have not tried to mislead anyone knowingly by providing false information or conclusions.

Also how can you insinuate my postings are so technical that they might confuse the not so knowledgeable ?


Well considering 49 out of a 130 people voted G80 certainly shows how much of a influence your post could have been on less knowledged forum users.
Besides turning away from the facts and carrying on with your assumptions and spreading this misinformation on almost every game forums out there in my eyes looks a lot like spreading propaganda.



I REPEAT my postings are all just pure speculation based on given reasoning & arguments that are expressed in basic laymen terms. There is no detailed technical talk from my behalf. In fact its you that comes ranting on this thread & coming out with numerical figures to compare against Xenos. If you & others wish to go off-track & get technical I can't control that.



Sure you are we'll just have to look at the thread title for that right?



Let me remind you

And me giving Sony a bad name ? What are you on ?


Stating the facts never hurts anyone.
However taking things out of your *** like you does hurt a brandname with gamers and hardcore gamers.



A while back everyone was very down in these forums due to all of the ongoing negative media coverage. I believe my thread has indirectly helped some people gather there thoughts again & understand what the PS3 truly is & why they were interested in it in the first place.


Yeah sure filling their heads with G80 dreams that's a good thing. :lol:



I think most people will agree if anything i have been most positive about Sony & its creation. As i have stated they are the only ones to give us a NextGen console - whatever RSX is. And my preorder is gladly placed.

You have made it very clear you love the XBOX360 very much. But that’s no excuse for you to go & try to put off others from debating matters relating to the PS3 .


So you're calling me a fanboy while you're stating such an obvious fanboy statement?
By the way we'll first have to wait and see if the PS3 really is any better than the Xbox 360 at all because most developers have been saying there's not much difference and if there is it goes in both directions.


Not once in this thread have you been straight up & said whether you agree or disagree with sensible reasoning to back it up. Maybe your head is too clouded by your love for XBOX360 & as a result you are not able to debate openly on matters relating to the PS3. Why don't you go & play something decent on XBOX360 instead ?

Fanboyism can even destroy the best of forums through non impartial opinions that do nothing other than justify there own personal interests & investments. THIS IS BECAUSE OF PEOPLE LIKE YOU!!!

Let me get this straight.
I'm not a fanboy for any system.
I have pre-ordered my PS3 well before the E3 of 2005.

Stating facts when they're negative towards the PS3 doesn't make you a fanboy.
Neglacting facts does make you one!

[archy141]

I have pre-ordered my PS3 well before the E3 of 2005

Thats really fresh ! All you ever do is put down the PS3 on these forums.

I don't have any console at the moment. In fact i have to confess i'm not a hard gamer- very much the casul one. Im buying the PS3 for all the other features it offers as well as ability to be a console. I did consider the Xbox360 but could not see any value that it would offer over having my PC upgraded. The way things stand with XBOX360 gaming i would instead be better off just getting a G80 for my PC. Non of this is fanboyism - just pure reasoning to what best suits my needs.

* Lets leave it at that & not mess up the thread please. I will not respond to any more comments that are personal from you


Archy

[mrboo]

This thread has gone from very funny to read to just sad

Why is it that 1 fanboys silly imagination can sucker in so many people?

[HiX]

We're speculating at best regarding RSX's performance. Xenos is a very good GPU, no doubt but since it's the first generation of it's architecture, it's not as fast as R580+ or R600. Granted, it is still capable of doing some of DX10 effects that R580 can't do.

I think it's safe to say RSX is faster and more powerful in terms of shader performance. Xbox fans may argue on that but the chances of it being superior is higher considering that it is not even out yet and already we've been seeing great-looking PS3 games on the horizon.

The details of Xenos has been set in stone. For RSX, it is not. From the look of things in the dev world, RSX is getting the nod for graphical superiority. We'll see how it is a few months from now.

[janenba352]

The slides that keep popping up even ill have to agree are old and meaningless. We have no idea if the specs could have changed or not. Theres a 50/50 chance that it changed or not. Lets wait for official word from Sony or Nvidea before we sit here argueing over the chances of the RSX being upgraded. Truly I do believe it was upgraded since Nvidea did release a more powerful architecture and they did promise Sony the most powerful video card so theres a darn good chance Nvidea kept it's promise.

[Jugix]

If nVidia can cram G80 to 300M transistors and pump out one million of them out in a month, then by all means we can have G80 in PS3. For now I think gf7900GTX is more believable alternative considering transistor count, mass production possibilities (old GPU) and devs comments about RSX.

I am glad if Sony suddenly proves me wrong! :wink:

[mrboo]

If nVidia can cram G80 to 300M transistors and pump out one million of them out in a month, then by all means we can have G80 in PS3. For now I think gf7900GTX is more believable alternative considering transistor count, mass production possibilities (old GPU) and devs comments about RSX.

I am glad if Sony suddenly proves me wrong! :wink:

Finally someone said it.

G80 will be atleast 2x the die space with perhaps 2x the transister's of RSX

[NUMA NUMA]

too... much.... tech.. jargin.... so confused :(

[janenba352]

If nVidia can cram G80 to 300M transistors and pump out one million of them out in a month, then by all means we can have G80 in PS3. For now I think gf7900GTX is more believable alternative considering transistor count, mass production possibilities (old GPU) and devs comments about RSX.

I am glad if Sony suddenly proves me wrong! :wink:

Doesnt mean anything. They could have changed the whole card for the PS3. Nvidea will deliver. Just a matter of believing. Even if its just a little stronger then the 7950GX2 id be happy.

[Knive]

Well maybe a new gfx card might explain the size increase that happened a few months back? They said they were redesigning the PS3, and all they did was increase the size....maybe for the better card?

[Guilty Bystander]

Doesnt mean anything. They could have changed the whole card for the PS3. Nvidea will deliver. Just a matter of believing. Even if its just a little stronger then the 7950GX2 id be happy.


7950GX2 is more than 560 million transistors.

[Aleman]

But he did! 3 goals had to be achieved by Nvidia with RSX:
1. Built the most powerful GPU


It was probably more like "build the most powerful GPU that will cost less than $X and consume less than Y watts and can fit into Z space."

[janenba352]

Doesnt mean anything. They could have changed the whole card for the PS3. Nvidea will deliver. Just a matter of believing. Even if its just a little stronger then the 7950GX2 id be happy.


7950GX2 is more than 560 million transistors.

I said a little stronger then the 7950GX2 and you do know the card is just two cards stuck together. So really each GPU has 280 transistors.

I almost garrentee that 300 transistors can make a strong GPU. Even if they have to overclock the darn thing they will make it the strongest card.

[Aleman]

You can't really judge a GPU's power by the transistor count.

In this case, you can guess about what Nvidia card the RSX might be similar to but it won't tell you how powerful it is.

[Organic_Shadow]

Oohhhhhh.... so transistor count defines what something is now? Just like GHz does with CPU's still huh. That's why my 2.0GHz Athlon64 3200+ (ancient socket 754) was spanking my room mate's P4 3.0GHz all over the place?

AMD invested money into coming up with ways to make the most per clock cycle, so I don't see why GPU manufacturers can't do the same with transistor count. ATi has come up with a way of making the most of the shaders-- unifying them. It may not be proven to be very effective yet but they are at least thinking differently instead of the same old "more is better" thinking that has caused so many of us to put up with loud, monstrous PC's or invest in silent water cooling.

[Lefein]

I'm going to have to side with Guilty Bystander. It's really not fair to ramp up peoples expectations like this. Heaven only knows how much a G80 GPU will cost when it finally does come out, and on top of that, NVidia hasn't even taped one out yet...

How in the hell is Sony supposed to redesign a processor that doesn't even exist to fit on a bus specification that is less than a year old? I respect what Sony does for hardware, but this is really over the edge.

[DarkChild]

Last I checked the point of this thread was to speculate, it's a lot better then sitting around here and not have anything to read about, so archy put out an idea, and clearly stated it was all speculation, last i checked speculation was unproven opinion. why the hell is everyone on here taking this as fact.

if people on here are unable to tell fact from opinion then we really have some dumb people on this board, and the reason i'm on this board is because i thought some were smart enough to tell fact from fiction.

Guilty Bystander, why are you so against someone's opinion, which he has stated numerous times as such. i for one find it more interesting to speculate then to just sit around and do nothing. that is all we've been doing since before E305, so why not now when there is no news coming out.

point is why don't we just discuss the thread topic rather then tell this person to stop what he is proposing. unless of course...you have other motives.(and the confusing others, give the kid some credit, he's convincing people of a possibility, you've been trying to disprove him the entire time with no success, he must be doing something right.)

[hoverbike]

This is the first time I read this topic and I think that the RSX will be more powerful but I doubt it would be in the same class as the G80. That would be so insane and great.

That guy's points are good but it's not to be taken as what's going to happen. I'm not going to get my hopes up about this but I will keep it in mind.

[jlippone]

You can't really judge a GPU's power by the transistor count.

In this case, you can guess about what Nvidia card the RSX might be similar to but it won't tell you how powerful it is.
NVidia made that one clear when they moved from NV40 to G70 (N47?).
Transistor count went down by over 20 million and the chip was quite a bit faster.

[GunTeng]

lol, keep your fingers crossed. At least the RSX will be more powerfull than a GeForce 400 MX :)
you can make more out of it but sadly i have to leave now... ;)

Is that supposed to make people feel good about the RSX? :lol:
I have a Geforce 400MX & it can't run games that were released in the last 5 years at least.:(

More powerful?...it had better be.:lol:

[panson feo]

Last I checked the point of this thread was to speculate, it's a lot better then sitting around here and not have anything to read about, so archy put out an idea, and clearly stated it was all speculation, last i checked speculation was unproven opinion. why the hell is everyone on here taking this as fact.

if people on here are unable to tell fact from opinion then we really have some dumb people on this board, and the reason i'm on this board is because i thought some were smart enough to tell fact from fiction.

Guilty Bystander, why are you so against someone's opinion, which he has stated numerous times as such. i for one find it more interesting to speculate then to just sit around and do nothing. that is all we've been doing since before E305, so why not now when there is no news coming out.

point is why don't we just discuss the thread topic rather then tell this person to stop what he is proposing. unless of course...you have other motives.(and the confusing others, give the kid some credit, he's convincing people of a possibility, you've been trying to disprove him the entire time with no success, he must be doing something right.)
The thing is, facts are being shoved in this guys face yet he just ignores them. Alot is known about the RSX. If you want more info just go read B3D once in a while, they know alot about RSX.

[archy141]

Last I checked the point of this thread was to speculate, it's a lot better then sitting around here and not have anything to read about, so archy put out an idea, and clearly stated it was all speculation, last i checked speculation was unproven opinion. why the hell is everyone on here taking this as fact.

if people on here are unable to tell fact from opinion then we really have some dumb people on this board, and the reason i'm on this board is because i thought some were smart enough to tell fact from fiction.

Guilty Bystander, why are you so against someone's opinion, which he has stated numerous times as such. i for one find it more interesting to speculate then to just sit around and do nothing. that is all we've been doing since before E305, so why not now when there is no news coming out.

point is why don't we just discuss the thread topic rather then tell this person to stop what he is proposing. unless of course...you have other motives.(and the confusing others, give the kid some credit, he's convincing people of a possibility, you've been trying to disprove him the entire time with no success, he must be doing something right.)


Thanks for saying what i could not myself. You have showed more sense than some of the other members that i expected of. You are hitting it DEAD RIGHT on the head. What is wrong with all you people ?

All i'm giving is basic logic & reasoning to something that we have all possibly overlooked ?

I will personally request the closure of this thread as soon as we have solid proof that dismisses this argument.

Just to repeat again. If the RSX is nothing more than the original spec's it will not bother me at all. I'm not a hardcore gamer but more of a technology enthusiast. I look forward to playing the odd AAA game, and utilising all the other Multimedia features to there fullest. As a result it has a lot more value to me than say XBOX360 -better to stick in a G80 in my PC.
And thats not a fanboy comment before anyone jumps on me - just saying i look to buy what meets my personal needs.

It seems to me some people are getting scared by my posting as they feel I’m setting them up for a fall. Remeber its just an OPINION based on SPECULATION with reasoning. I'm glad I did my research & made the posting because whatever happens the anomalies I pointed out are true & exist. Whether Sony follows the original laid out plans or pull a surprise like the Motion controller remains to be seen.

Can anyone remember how many Developers new forehand about the Motion Controller change before E3 ?

Z E R O !

Warhawk devs were given the motion controller last minute to allow Sony to demonstrate it. Key thing to note here is even SCE developers were kept in the dark till last minute. Sony’s ability to keep secrecy on PS3 hardware is extremely high. They will definitely not put secrets in ANY developers hands if they have marketing value. Even Kojima was out of the picture with the motion controller –some may say disrespect & arrogance to the great man. But thats coperate business.

However much I respect the Beyond3D developer community for there talents, I take there knowledge on Final RSX spec’s with a pinch of salt. I will not waste my time with them –closed minds. Also there is the NDA matter. There is no way anyone would come forth on an open public forum.
Some people laughed at the response to my posting on Beyond3D.
I have no regrets what so ever & did not expect much more from there. I just wanted to put my theory out myself before anyone else did for me.

I'm no tech guru of any sort. My reasoning is simple logic. However passionate i am i do not allow it to cloud my judgement or reasoning.



If everyone waited to have definitive specs -Their wouldn't be much need to speculate nor talk about since the specs are right there.

Speculation breaths a lot of life into threads, with newer information seeping in little by little to help direct discussion.




Archy

[panson feo]

I'm no tech guru of any sort. My reasoning is simple logic. However passionate i am i do not allow it to cloud my judgement or reasoning.

Speculation breaths a lot of life into threads, with newer information seeping in little by little to help direct discussion. [/size][/color]

Explain this then.


IGN: What is it like to program and design the game for Xbox 360 versus the PlayStation 3? We're not looking to start a blood war here. We're more interested in what it's like to program for each, and what each one's strengths and weaknesses seem to be. Also, how do you foresee the two versions differing in the final product visually, gameplay-wise, and sound wise?

Magnus Högdahl: The PS3 will have a content size advantage with Blu-ray and a CPU advantage for titles that are able to utilize a lot of the SPUs.The Xbox 360 has a slight GPU advantage and its general purpose triple-core CPU is relatively easy to utilize compared to SPUs. I expect that it will be near impossible to tell Xbox 360 and PS3 screenshots apart.

http://ps3.ign.com/articles/704/704524p2.html
If the RSX is G80 based there is no way a dev would say XENOS has the advantage.

[Hexadecimal]

How can you say that? He didn't even mention what the advantage was.

[Jugix]

If everyone waited to have definitive specs -Their wouldn't be much need to speculate nor talk about since the specs are right there.

Speculation breaths a lot of life into threads, with newer information seeping in little by little to help direct discussion.




I bet when we get our eyes on offical RSX specs, the speculation only boosts from there! Xenos specs are known for ages now but still people are speculating is it faster than Radeon X800 or not! ;)

[PS3 rocks]

How in the hell is Sony supposed to redesign a processor that doesn't even exist to fit on a bus specification that is less than a year old? I respect what Sony does for hardware, but this is really over the edge.

Yeah thats a good point, sony is good but not this good. I'm betting it's a modified 7900.

[archy141]

http://ps3.ign.com/articles/704/704524p2.html
If the RSX is G80 based there is no way a dev would say XENOS has the advantage.


Did you you actually read ALL of what i wrote ?

Certainly doessn't look you understood it.



Can anyone remember how many Developers new forehand about the Motion Controller change before E3 ?

Z E R O !

Warhawk devs were given the motion controller last minute to allow Sony to demonstrate it. Key thing to note here is even SCE developers were kept in the dark till last minute. Sony’s ability to keep secrecy on PS3 hardware is extremely high. They will definitely not put secrets in ANY developers hands if they have marketing value. Even Kojima was out of the picture with the motion controller –some may say disrespect & arrogance to the great man. But thats coperate business.


I don't want really to repeat -but if you could not grasp it I will try to break it down further just for you.

What i was educating all is that Sony has shown that they are successfully able to keep hardware secrets from everyone including there own SCE developers. Hardware secrets have strategic marketing value where time is the essence of the matter. We are not 100% sure that any of the developers out there are using the final RSX hardware in there Kits.
So the opinions you are hearing in the media do not necessary reflect what MIGHT be the case later on. I hope that is clear enough.

To iterate again. We don't necessaily have to end up with G80.


It is probable that the RSX is G70 (N47) based but it is customised & shares the features of G80 -not necessarily all. A Hybrid somewhere between the two.

I wish we had the BANGING HEAD AGAINST WALL EMOTICON.

*PS3 Rocks*Lefein* The G80 & RSX developments are for complety different architecture platforms - G80 for the crippled PC architecture. The RSX would only need to be a subset of the G80 architecture to get the same out of PS3's refined architecture. It would most likely than be smaller & easier to produce. BTW in case you missed it G80 is supposed to have TAPED OUT a while back, in 90nm scale production & rumoured to be out sometime after september. That is a PC GFX card we are talking about. Not an integrated chip solution like RSX most likely will be. Would be a great oppertunity for Nvidia/Sony to do dual marketing.



Archy

[Guilty Bystander]

I said a little stronger then the 7950GX2 and you do know the card is just two cards stuck together. So really each GPU has 280 transistors.

I almost garrentee that 300 transistors can make a strong GPU. Even if they have to overclock the darn thing they will make it the strongest card.


That's not true because the 7950GX2 has 48 Texture units, 48 Pixel Shaders, 16 Vertex Shaders and 2 256bit memory interfaces.
No way the RSX can get close to that.
Don't even get me started on Quad SLI.


I bet when we get our eyes on offical RSX specs, the speculation only boosts from there! Xenos specs are known for ages now but still people are speculating is it faster than Radeon X800 or not!

Wow wow hold your horses there!
Don't think Xenos is the same as a X800 just because their fillrates are the same because there's more to it than that.
The Xenos has more RAW power than a X1800XT but less than a X1900XTX but in shader complexity and executions the Xenos is a lot better than the X1900XTX.
A X1800XT only does 272Flops (192Flops Pixel and 80Flops Vertex) per clock x 650MHz = 176,8GFlops, a X1900XTX does 656Flops (576Flops Pixel and 80Flops Vertex) per clock x 650MHz = 426,4GFlops.
The Xenos does 480Flops (48 ALU's x 10 instructions) per clock x 500MHz = 240GFlops.
A X850XT PE only does 156Flops (96Flops Pixel and 60Flops Vertex) x 540MHz = 84,24GFlops.

RSX as that states does 384Flops per clock x 550MHz = 211,2GFlops.
RSX vs Xenos would be 211,2GFlops vs 240GFlops.

[GunTeng]

From what I can see is that everyone has gotten into a rut to attempt to prove points of view & re-iterate what they feel has not been understood.

Time to move on people.
You've had your say & its time to move on.

As I suggested ( not clearly however;)) before, it would be reasonable to assume that there is more going on behind the scenes with SONY than anyone here can guess at.
We can hope for the best of the best & the next generation graphics chips before PC's get the equivelent. But until more info is released & reality kicks in, we should be allowed our dreams. It is our choice & we choose to dream with our own free will.
If you choose to believe it is as insubstantial as a dream that will never materialise, then that is your choice. No one has the right to judge someone elses hopes & dreams, nor have they the right to destroy them.

I hope for the best, but at the same time realise it may be asking too much. So at best I will be elated over the reality , but I will not be disappointed with it reguardless which reality turns out to be true.

Last point: If you don't believe in these assumptions, far fetched dreams, or the possibilty that SONY has more positive elements for the PS3 up their sleeves, then state an arguement against.
State another possibility for their quietness, coincidences where timing & tech are concerned.
BUT never point the finger at people and ACCUSE them of wrong doing where none exist. The only wrong-doing in this thread are the people who are pointing the finger & attempting to put blame onto the shoulders of another.

If someone takes the thread wrongly, then that is their fault for not reading it in the context it was so clearly posted.
That is their choice.
If they moan at SONY, that is their choice.
In that case, they would be in the wrong just as they were with the PS2.

Thats all folks :)

[Jugix]

I bet when we get our eyes on offical RSX specs, the speculation only boosts from there! Xenos specs are known for ages now but still people are speculating is it faster than Radeon X800 or not!

Wow wow hold your horses there!

RSX vs Xenos would be 211,2GFlops vs 240GFlops.

You did notice my WINKY-WINKY smiley at the end of the sentence!? ;) So if FLOPS are the key factor in GPUs power, wouldn't that make a Cell as good or almost better than todays high-end GPUs aka. RSX and Xenos in GPU functions?

PS3 Cell theoretical max: 218GFLOPS
RSX pixel shader max: 211GFLOPS
Xenos 48ALU max: 240GFLOPS

Wow, cell is insane! No wonder devs are using it as a pixel/vertex shader! :shock:

[archy141]

You can't really judge a GPU's power by the transistor count.

In this case, you can guess about what Nvidia card the RSX might be similar to but it won't tell you how powerful it is.

Aleman is absolutly correct. Let me share something very interesting that i came across:


Keep in mind that USA require more complex GPU and *much* more transistors, lower clocks, bigger temperatures and less processing power than fixed (classic) architecture. So R600 64 shader units is not faster than G80 48 (16+32).
http://www.nvnews.net/vbulletin/showthread.php?t=71762&page=6

USA=Unified Shader Architecture - as utilised in Xenos

I can't vouch for his expertise on the matter.



Archy

[Shred0r]

[...]besides the fact that first party devs had their final dev kits before E3.
http://www.beyond3d.com/forum/showthread.php?t=29579

Wouldn't put my money on that one...

Phil Harrison @ GDC '06:


Final versions of their SDK kits will ship to developers in June.
Although this is only software not hardware related...

www.firingsquad.com/news/newsarticle.asp%3Fsearchid%3D9485+final+sdk+ps3+in somniac&hl=de&gl=de&ct=clnk&cd=5&client=firefox-a (http://66.102.9.104/search?q=cache:WQEo9-2iop0J:

now here's another info on the relase of the "final environment":


The SDK will be delivered to devs. next week with the second release in April and the final production environment available in September.

final production environment = final hardware, final SDK, final online service ???


Middleware: Epic Unreal Engine, PhysX, HAVOK, all optimized for PS3. SN Systems ProDG Debugger, Tuner, Compiler, etc. Final technology delivered to developers by E3.

http://66.102.9.104/search?q=cache:lcm2s5_UvFMJ:www.joystiq.com/2006/03/22/gdc-the-ps3-keynote-blow-by-blow/+final+sdk+ps3+insomniac&hl=de&gl=de&ct=clnk&cd=2&client=firefox-a GDC keynote][/url]


PlayStation 3 development units, he said, are "absolutely on track," with final hardware being shipped to developers in June.

http://66.102.9.104/search?q=cache:Wmu5iGLKfjAJ:www.gamasutra.com/php-bin/news_index.php%3Fstory%3D8628+gdc+final+hardware&hl=de&gl=de&ct=clnk&cd=5&client=firefox-a

Furthermore Ted Price, the CEO of insomniac, if i am right, said in an interview at the time of GDC that they just have received the final hardware. So in this case not before E³ 06...

But on the other side there is one interesting comment on the RSX being already used at the time of the GDC 06 for hardware rendering:


What we have above is a clip from the developer demo of Warhawk for the PS3. A voiceover describes the "ambient warfare" to be used in the air-combat simulation game, as well as the innovative mix of "Cell-based software rendering with RSX-based hardware rendering" to produce the graphics. For all you coders out there, it was also interesting to note that the programs running on the Cell's SPUs were not written in low-level assembly (easing development, but possibly degrading performance).

http://66.102.9.104/search?q=cache:HDkCeY5sXH8J:www.joystiq.com/category/gdc/+gdc+final+hardware&hl=de&gl=de&ct=clnk&cd=4&client=firefox-a

So my guess is, that the first party developers (Sony owned studios) got their final PS3 hardware some time before GDC 06 BUT after E³ 06.
Whereas the 2nd and 3rd party developer studios like Insomnaic got it a bit delayed at the time of GDC...

The real question is when Nvidia taped out their DX10 hardware for the first time and if theres a practical possibility that they have implemented the DX10 parts into the RSX before the devs got their final PS3 units before GDC/after E³...

P.S: How the hell was this html code to shorten an url link, cause its a bit annoying to have such long links...

[Guilty Bystander]

You did notice my WINKY-WINKY smiley at the end of the sentence!? So if FLOPS are the key factor in GPUs power, wouldn't that make a Cell as good or almost better than todays high-end GPUs aka. RSX and Xenos in GPU functions?

PS3 Cell theoretical max: 218GFLOPS
RSX pixel shader max: 211GFLOPS
Xenos 48ALU max: 240GFLOPS

Wow, cell is insane! No wonder devs are using it as a pixel/vertex shader!


Actually no because those figures for RSX and Xenos are programmable GFlops whereas Cell's are fixed functions.
Xenos would do about 2TFlops fixed functions and the RSX about 1,8TFlops (hey sounds familiar right :P)
Those GFlops for the RSX are the whole package (Vertex and Pixel shaders) not just Pixel shaders.

A GeForce FX5600/5700 Ultra would do about 200GFlops fixed functions so you can look at Cell as one of them in realtime rendering.
Cell can be great at Pre and/or Post processing with 7800/7900 like power.


Keep in mind that USA require more complex GPU and *much* more transistors, lower clocks, bigger temperatures and less processing power than fixed (classic) architecture. So R600 64 shader units is not faster than G80 48 (16+32).

Actually Archy that guy is totally wrong.
A unified shader architecture requires less transistors for the same results as with fixed shaders
With a unified shader architecture Pixel shaders only require 1 FMADD instead of 1 FMADD + 1 FADD.
Thus requiring only 1 ALU for a Pixel shader instead of 2 ALU's.
With a 400-500 million transistor GPU like the G80/R600 you can have lots of unified shader ALU's and thus lots of Vertex or Pixel shaders.
Don't get me wrong it's not like you can have twice the shaders with the same amount of transistors but you can have more.

[archy141]

Off topic how can i get hold of a MOD ?

[xen]

You did notice my WINKY-WINKY smiley at the end of the sentence!? So if FLOPS are the key factor in GPUs power, wouldn't that make a Cell as good or almost better than todays high-end GPUs aka. RSX and Xenos in GPU functions?

PS3 Cell theoretical max: 218GFLOPS
RSX pixel shader max: 211GFLOPS
Xenos 48ALU max: 240GFLOPS

Wow, cell is insane! No wonder devs are using it as a pixel/vertex shader!


Actually no because those figures for RSX and Xenos are programmable GFlops whereas Cell's are fixed functions.
Xenos would do about 2TFlops fixed functions and the RSX about 1,8TFlops (hey sounds familiar right :P)
Those GFlops for the RSX are the whole package (Vertex and Pixel shaders) not just Pixel shaders.

A GeForce FX5600/5700 Ultra would do about 200GFlops fixed functions so you can look at Cell as one of them in realtime rendering.
Cell can be great at Pre and/or Post processing with 7800/7900 like power.


Keep in mind that USA require more complex GPU and *much* more transistors, lower clocks, bigger temperatures and less processing power than fixed (classic) architecture. So R600 64 shader units is not faster than G80 48 (16+32).

Actually Archy that guy is totally wrong.
A unified shader architecture requires less transistors for the same results as with fixed shaders
With a unified shader architecture Pixel shaders only require 1 FMADD instead of 1 FMADD + 1 FADD.
Thus requiring only 1 ALU for a Pixel shader instead of 2 ALU's.
With a 400-500 million transistor GPU like the G80/R600 you can have lots of unified shader ALU's and thus lots of Vertex or Pixel shaders.
Don't get me wrong it's not like you can have twice the shaders with the same amount of transistors but you can have more.


Sorry what?!!

The theorectical GFLOPS for Cell are also programmable GFLOPS. Meaning you can use these Gflops for whatever the programmer chooses.

And Xenos's total GFLOPS including fixed function is below a 1TFLOP.

Remember Microsoft officially rates Xbox360 at 1 TFLOP combined system performance (GPU GFlops + CPU 115 GFlops)

RSX is 1.8 TFLOPS on it own. Then add Cell gives you the 2 TFLOPS.

Total

Playstation 3
CPU 218GFLOPS + RSX 1.8 TFLOPS = 2+ TFLOPS

XBOX 360
CPU 115GFLOPS + XENOS (885? GFLOPS) = 1 TFLOP

[Aleman]

I don't see how the RSX could be 1.8 TFLOPS. That doesn't make any sense considering I think I saw that the 7800GTX was ~300 GFLOPS somewhere. I have a hard time believing that the RSX can do > 5x the FLOPS than a 7800GTX.

[Jugix]

Guilty, what does fixed function mean in a general purpose processor as Cell? :roll: As far as I have studied Cell architecture and programming, pretty much all of those "200GFLOPS" are at programmers disposal any time they want. Just cram T&L vertex code down the SPE pipe and you got yourself a gf5600 killer right there in one SPE.

[xen]

I don't see how the RSX could be 1.8 TFLOPS. That doesn't make any sense considering I think I saw that the 7800GTX was ~300 GFLOPS somewhere. I have a hard time believing that the RSX can do > 5x the FLOPS than a 7800GTX.

Those 300 GFlops for a 7800GTX are the programmable flops. Its 1.3 TFlops combining all fixed function hardware Flops.

RSX is 1.8 TFlops combined (programmable + fixed function)

Now the big money question is how does RSX gain another 500 GFlops over a 7800GTX?

[WASHIMUL]

It is NVIDIA that stated the 1.5 gflops figure for 7800GTX...........later some students at stanford found the practical figure to be 1.3 gflops

NVIDIA themselves have stated the 1.8TF figure for RSX so i think the practical count would be close to that margin

as for XBOX 360 the total flop count is 1 (115 gflops for xenon + rem for XENOS)

however this is a deflated figure..............the total practical gflops count for xenon is 77gflops.............much lower than 115 claimed not by IBM but by MS

[1080p]

I don't get this argument!
it is officially announced as Geforce 7900, they said its a modified 7800 before the release of the 7900 series!
Why you guys arguing?!
Even the spec sheet shows its a 7900! Too bad its going to be an out dated GPU when it is released :\

[xen]

I don't get this argument!
it is officially announced as Geforce 7900, they said its a modified 7800 before the release of the 7900 series!
Why you guys arguing?!
Even the spec sheet shows its a 7900! Too bad its going to be an out dated GPU when it is released :\

The spec sheet also says the RSX is 1.8 TFlops

So that 7900 series GPU gained some power since going into PS3

Between 300-500 GFlops extra infact.

For comparison the Dell quad SLI 7800GTX 512 provides 5.2 TFlops combined.

http://www.pcper.com/article.php?aid=195

Now dividing by 4 measures each card equaling 1.3 TFlops

So.

According to Nvidia a 7800GTX 512 = 1.3 TFlops
According to Nvidia the RSX = 1.8 TFlops

So to stop all the arguing please explain how RSX has gained these extra GFlops? :wink:

[Aleman]

I don't think you can just divide Quad SLI by 4 to get the FLOPS number for one card. SLI performance doesn't scale linearly.

[xen]

Thats right the REAL WORLD performance doesnt scale linearly.

But Theoretical marketing figures sure do.

[HighPs3]

Mmmm flip flops....

Anyway I doubt its a brand new chip but Im with you it would be great but just not going to happen in my book. Be happy with the cell paired with the 7800 whatever. Cross my fingers though your idea sounds a lot cooler. Just to cool.

[janenba352]

I said a little stronger then the 7950GX2 and you do know the card is just two cards stuck together. So really each GPU has 280 transistors.

I almost garrentee that 300 transistors can make a strong GPU. Even if they have to overclock the darn thing they will make it the strongest card.


That's not true because the 7950GX2 has 48 Texture units, 48 Pixel Shaders, 16 Vertex Shaders and 2 256bit memory interfaces.
No way the RSX can get close to that.
Don't even get me started on Quad SLI.


I bet when we get our eyes on offical RSX specs, the speculation only boosts from there! Xenos specs are known for ages now but still people are speculating is it faster than Radeon X800 or not!

Wow wow hold your horses there!
Don't think Xenos is the same as a X800 just because their fillrates are the same because there's more to it than that.
The Xenos has more RAW power than a X1800XT but less than a X1900XTX but in shader complexity and executions the Xenos is a lot better than the X1900XTX.
A X1800XT only does 272Flops (192Flops Pixel and 80Flops Vertex) per clock x 650MHz = 176,8GFlops, a X1900XTX does 656Flops (576Flops Pixel and 80Flops Vertex) per clock x 650MHz = 426,4GFlops.
The Xenos does 480Flops (48 ALU's x 10 instructions) per clock x 500MHz = 240GFlops.
A X850XT PE only does 156Flops (96Flops Pixel and 60Flops Vertex) x 540MHz = 84,24GFlops.

RSX as that states does 384Flops per clock x 550MHz = 211,2GFlops.
RSX vs Xenos would be 211,2GFlops vs 240GFlops.

Why are you starting with the SLI and quad SLI crap.. we all know we dream of the RSX in SLI or Quad SLI it just isnt going to happen ever.

Anyways Do you know the exact specs of the Final RSX? If you do please tell me. Otherwise your going by old info. Like I said lets wait for the official announcement of the card's specs.

[xen]

janenba352, i know you didnt quote me in your post but i did mention the quad SLI set up only to try and gain some form of performance comparison of each card (in the quad) then compare a single cards theorectical power to the the RSX theorectical figures.

From Nvidia Quad SLI 7800 = 5.2 TFlops of performance
Now this makes each card = 1.3 TFlops of performance

I know its not a scientific comparison and if anyone else could point me to a better breakdown of the 7800GTX (programmable + fixed) combined theorectial figures i'd appreciate it.
(Ive looked but Nvidia hasnt provided that information)

I think this is very important as everybody seems so eager to label the PS3 with just a bog standard card (7800/7900) without looking at the cards theorectical capabilities.

It's really simple what i'm trying to do

So for now we have 7800GTX at 1.3 TFlops

And RSX at 1.8 TFlops

Again how did the RSX gain 500 GFlops extra with a 550Mhz clock?

[janenba352]

Again how did the RSX gain 500 GFlops extra with a 550Mhz clock?

Hmm thats a good point. It wouldnt make much sence if its that slow of a clock to be getting that. Its faster then the specs stated. Its just that we have no clue what the architexture is.

I think most of us will have to agree it has to have a faster clock then 550Mhz

[Aleman]

It would have to have lots more pipelines... but the transistor number suggests it has the same number of pipelines as a 7800/7900.

[xen]

It would have to have lots more pipelines... but the transistor number suggests it has the same number of pipelines as a 7800/7900.

The transistor count was never clearly defined.

The spec said 300+ million transistors.

That could mean
301 million
350 million
399 million you get the idea.

Also bare in mind a 7900 is just 278 million transistors for comparison

And also includes many millions of PC related stuff PS3 wont include (pureVideo etc)

Clocking it higher could account for the difference but then PS3 would need to be a fridge to contain the extra heat.

[janenba352]

according to Sony the system as a huge cooling system so it should stay pretty cool.

Also 300+ million transistors doesnt give much of an idea on what the actual count is. So I agee with you there could be over 700 million transistors on the thing and we have no idea. Do we even have a clue on how many pipelines there officially supposed to be.

[DarkChild]

I went back to watch the e305 show where the ceo of Nvdia presented the RSX and at the 29:10 min mark he goes on to say that to provide the shading power needed for the RSX to do what it does, they incorporated a "farm of shader processors" on the chip. now a while back udontkneedtoknow tried explaining this but it was on paper,it was interesting that he said a "FARM" of shader processors

[archy141]

I went back to watch the e305 show where the ceo of Nvdia presented the RSX and at the 29:10 min mark he goes on to say that to provide the shading power needed for the RSX to do what it does, they incorporated a "farm of shader processors" on the chip. now a while back udontkneedtoknow tried explaining this but it was on paper,it was interesting that he said a "FARM" of shader processors

interesting... I wonder if that was just marketing buzz words or actual hint to the final RSX. If someone has time maybe they can go through that G80 patent link i supplied. Maybe there is a relation in one of those 40+ pages.

I must dig udontkneedtoknow's discussion, on this. In fact he has not commented on this thread as yet.


Archy

[Aleman]

Well, the Geforce 7800 and 7900 each have 32 shader pipelines. That could count as a "farm."

[DarkChild]

I went back to watch the e305 show where the ceo of Nvdia presented the RSX and at the 29:10 min mark he goes on to say that to provide the shading power needed for the RSX to do what it does, they incorporated a "farm of shader processors" on the chip. now a while back udontkneedtoknow tried explaining this but it was on paper,it was interesting that he said a "FARM" of shader processors

interesting... I wonder if that was just marketing buzz words or actual hint to the final RSX. If someone has time maybe they can go through that G80 patent link i supplied. Maybe there is a relation in one of those 40+ pages.

I must dig udontkneedtoknow's discussion, on this. In fact he has not commented on this thread as yet.


Archy

I believe in udontkneedtoknow's discussion they came to the conclusion that the word "processor" may just have been used in a way to represent pipes. but what if the RSX is a 7800 core with multiple 6800 cores acting as SPE's? remember all along sony wanted to do something similar with the cell, have one as a cpu and one as a GPU. what if they did that, only they used true gpu cores?

[nick²]

I doubt there is any change...

Its going to be a 7800 GTX 512 with half the memory bandwidth and half the memory...

[hisame]

Like I always said comparing PS3 with PC is just no good.
As for RSX we know so little of it the only thing we knew was like archy141 posted,
48 pipe line with shader 4.0, and the shader will be fix shader.
Frankly now even that is obsolete, sony rmove all spec for cell and RSX on march.
So all we know is RSX is a fix shader chip with the GDDR3 or 4, and FlexIO bus to Cell.
All we know is RSX will be at least as good as 7900 GTX,
that's what sony put on version 2 SDK and they were saying it is too slow compare to RSX.

A lot also compare RSX with 360 GPU by pipe line number that is stupid.
Nvidia CPOU has lower pipeline count then ATI GPU even with par performance.
1900XTX par with 7900 GTX where 1900 has 56 pipe lines and 7900 only has 32.
If RSX at leas has 48 Nvidia pipeline which was the old RSX spec,
by ratio we are looking at somthing much faster then 7950 GX2.
Also Nvidia has the habbit of putting pixel to vertex shader ratio of 3:1.
Which is very good for Cell since SPE can do vertex processing.
Maybe we are looking at 36 fragment pipeline with 12 vertex pipeline.

All my old post I used 7900 GTX as replacement of RSX since that is what SDK ver.2 had.
But it is also a fact that sony said 7900 is only 30% power of RSX.
If we take that literally RSX is able to do at least 48B pixel/sec, and 4.2B vertex/sec.
Even just compare the pipeline number RSX should be at least as good as 7950 GX2.
There is very limited on what we know about PS3 at the moment since sony remove all spec toward cell and RSX at march.
All I can say is look forward to TGS, everything should be clear by then.
And RSX should be at least as good as 7950 GX2.

[NUMA NUMA]

If someone could put this in laymans terms please, I'm confused :?

[razorblade416]

If someone could put this in laymans terms please, I'm confused :?

RSX= beats da crap outta your PC graphics card :lol:

[Mason666]

Razorblade416 that is a stupid comment :?, I'm not a PC fanboy so don't flame me and I've already ordered a PS3.
However I know a PC will cream a PS3 it just comes at a price, usually 3 or 4 times that of PS3 but still. The RSX is meant to be equivalent to 2 Nvidia 6800s in SLi, which in PC terms is old technology. Set ups now using 2 7900GTXs in SLi are literally 3 to 4 times faster than that of the 6800s, which is what the RSX is claimed to perform as (in the E3 '05 Sony Conference).

[THEREALNEXTGEN]

Razorblade416 that is a stupid comment :?, I'm not a PC fanboy so don't flame me and I've already ordered a PS3.
However I know a PC will cream a PS3 it just comes at a price, usually 3 or 4 times that of PS3 but still. The RSX is meant to be equivalent to 2 Nvidia 6800s in SLi, which in PC terms is old technology. Set ups now using 2 7900GTXs in SLi are literally 3 to 4 times faster than that of the 6800s, which is what the RSX is claimed to perform as (in the E3 '05 Sony Conference).

Have you not read the posts in the topic. You are opening yourself to some flaming buddy. You know Sony isn't going to have it outdated like that. The information presented by many members from reputable sources and Sony's plan dispells that the RSX will be like 2 SLI 6800's. Please read in the future and don't just post crap and common knowledge.

[DarkChild]

Well, the Geforce 7800 and 7900 each have 32 shader pipelines. That could count as a "farm."

you may very well be right. but here's the problem...neither one of us knows for sure. I am however inclined to believe that they may have gone a multi core route. because they worked with sony on the project. it's not like sony went to them and left them to come up with something on their own.(not saying any of this is true just just why I took interest in the "farm" comment)

[DarkChild]

Well, the Geforce 7800 and 7900 each have 32 shader pipelines. That could count as a "farm."

you may very well be right. but here's the problem...neither one of us knows for sure. I am however inclined to believe that they may have gone a multi core route. because they worked with sony on the project. it's not like sony went to them and left them to come up with something on their own.(not saying any of this is true just just why I took interest in the "farm" comment)

[DarkChild]

I went back to watch the e305 show where the ceo of Nvdia presented the RSX and at the 29:10 min mark he goes on to say that to provide the shading power needed for the RSX to do what it does, they incorporated a "farm of shader processors" on the chip. now a while back udontkneedtoknow tried explaining this but it was on paper,it was interesting that he said a "FARM" of shader processors

interesting... I wonder if that was just marketing buzz words or actual hint to the final RSX. If someone has time maybe they can go through that G80 patent link i supplied. Maybe there is a relation in one of those 40+ pages.

I must dig udontkneedtoknow's discussion, on this. In fact he has not commented on this thread as yet.


Archy

I believe in udontkneedtoknow's discussion they came to the conclusion that the word "processor" may just have been used in a way to represent pipes. but what if the RSX is a 7800 core with multiple 6800 cores acting as SPE's? remember all along sony wanted to do something similar with the cell, have one as a cpu and one as a GPU. what if they did that, only they used true gpu cores?

edit: sorry about the double and tripple posting I'm on my phone touch screen and it can get phony at times

[janenba352]

Razorblade416 that is a stupid comment :?, I'm not a PC fanboy so don't flame me and I've already ordered a PS3.
However I know a PC will cream a PS3 it just comes at a price, usually 3 or 4 times that of PS3 but still. The RSX is meant to be equivalent to 2 Nvidia 6800s in SLi, which in PC terms is old technology. Set ups now using 2 7900GTXs in SLi are literally 3 to 4 times faster than that of the 6800s, which is what the RSX is claimed to perform as (in the E3 '05 Sony Conference).

umm... the cell processor is already as strong as most PCs now a days with 6800GTs installed. And the RSX is supposed to be the one of the strongest video cards made at system launch. So how exactly does current PCs beat the PS3. The will with ATI's and Nvidea's newest video card released at the end of the year when it comes to GPUs but not CPUs. The Cell still makes the difference.

Also stop refering to old news. E3 '05 is old news and E3 '06 came and gone with no info on specs. Lets wait for TGS or any of the other smaller shows to see if anything new about the specs are released. Remember Devs said the amount of ram in the final dev kits were increased as well as the power of the kits.

Eventually one of the tech sites with get their hands on one of the dev kits then we will see whats in them if Sony doesnt announce the specs before that.

[Guilty Bystander]

umm... the cell processor is already as strong as most PCs now a days with 6800GTs installed. And the RSX is supposed to be the one of the strongest video cards made at system launch. So how exactly does current PCs beat the PS3. The will with ATI's and Nvidea's newest video card released at the end of the year when it comes to GPUs but not CPUs. The Cell still makes the difference.

Also stop refering to old news. E3 '05 is old news and E3 '06 came and gone with no info on specs. Lets wait for TGS or any of the other smaller shows to see if anything new about the specs are released. Remember Devs said the amount of ram in the final dev kits were increased as well as the power of the kits.

Eventually one of the tech sites with get their hands on one of the dev kits then we will see whats in them if Sony doesnt announce the specs before that.


SLI, Crossfire and Quad SLI easily beats Xenos and RSX.
Also don't be too confident about Cell because Intel's Conroe will be available in quad core formation at the end of this year.

[DarkChild]

umm... the cell processor is already as strong as most PCs now a days with 6800GTs installed. And the RSX is supposed to be the one of the strongest video cards made at system launch. So how exactly does current PCs beat the PS3. The will with ATI's and Nvidea's newest video card released at the end of the year when it comes to GPUs but not CPUs. The Cell still makes the difference.

Also stop refering to old news. E3 '05 is old news and E3 '06 came and gone with no info on specs. Lets wait for TGS or any of the other smaller shows to see if anything new about the specs are released. Remember Devs said the amount of ram in the final dev kits were increased as well as the power of the kits.

Eventually one of the tech sites with get their hands on one of the dev kits then we will see whats in them if Sony doesnt announce the specs before that.


SLI, Crossfire and Quad SLI easily beats Xenos and RSX.
Also don't be too confident about Cell because Intel's Conroe will be available in quad core formation at the end of this year.

the quad cores will be nice I'm looking foward to them. but in terms of floating point math and decoding video stream HD or standard I don't think the conroe will come close. let's not forget the quad core is more of an integer math monster(general purpose)

edit: I'm going out on a limb here and say that I think one of the cells spe's will out do the conroe in floating point math

[DarkChild]

umm... the cell processor is already as strong as most PCs now a days with 6800GTs installed. And the RSX is supposed to be the one of the strongest video cards made at system launch. So how exactly does current PCs beat the PS3. The will with ATI's and Nvidea's newest video card released at the end of the year when it comes to GPUs but not CPUs. The Cell still makes the difference.

Also stop refering to old news. E3 '05 is old news and E3 '06 came and gone with no info on specs. Lets wait for TGS or any of the other smaller shows to see if anything new about the specs are released. Remember Devs said the amount of ram in the final dev kits were increased as well as the power of the kits.

Eventually one of the tech sites with get their hands on one of the dev kits then we will see whats in them if Sony doesnt announce the specs before that.


SLI, Crossfire and Quad SLI easily beats Xenos and RSX.
Also don't be too confident about Cell because Intel's Conroe will be available in quad core formation at the end of this year.

the quad cores will be nice I'm looking foward to them. but in terms of floating point math and decoding video stream HD or standard I don't think the conroe will come close. let's not forget the quad core is more of an integer math monster(general purpose)

[archy141]

Like I always said comparing PS3 with PC is just no good.
As for RSX we know so little of it the only thing we knew was like archy141 posted,
48 pipe line with shader 4.0, and the shader will be fix shader.
Frankly now even that is obsolete, sony rmove all spec for cell and RSX on march.
So all we know is RSX is a fix shader chip with the GDDR3 or 4, and FlexIO bus to Cell.
All we know is RSX will be at least as good as 7900 GTX,
that's what sony put on version 2 SDK and they were saying it is too slow compare to RSX.

A lot also compare RSX with 360 GPU by pipe line number that is stupid.
Nvidia CPOU has lower pipeline count then ATI GPU even with par performance.
1900XTX par with 7900 GTX where 1900 has 56 pipe lines and 7900 only has 32.
If RSX at leas has 48 Nvidia pipeline which was the old RSX spec,
by ratio we are looking at somthing much faster then 7950 GX2.
Also Nvidia has the habbit of putting pixel to vertex shader ratio of 3:1.
Which is very good for Cell since SPE can do vertex processing.
Maybe we are looking at 36 fragment pipeline with 12 vertex pipeline.

All my old post I used 7900 GTX as replacement of RSX since that is what SDK ver.2 had.
But it is also a fact that sony said 7900 is only 30% power of RSX.
If we take that literally RSX is able to do at least 48B pixel/sec, and 4.2B vertex/sec.
Even just compare the pipeline number RSX should be at least as good as 7950 GX2.
There is very limited on what we know about PS3 at the moment since sony remove all spec toward cell and RSX at march.
All I can say is look forward to TGS, everything should be clear by then.
And RSX should be at least as good as 7950 GX2.

Great POST ! I have been waiting quite a while for you to make a posting here as i value your opinion. Do you any links pointing to this ?


Sony said 7900 is only 30% power of RSX

I would like to clarify what HISAME said about PS3 & PC architecture.It is not correct to compare for instance a NV47 core in a PS3 with one in a PC. One will be running in an extremely REFINED architecture that is optimised for graphics work & the other in a relatively crippled architecture. Also the nv47 in the PS3 will be in a CLOSED setup & the developers wil be able to squeeze SO MUCH more out of it unlike PC. PC games programming has to rely on unoptimised open ended programming to ensure compatiablity across a rangeof different PC configurations.



Sony PlayStation 3 Hardware May Not Be Finalized – Sources.
Developers Use Personal Computers to Show PS3 Advantages
[ 06/06/2006 | 09:49 AM ]

..The game developers who did not run their demos on the pre-release Sony PlayStation 3 hardware, used high-performance personal computers, such as those featuring high-speed Nvidia GeForce 7900 graphics cards in SLI mode or even ATI’s Radeon X1900 in CrossFire mode...

[razorblade416]

I would like to clarify what HISAME said about PS3 & PC architecture.It is not correct to compare for instance a NV47 core in a PS3 with one in a PC. One will be running in an extremely REFINED architecture that is optimised for graphics work & the other in a relatively crippled architecture. Also the nv47 in the PS3 will be in a CLOSED setup & the developers wil be able to squeeze SO MUCH more out of it unlike PC. PC games programming has to rely on unoptimised open ended programming to ensure compatiablity across a rangeof different PC configurations.

That can be said about any part on any console. That's what makes consoles so great, even if they don't have the best hardware, the devs can tailor to the parts and really make a better game than one running on a higher end PC.

Take the PS2 for example, 300mhz, 4MB gfx card ,nothing special compared to the PCs of today. And yet, since devs have worked on sqeezing every ounce of power of the machine, it can run more powerful games than PCs with MUCH higher specs than the console 8)

[archy141]

POSSIBLY VERY IMPORTANT FIND







March 14, 2006 - At the PlayStation Business Briefing 2006, Ken Kutaragi has just added another infamous quote to his belt, this time stating that PS3 games are "live" and that the PS3 concept is "4D" http://uk.ps3.ign.com/articles/696/696060p1.html


At the time of hearing this whole world thought this MAN has really gone MAD.
4D ?? ! ! I personally became little fearful for Sony to see someone so high up talking gibberish.

Than yesterday as I was scouring the web for information on the RSX, I came across a forum where a member had discreetly left created a thread, left a message & practically disappeared –still no chance to thanks him. It immediately rang BELLS & WHISTLES *&^%$ ! .

It was a link to a PATENT that connected Kens infamous words from E3 2005 ‘4D’ to a PATENT that a number of technical guys verified that the patents decisively relate to NVIDIA G80 !

Hang on we have MAD Ken shouting ‘4D’ ] to world media & we have a NVIDIA GPU PATENT shouting ‘4D’ technology.

See quote from patent.




Originally Posted by version


4D !

"A discrete BRDF is naturally stored in a 4D lookup table.
While commercially available graphics hardware has long had
2D texture mapping capability, no hardware currently in use
has the 4D texture mapping capability necessary to implement
this natural representation. Even if 4D texture mapping were
implemented, the tables required to hold a measured BRDF would
be quite large. At four bytes per sample (single-precision
floating point) and a modest sampling rate of approximately
twenty samples per radian (or a parabolic parameterization of
64.times.64 samples), a BRDF requires 64 MB of memory. Since
a modern graphics accelerator board has only 64 MB of total
texture memory, this storage requirement is prohibitively large.
Conventional texture compression algorithms can reduce the space
requirement, but not by more than an order of magnitude.
"
http://patft.uspto.gov/netacgi/nph-P...s&RS=Papakipos


I believe Crazy Ken was ranting about this EXACT 4D TEXTURE technology that the PS3 would come to possess i.e a RSX feature.
The man is not crazy but just obsessed about creating the first ever True NextGen console.


The 4D is back in the equation !




17. Physically Based Reflectance for Games
Monday
Half Day, 8:30 am - 12:15 pm
Level: Intermediate
Room 205

How to use the physical principles of reflectance to increase game realism while acknowledging real-world production issues such as performance and ease of content creation. In this course, game developers learn how to incorporate realistic reflectance in their games, and graphics researchers learn how to evaluate the applicability of their research to game development.

Prerequisites
Some knowledge of a low-level graphics API such as DirectX or OpenGL, or experience with modern graphics hardware. Basic understanding of calculus is recommended.

Intended Audience
Game developers and researchers whose focus is real-time graphics.

Organizer
Nathaniel Hoffman
Naughty Dog, Inc.

http://now.siggraph.org/s2006/main.php?f=conference&p=courses&s=17

Just in case you are not aware Naughty Dog, Inc. are first party Sony developers.

So what does "Physically Based Reflectance for Games" have to do with ‘4D’ ?




Light reflectance functions describe how light is scattered at the surface of an object. The amount of reflected light typically varies with the incident direction of the incoming light and with the measurement direction for the reflected light. As such, the bidirectional reflectance distribution function (BRDF) at given point on a surface and for a given wavelength is 4-dimensional.
http://www.graphics.cornell.edu/~eric/


So in case you cannot put the links together for yourself.

1. We have (what I think is a G80 related) Nvidia Patent talking about supporting ‘4D’ BRDF

2. We have Sony comming out with bizaar uses of the phrase ‘4D’ in there marketing

3. We have Sony's Naughty Dog organising courses on "Physically Based Reflectance for Games"

4. We have a confirmation that Physically Based Reflectance is related to ‘4D’ BRDF

So to summarize Sony are running Siggraph courses that invlolve use of ‘4D’ BRDF in games. Coincidentally the G80 patents show that it is uniquely capable of doing ‘4D’ BRDF in hardware.



Hmm... things hotting up & my heads going to EXPLODE !
Is this all coincidental or what ? Anyone else see the links or am i going Crazy like Ken ?

[NUMA NUMA]

^Whoa, that still dosen't change the fact that Ken is absolutly nuts.

[hisame]

I have post about the 4D thing long time ago.
It has been explained more clearly by IBM to a IEEE Journalist.
It is always easier for pro to understand pro.
In last year's "specturm" magazine, a pro mag for IEEE members.
I am a member of IEEE. In the mag the IBM "Cell" PD team leader show how great Cell is.
Only using a single Cell, it was bale to DL a saterllite picture and height data real time from saterllite.
Using this info the Cell box draw up a real time flight simulation of the moutain terrain.
Yes, PS3 will be able to DL real time data from the net and use them real time in games.
You can drive GT game in Tokyo or NY with real time traffic and weather info, statically fit to your game.
That's what sony try to show with that London demo, which no one understand,
most "tech" people only mention how good the 3D simulation was.
Without knowing that demo was DL real time from the net of all the data about London,
with pre install Londo 3D frame map, PS3 was able to real time simulate a real time Londo.
Stastically correct like traffic, population time and weather.

[archy141]

I have post about the 4D thing long time ago.
It has been explained more clearly by IBM to a IEEE Journalist.
It is always easier for pro to understand pro.
In last year's "specturm" magazine, a pro mag for IEEE members.
I am a member of IEEE. In the mag the IBM "Cell" PD team leader show how great Cell is.
Only using a single Cell, it was bale to DL a saterllite picture and height data real time from saterllite.
Using this info the Cell box draw up a real time flight simulation of the moutain terrain.
Yes, PS3 will be able to DL real time data from the net and use them real time in games.
You can drive GT game in Tokyo or NY with real time traffic and weather info, statically fit to your game.
That's what sony try to show with that London demo, which no one understand,
most "tech" people only mention how good the 3D simulation was.
Without knowing that demo was DL real time from the net of all the data about London,
with pre install Londo 3D frame map, PS3 was able to real time simulate a real time Londo.
Stastically correct like traffic, population time and weather.


WOOOwww Thats totally aamazzing ! ! ! Almost scary !



-So does that mean the links to G80 4D BRDF patent are not related to RSX in any way as i was thinking ?

-Those course that Sony are doing on 4D BDRF are not because they are related to a technology that the RSX might possess (feature from G80) ?

-Is it possible that 4D relates to RSX as i am speclating & 4D also relates to Cell in a different way as you explained ?



Archy.

[razorblade416]

Not to big of deal for the weather. My Flight Simulater 2004 could do real-time weather for where you were flying.

But real time TERRIAN FROM SATELLITES!?!?!?!?! I'm sry but that freakin' amazing :shock: :shock:

Think----GTA4 8)

[archy141]

But would there be any computational power left to do anything else ?? SFX, AI, Physics, etc etc

[janenba352]

you do know 4D is time right... so unless to can control time its not real 4D lol.

http://www.tenthdimension.com/flash2.php

[udontneed2know]

What Ken Kutaragi meant by 4D gaming is simply high level 3D gaming done in complete real-time by the program. He's essentially talking about the coming age when Procedural techniques will begin dominating the programmers workload. We have entered the era where CPU's and GPU's have enough computational power to begin assembling various sections of a single object on its own, which in the most general of terms, is what Procedural Generation is all about. The processor is given a list of object, like pieces of a puzzle, and through math the processor knows what pieces to put together and when and it will draw the finished object on the screen in real-time.

EA is using Procedural Generation for animation in this new generation, as are alot of other development studios. Warhawk has procedurally generated water effects. Spore has the procedurally generated character modeller and animation software. Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.

So thats 4D gaming. 3D gaming handle in real-TIME by the processors themselves.

Both Nvidia and Sony have been pretty vague about the overall innerds of the RSX processor. Its definately not a straight port of any existing PC Nvidia GPU. It definately has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

[Phantasmn]

you do know 4D is time right... so unless to can control time its not real 4D lol.

http://www.tenthdimension.com/flash2.phpI think i just got a headache from that 10 dimensional stuff about string theory... It seems to make sense at first, but when you think about it, it seems different. Stupid ants!

[janenba352]

you do know 4D is time right... so unless to can control time its not real 4D lol.

http://www.tenthdimension.com/flash2.phpI think i just got a headache from that 10 dimensional stuff about string theory... It seems to make sense at first, but when you think about it, it seems different. Stupid ants!

But its a good little video. But yea like you said udontneed2know he is talking about real-time. real-time weapon change, real-time shifting?, real time motion?, real time... physics.... im guessing. Im just saying he is meaning time when he says 4D gaming.

plus that video is kinda cool kinda explain how the prince of persia controls time but popping in and out of the 5th dimention or was it 4th... damn have to re-watch it...

[Jabjabs]

Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.


I didn't shout this from the Hills, but I got a new job last week. Moved from that dead beat (it was okay) dev house to a much smaller independent developer, all 8 of us. The idea is to eventually get a game on to XBLA, anyway thanks to that goal I have finally got my first taste of procedural graphics and damn, I love it already. This is the future of game technology.


I think i just got a headache from that 10 dimensional stuff about string theory... It seems to make sense at first, but when you think about it, it seems different. Stupid ants!

I thought there where 11 dimensions? Time to read up on this again. :|

[Phantasmn]

Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.


I didn't shout this from the Hills, but I got a new job last week. Moved from that dead beat (it was okay) dev house to a much smaller independent developer, all 8 of us. The idea is to eventually get a game on to XBLA, anyway thanks to that goal I have finally got my first taste of procedural graphics and damn, I love it already. This is the future of game technology.


I think i just got a headache from that 10 dimensional stuff about string theory... It seems to make sense at first, but when you think about it, it seems different. Stupid ants!

I thought there where 11 dimensions? Time to read up on this again. :|Your right, it is 11. I forgot about the 0 dimension in the video.... time to read up on string theory lol

[bwogowly]

Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.


I didn't shout this from the Hills, but I got a new job last week. Moved from that dead beat (it was okay) dev house to a much smaller independent developer, all 8 of us. The idea is to eventually get a game on to XBLA, anyway thanks to that goal I have finally got my first taste of procedural graphics and damn, I love it already. This is the future of game technology.


I think i just got a headache from that 10 dimensional stuff about string theory... It seems to make sense at first, but when you think about it, it seems different. Stupid ants!

I thought there where 11 dimensions? Time to read up on this again. :|Your right, it is 11. I forgot about the 0 dimension in the video.... time to read up on string theory lol

You don't even have to read up on String Theory, which talks about 10 dimmensions, nor M-Theory, which explains the String Theory to all coinsist by adding the 11th dimmension, because Gravity Theory had it right to begin with, with 11 dimmensions to explain the universe with. I saw it on some Discovery channel or something, this guy made up Gravity Theory but no one bought it because everyone else had only 10 dimmensions with String Theory, which didn't get completed until a guy came up with M-Theory, by adding another dimmension, adding up to the same amount of Gravity Theory that had it right all along. Since String Theory and M-Theory are so complicated (I think they'd have to be more complicated than Gravity Theory because they were stuck with only 10 dimmensions for a while), I suggest to just read up on Gravity Theory because it was right to begin with. Both M-Theory and Gravity Theory will do fine, but I would prefer to give credit where credit is due. Credit should go to the person that came up with an idea first, so Gravity Theory is the choice for me. :)

[Ramza]

Bending the fifth dimension through the sixth to jump timelines... hmmm has anyone seen the movie Butterfly Effect. (That string theory is good stuff.)

I guess I should contribute to the topic too.

Considering the extreme secrecy at this point I do not think there is any chance that the RSX is as given at E3 05. However, I think that all of the speculation about an entirely new design is a bit overboard. I think increased clock speeds, additional pipelines, and maybe more RAM are the upgrades that can be expected. The only thing Sony really needs is enough to make RSX > Xenos. If they can assure that each part of the PS3 is a step up (regardless of now small) from the 360 it will go a long way for justifying their price point, even for the people who don't care about Blu-ray.

I find it very interesting that devs cannot even answer the question; "Has the spec changed from the E3 05 numbers" Archy's squirrelly little friend over at the NT forums has been repeatedly ignored when he tries to get an answer. If the spec was the same as E3 05 I don't see why Sony would have such tight security on it.

A summary of my thoughts:
I will be happy with the E3 05 RSX.
I expect some minor upgrades from E3 05 to make RSX >= Xenos.
I will be thrilled if they add another 256MB of VRAM.
I will shit myself if the entire architecture is changed.

[Phantasmn]

Bending the fifth dimension through the sixth to jump timelines... hmmm has anyone seen the movie Butterfly Effect. (That string theory is good stuff.)

I guess I should contribute to the topic too.

Considering the extreme secrecy at this point I do not think there is any chance that the RSX is as given at E3 05. However, I think that all of the speculation about an entirely new design is a bit overboard. I think increased clock speeds, additional pipelines, and maybe more RAM are the upgrades that can be expected. The only thing Sony really needs is enough to make RSX > Xenos. If they can assure that each part of the PS3 is a step up (regardless of now small) from the 360 it will go a long way for justifying their price point, even for the people who don't care about Blu-ray.

I find it very interesting that devs cannot even answer the question; "Has the spec changed from the E3 05 numbers" Archy's squirrelly little friend over at the NT forums has been repeatedly ignored when he tries to get an answer. If the spec was the same as E3 05 I don't see why Sony would have such tight security on it.

A summary of my thoughts:
I will be happy with the E3 05 RSX.
I expect some minor upgrades from E3 05 to make RSX >= Xenos.
I will be thrilled if they add another 256MB of VRAM.
I will *expletive deleted* myself if the entire architecture is changed.

On Butterfly Effect, that was a good movie and they used the multi-dimensional theories well also.

But back on topic, I'm sure that I speak for many people here that we all wish that RSX will have some sort of improvement, especially more RAM.

Now, i don't want to be seen as raining on anyone's parade, but, for all of those saying that you would like to see GDDR4 for the RSX, there is almost no way that SONY would be able to secure enough this late in the game. And besides, there is already one type of unproven memory, isn't that enough risk?

Also, for everyone hoping for more pipelines to be added, that would most likely cause an increase in heat. Now, I'm not exactly the most qualified person to speak about video cards and whatnot, but with all of the hot stuff under the hood of the PS3 as it is, that box would have to increase in size, weight, and noise in order to keep that beast cool. I don't know about you, but I would rather not have to deal with the Jumbo Jet landing after 15 minutes of turning on the system. Quite a few of my friends have 360's, and I plan on getting one too, just not such a loud one without an HDMI port (whole other topic).

Just my 2 cents.

[CaptHavok]

Now, i don't want to be seen as raining on anyone's parade, but, for all of those saying that you would like to see GDDR4 for the RSX, there is almost no way that SONY would be able to secure enough this late in the game. And besides, there is already one type of unproven memory, isn't that enough risk?




If your refering to the XDR as unproven memory I think your mistaken for IBM has been proving that memory for sometime now its not in the class with GDDR4. XDR is made by Rambus and its a RDRAM device not a DDR device and XDR has been clocked between 3Ghz to 8Ghz and thats just to start there are later versions planned for PC base that will run even faster.

IBM even went as far as to super freeze a processor to see how fast it can go and it went into the 100s of Ghz so I feel confident in IBM scientist to test the hardware that will be in my PS3.

[bwogowly]

Wow, that totally makes sense, I have to read up on G80 now and Shader Model 4.0! (Frantic...) YAY!!!

"Also, for everyone hoping for more pipelines to be added, that would most likely cause an increase in heat. Now, I'm not exactly the most qualified person to speak about video cards and whatnot, but with all of the hot stuff under the hood of the PS3 as it is, that box would have to increase in size, weight, and noise in order to keep that beast cool. I don't know about you, but I would rather not have to deal with the Jumbo Jet landing after 15 minutes of turning on the system. Quite a few of my friends have 360's, and I plan on getting one too, just not such a loud one without an HDMI port (whole other topic)," Phantasmn. For one so worried about hotness, I can't see how you want a 360, which some catch themselves on fire. I even have a friend that saw one of them do it.

"IBM even went as far as to super freeze a processor to see how fast it can go and it went into the 100s of Ghz so I feel confident in IBM scientist to test the hardware that will be in my PS3," CaptHavok. Here's more info on that: http://arstechnica.com/news.ars/post/20060622-7117.html

[WelshBluebird]

Also don't be too confident about Cell because Intel's Conroe will be available in quad core formation at the end of this year.

tbh, that won't make a difference for a while.
Most PC games atm can't use two cores, let alone 4. Actually, the only came I Know of which is being developed with multi cores in mind is crysis.

[Hexadecimal]

Oblivion is designed to take advantage of multi-core CPUs, infact it was the first Xbox 360 games to use all 3 cores of the Xbox 360. There are many games that have support for multi-core CPUs, such as Age of Empires 2, Battlefied 2, Black and White 2, Call of Duty 2, F.E.A.R, Quake 4, Quake 5, Tom Clancy's Rainbow Six 3, World of Warcraft

[Lefein]

Heh, watch out for AMD, they have a memory controller on-die whereas Conroes is off-die. AMDs AM2 architecture is a lot more flexible and can be pushed for more gains due to the on-die design.

Intel simply put a band-aid to a larger problem. It will perform better than the FX-62, but there's nothing stopping AMD from dropping to 65nm and redesigning AM2 chips to stomp Conroe..

As far as Cell is concerned, its not even x86 based, so any comparison between Intel and Cell is just kinda silly.

[archy141]

...So that’s 4D gaming. 3D gaming handle in real-TIME by the processors themselves...
Both NVIDIA and Sony have been pretty vague about the overall innards of the RSX processor. Its definitely not a straight port of any existing PC NVIDIA GPU. It definitely has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

Thanks for your contribution.

Does that possibly mean the links I have been making between NVIDIA’s 4D texture mapping patents & Sony are incorrect ?

Additionally I pointed out the fact that Sony are organizing Siggraph 2006 courses on 'Physically Based Reflectance for Games'. This subject is also related to 4D texture mapping.

I guess they could just all be sheer coincidence.

Maybe 4D applies to both situations. It is possible that Sony's bizarre marketing Buzz word '4D' actually took its life from the unique hardware feature of there GPU.

I would also like to touch another RSX design speculation that’s been in the rumour circles for lot longer.



http://www.geocities.com/yerbrowneyedgurl/lefein/RSX1_sensitive.jpg


If just say this exotic design did turn out to reflect the final RSX design, wouldn’t there be huge incompatibility in running the already developed games as it looks way too far architecturally different form the classic NV47 design ?



Archy

[Makagoto]

nvidia did a console gpu before, it was custom but based on 2 designs. one design of the future and one of the past. it had little more than the past product and a little less than the future product. guess rsx will turn out with a similar progress.

[razorblade416]

What Ken Kutaragi meant by 4D gaming is simply high level 3D gaming done in complete real-time by the program. He's essentially talking about the coming age when Procedural techniques will begin dominating the programmers workload. We have entered the era where CPU's and GPU's have enough computational power to begin assembling various sections of a single object on its own, which in the most general of terms, is what Procedural Generation is all about. The processor is given a list of object, like pieces of a puzzle, and through math the processor knows what pieces to put together and when and it will draw the finished object on the screen in real-time.

EA is using Procedural Generation for animation in this new generation, as are alot of other development studios. Warhawk has procedurally generated water effects. Spore has the procedurally generated character modeller and animation software. Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.

So thats 4D gaming. 3D gaming handle in real-TIME by the processors themselves.

Both Nvidia and Sony have been pretty vague about the overall innerds of the RSX processor. Its definately not a straight port of any existing PC Nvidia GPU. It definately has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

So, could Procedural Generation make an infinite about of diffent people such as in a crowd in a football game or people roaming the streets in GTA? That would be AMAZING! :shock:

[mrboo]

What Ken Kutaragi meant by 4D gaming is simply high level 3D gaming done in complete real-time by the program. He's essentially talking about the coming age when Procedural techniques will begin dominating the programmers workload. We have entered the era where CPU's and GPU's have enough computational power to begin assembling various sections of a single object on its own, which in the most general of terms, is what Procedural Generation is all about. The processor is given a list of object, like pieces of a puzzle, and through math the processor knows what pieces to put together and when and it will draw the finished object on the screen in real-time.

EA is using Procedural Generation for animation in this new generation, as are alot of other development studios. Warhawk has procedurally generated water effects. Spore has the procedurally generated character modeller and animation software. Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.

So thats 4D gaming. 3D gaming handle in real-TIME by the processors themselves.

Both Nvidia and Sony have been pretty vague about the overall innerds of the RSX processor. Its definately not a straight port of any existing PC Nvidia GPU. It definately has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

So, could Procedural Generation make an infinite about of diffent people such as in a crowd in a football game or people roaming the streets in GTA? That would be AMAZING! :shock:

Procedual generation was used on Evolution Studio's WRC games on PS2, thats how the games were able to have an enourmous amount of tree/vegatation on screen. Procedual game desighn is the future.

[EbonySeraphim]

...So that’s 4D gaming. 3D gaming handle in real-TIME by the processors themselves...
Both NVIDIA and Sony have been pretty vague about the overall innards of the RSX processor. Its definitely not a straight port of any existing PC NVIDIA GPU. It definitely has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

Thanks for your contribution.

Does that possibly mean the links I have been making between NVIDIA’s 4D texture mapping patents & Sony are incorrect ?

Additionally I pointed out the fact that Sony are organizing Siggraph 2006 courses on 'Physically Based Reflectance for Games'. This subject is also related to 4D texture mapping.

I guess they could just all be sheer coincidence.

Maybe 4D applies to both situations. It is possible that Sony's bizarre marketing Buzz word '4D' actually took its life from the unique hardware feature of there GPU.

I would also like to touch another RSX design speculation that’s been in the rumour circles for lot longer.

[IMAGE]

If just say this exotic design did turn out to reflect the final RSX design, wouldn’t there be huge incompatibility in running the already developed games as it looks way too far architecturally different form the classic NV47 design ?



Archy
That image design is pretty much a copy of the Cell processor except with a few numbers changed to match the clock speed of the Cell. It's pretty much a speculative diagram assuming the RSX wasn't based off of anything nVidia has already done or is doing that replicates the Cell's design.

[archy141]

I personally don't have any faith in that design happening but others did bring it to my attention so i thought it would be good to discuss its merits & falls openly.


Also


Oh, new vocabulary term...

"Synergistic Media Element"

To all who were left curious, thats the SME in the block diagram above.




Archy

[Aleman]

That diagram doesn't make any sense. GPU's already have arrays of pixel and vertex pipelines that do parallel processing.

[hisame]

Sony said they will alter the spec of cell and RSX by a little this is why they took down the spec in March.
The rumor is PS3 will use heat pipe like top end MOBO.
COuld we expect higher clock cell and RSX? possible i suppose.
But I don't think it will be any design change on the chip, it too late for that.

[archy141]

Sony said they will alter the spec of cell and RSX by a little this is why they took down the spec in March..

Im not so sure i agree with that. From the interview it seems Ken is saying they will be using common heat pipe solution -something that shouldn't be a surprise by. He didn't say it was something special or proprietary that was going to be put in place or it was a change to the original cooling plans. Even the XBOX360 uses heat pipes of some form to do cooling


Lastly, Kutaragi addressed cooling the Cell. "We'll use heat pipes and a custom cooling solution, but the methods used will be common," he said. "We definitely aren't using any proprietary methods... We certainly couldn't fit a liquid cooling system [inside the PS3]... We're spending a lot on heat and electromagnetic interference [management]. The power supply could almost be sold separately."
http://uk.gamespot.com/news/6152738.html?sid=6152738

If anything it seems to have been made a lower priority over other areas that they feel are much more important.



Archy

[jlippone]

What Ken Kutaragi meant by 4D gaming is simply high level 3D gaming done in complete real-time by the program. He's essentially talking about the coming age when Procedural techniques will begin dominating the programmers workload. We have entered the era where CPU's and GPU's have enough computational power to begin assembling various sections of a single object on its own, which in the most general of terms, is what Procedural Generation is all about. The processor is given a list of object, like pieces of a puzzle, and through math the processor knows what pieces to put together and when and it will draw the finished object on the screen in real-time.

EA is using Procedural Generation for animation in this new generation, as are alot of other development studios. Warhawk has procedurally generated water effects. Spore has the procedurally generated character modeller and animation software. Spore is really the centerpiece game to why Procedural Generation is going to help development studios handle the artistic costs of this new generation. Instead of artists having to draw each little piece of a character model or tree model and then place it, instead the processors themselves will handle all that on its own and do it all in real-time.

So thats 4D gaming. 3D gaming handle in real-TIME by the processors themselves.

Both Nvidia and Sony have been pretty vague about the overall innerds of the RSX processor. Its definately not a straight port of any existing PC Nvidia GPU. It definately has likenesses to their latest work, but it will have several differences since your dealing with a fixed system and the Cell processor in general.

So, could Procedural Generation make an infinite about of diffent people such as in a crowd in a football game or people roaming the streets in GTA? That would be AMAZING! :shock:

Procedual generation was used on Evolution Studio's WRC games on PS2, thats how the games were able to have an enourmous amount of tree/vegatation on screen. Procedual game desighn is the future.
Procedural content creation has been used for ages, if you want good examples try to check on Elite and Frontier.

[hisame]

By the sound of it PS3 will have a common heat pipe with 2 fans air cooling.
I was expecting clock up because of the delay sony put on PS3,
it is very psosible that sony can do a better and cleaner yeild of Cell and RSX.
Most of you might not know by chip maker even with the same chip it is possible to reduce heat and put clock higher.
As the production yeild gets better, this is why you get the pre-overclock video cards these days.
As the process improve by time it is possible to get better chip with same process.
I was expecting the same on cell and RSX.

Back to RSX like archy141 said, it is not possible for RSX to be 3 time mre powerful then 7900GTX.
By the pipe line number and clock speed I would expect RSX to be as good as 7950GX2.
But you still got to remeber the bus between RSX and Cell is far more faster then on PC.
The stuff we saw on E3 2006 were made by ver.2 SDK,
which is a IBM cell server with a 512MB of DDR2,and a 7900GTX 256MB DDR3 with x16 PCIE.
Frankly most game dev has problem using the 7900GTX, but at least the code can run.
Not untill after E3 with Ver.3 SDK, the devs has a working full spec PS3.
The joke is Ver.3 SDK has 2 HDMI and full spec parts on it just like sony listed on E3 2005.

[turdhat]

But with a 128 bit bus, surely thats going to be a handicap for performance isn't it?

[Aleman]

But with a 128 bit bus, surely thats going to be a handicap for performance isn't it?

Yes. That's something that's been heavily overlooked on these boards.

[hisame]

Unlike PC which is mainly 32bit base and now going to 64bit on some parts.
PS3 is design down to instruction set, register size, bus code, cache size and RAM are all 128 bit base.
I would expect minimum handicap if there is any.

[Aleman]

Unlike PC which is mainly 32bit base and now going to 64bit on some parts.
PS3 is design down to instruction set, register size, bus code, cache size and RAM are all 128 bit base.
I would expect minimum handicap if there is any.

PC GPU's have 256-bit GDDR3 local memory that gives up to ~50 GB/s of memory bandwidth. The RSX will has 128-bit GDDR3 memory at 22.4 GB/s (equivalent to an ATI Radeon 9800 Pro circa 2003).

[CaptHavok]

What I see for PCs if we are considering them when talking about game machines I myself consider a counsole a better choice for gaming than a tax machine with a awsome graphics card. Even tho thos cards can cost as much as a PS3 but thats not for debate here.

GDDR3 256 has up to thats up to not actually used 50GB/s for only the graphics card which is a comparison at that not a good one as a PC with that kind of card doesnt have a multiprocessing supercomputer with RDRAM supporting it.

most afordable PCs have busses in the range of 800MB and with a 2 bit process to grab data per cycle with wait states to not lose data it all seems to me that the PC is a bottleneck to the graphics.

Mark Rein himself even stated to the fact that PCs are being left behind and that customers have to pay outrageous amounts to play the games available to the next gen consoles.

So what can be learned of this that with two excelant graphic devices in one machine and the memory thats fast enough to match will definatly be a fine machine to pay top dollar for.

[hisame]

That's one the thing I am not so sure about.
Since DDR only allows you to run on double channel.
What if you put 4 chips instead of 2?
I know rum bus will goes up to 4 channel but will GDDR do the same?

Anyway, low memory bandwidth on GPU has 2 effects.
Reduce ability when running in high resolusion, and larger texture.
The rule of thumb is how many 10GB of band width is how many M pixel the limit is.
So with 22.5Gb/s you are looking at 2M pixel,
so PS3 will just start to drop it's frame rate at 1080p, very much expected.

But you got to also remeber RSX can also use part of the main memory via another bus line.
Which means 22.5Gb/s is not the top max bandwidth for PS3.
Frankly EPIC and a number of sony first party are already considering this option on launch titles.
I meant console are always like these working on the knife edge.

[mrgtir]

But with a 128 bit bus, surely thats going to be a handicap for performance isn't it?

Yes. That's something that's been heavily overlooked on these boards.

You do realise that it has 2 128bit buses. It has one going in and one giong out. So its not going to choke as bad as if it had one 128bit bus.

[Aleman]

But with a 128 bit bus, surely thats going to be a handicap for performance isn't it?

Yes. That's something that's been heavily overlooked on these boards.

You do realise that it has 2 128bit buses. It has one going in and one giong out. So its not going to choke as bad as if it had one 128bit bus.

It has 2 128 bit buses? Since when?

[Knuckles126]

^^ I thought it only had one? :?

[The_One]

mrgtir is right on one point, there are two "buses" to the GDDR3, however, it is not 128bit each. There are only 64bit buses, not 128bit or 256bit buses, they are simply made up of smaller 64bit buses (So, 2 for 128bit and 4 for 256bit). Overall, though, Aleman is correct. The total is still 128bit. However, I'm not sure if it's full duplex (which may be what mrgtir is referring to).

@Aleman: I don't think we're overlooking the "bit" of the bus. After all, both consoles are in the same boat when it comes to the bus size. It's not a disadvantage to have a 128bit bus when your competition also has the same. It's going to be a bottleneck for both consoles, no doubt, but we'll see how things work out.

[mynd]

Not quite in the same boat. The 360 uses edram to alleviate bus traffic.
When your not having to write everything back to the memory, the 128 bit bus becomes less of a problem. If anything, some embedded ram like the 360 would be a nice feature to add to the RSX.

[archy141]

Not quite in the same boat. The 360 uses edram to alleviate bus traffic.
When your not having to write everything back to the memory, the 128 bit bus becomes less of a problem. If anything, some embedded ram like the 360 would be a nice feature to add to the RSX.

Absolutely agree. Having seen that Sony do watch & learn from there competitors & listen to developers, a solution to address this shortcoming is a real possibility.

This area needs to be covered & don't be surprised if they do a similar patch solution to XBOX360. In the event this happens - before people cry COPYCAT i believe this type of solution was previously implemented by ATI for Nintendo Gamecube.

Alternatively to using the ATI solution, i think it would be plausible for Sony to just add more cache to the RSX.

Also many people are guessing that more RAM is one of the upgrades with are likely to see with the RSX rather than any other changes.

What are the Pro's & Con's of:


1. Just adding to GDDR3 ?

2. Just Adding to the XDRAM ?



Archy

[Guilty Bystander]

mrgtir is right on one point, there are two "buses" to the GDDR3, however, it is not 128bit each. There are only 64bit buses, not 128bit or 256bit buses, they are simply made up of smaller 64bit buses (So, 2 for 128bit and 4 for 256bit). Overall, though, Aleman is correct. The total is still 128bit. However, I'm not sure if it's full duplex (which may be what mrgtir is referring to).

@Aleman: I don't think we're overlooking the "bit" of the bus. After all, both consoles are in the same boat when it comes to the bus size. It's not a disadvantage to have a 128bit bus when your competition also has the same. It's going to be a bottleneck for both consoles, no doubt, but we'll see how things work out.


The GDDR3 gets adressed by 2 64bit memory controller while the XDRAM gets adressed by 1 64 bit memory controller.
1400MHz x 128bit : 8 = 22,4GB/s
3200MHz x 64bit : 8 = 25,6GB/s
The Xbox 360 doesn't have any bandwidth problems because all bandwidth hungry effects (HDR, MSAA etc.) gets placed into the eDRAM while textures, sound, CPU processes get placed into the GDDR3 where 22,4GB/s is sufficient of bandwidth for.



Absolutely agree. Having seen that Sony do watch & learn from there competitors & listen to developers, a solution to address this shortcoming is a real possibility.

This area needs to be covered & don't be surprised if they do a similar patch solution to XBOX360. In the event this happens - before people cry COPYCAT i believe this type of solution was previously implemented by ATI for Nintendo Gamecube.

Alternatively to using the ATI solution, i think it would be plausible for Sony to just add more cache to the RSX.

Also many people are guessing that more RAM is one of the upgrades with are likely to see with the RSX rather than any other changes.

What are the Pro's & Con's of:


1. Just adding to GDDR3 ?

2. Just Adding to the VRAM ?



Archy


Putting eDRAM on the RSX certainly would have helped a lot.
However the biggest problem of the PS3 is the fact it's graphics has to be spread onto the 2 different memory types with different bandwidths.

I've said it before and I'll say it again.
Sony needs to lower the XDRAM to 128MB and put in 512MB GDDR3 with the memory controller fully dedicated to the GDDR3.
You'll then get a 256bit memory interface giving you memory bandwidth (1400MHz x 256bit : 8bit) 44,8GB/s instead of 16,5GB/s on the XDRAM and 22,4GB/s on the GDDR3 which you can't add up by the way like Sony and Microsoft PR like to do.
Developers because of this will have a much easier time adressing the memory since it's 1 pool with the same bandwidth and because the bandwidth is higher there's more bandwidth for the framebuffer.
Effects like MSAA and HDR also will be much better to do with less of a beating on performance.
So no XDRAM link with the RSX and more GDDR3 with a 256bit memory interface.

Too bad we won't see this happen though as Sony always likes to do things the hard way and make developers hate their jobs.

1. Adding GDDR3 will enable developers to use more or higher resolution textures.
Staying with a 128bit memory interface to the GDDR3 won't allow you to do more MSAA and better HDR however.

2. Adding VRAM like eDRAM as with the Xenos would be great as you won't have bandwidth problems.
Also MSAA and HDR won't eat of your GDDR3 leaving you with less RAM for textures.

IMO adding 10MB or more eDRAM similar to that of Xenos would be better than adding more GDDR3 with the same 128bit memory interface.
Best ofcourse would be both.

[archy141]

256 bit bus upgrade would be ideal but probably too expensive & difficult to create -Probably same reason why Microsoft also avoided it.

I also think that the eDram solution would work great but again on the very expensive side.

What about increasing Cache on RSX as suggested by some. Can that help & how ?


Archy

[xen]

Not quite in the same boat. The 360 uses edram to alleviate bus traffic.
When your not having to write everything back to the memory, the 128 bit bus becomes less of a problem. If anything, some embedded ram like the 360 would be a nice feature to add to the RSX.

To keep costs down and enable better die strinks down the line both console GPUs have been designed to use 128bit busses instead of the 256-512bit busses found in modern GPUs

To help compensate both GPUs are designed to use two busses for total working bandwidth.

On Xenos
1 bus to GDDR3 for textures, geometry, front buffer
1 bus to EDRAM for AA, stencil and Z tests, blending (this is the xenos backbuffer)

On PS3
1 bus to GDDR3 for textures, geometry, front/back buffer, AA, Z depth, blending
1 bus to XDRAM for textures, geometry, front/back buffer, AA, Z depth, blending

You can see a difference here. The bus to EDRAM on xenos is its back buffer. You cant hold textures or geometry here. This area is for a rendered frame only ready to be processed and copied back to GDDR3 to display. Meaning all textures, geometry, texture filtering, shaders etc has to come from GDDR3.

Whereas on PS3 both busses to RSX can provide it with textures, geometry, texture filtering, shaders etc

This would mean during rendering* PS3 would have upto twice as much bandwidth for shading and textures than 360.

upto 48GB from both busses to RSX
upto 22.4GB from the GDDR3 bus to xenon

Please remember this is during the shader fragment stage of a frame before applying AA etc. And that you cant hold textures in Xenos EDRAM

And before someone comes in saying
"yeah but XDR bandwidth is shared with Cell" or
"RSX has to go through Cell to get to XDR"

please refer to how Xenon has to go through the GPU to get to its Ram. And that its textures/shaders/geometry bandwidth is shaded with the CPU.

* the rendering of shader fragments, applying textures, shadows, texture filtering etc basically everything you see before the AA stage.

[nick²]

Everyone keeps saying how difficult it is to make a console with a 256-bit graphics bus... All it is, is copper on a PCB. Sure there is a little extra cost to it, but people overblow it... They did it to save money on the memory. a 128-bit memory setup will cost less because they can use GDDR3 that works at half bandwidth (effectively at half speed) per pin.

Also, the main system XDR won't be used much, and will not provide much extra bandwidth. Firstly, it cannot provide 25 GB/s to the RSX Its clocked in a bit slower due to having to go through the cell's memory controller (I think it was 18 GB/s?), not to mention the cell will be using it as well. You also have to average the bandwidth, not add.. Most of the system memory will be used by applications, so if you have 50 MB of data to spare in XDR, you are lucky. In practice, it probably will not help the memory bandwidth situation a great deal, and in some situations could actually turn out slower to do this way. You can pretty much guarentee they will be filling up the GPU memory before they touch the system memory at all.

[xen]

Everyone keeps saying how difficult it is to make a console with a 256-bit graphics bus... All it is, is copper on a PCB. Sure there is a little extra cost to it, but people overblow it... They did it to save money on the memory. a 128-bit memory setup will cost less because they can use GDDR3 that works at half bandwidth (effectively at half speed) per pin.

Also, the main system XDR won't be used much, and will not provide much extra bandwidth. Firstly, it cannot provide 25 GB/s to the RSX Its clocked in a bit slower due to having to go through the cell's memory controller (I think it was 18 GB/s?), not to mention the cell will be using it as well. You also have to average the bandwidth, not add.. Most of the system memory will be used by applications, so if you have 50 MB of data to spare in XDR, you are lucky. In practice, it probably will not help the memory bandwidth situation a great deal, and in some situations could actually turn out slower to do this way. You can pretty much guarentee they will be filling up the GPU memory before they touch the system memory at all.

I thought including a 256bit bus increases the number of pins on the chip. Therefore making it harder to make smaller.

I'm getting tired of people saying XDR cant be used by RSX.

There's 2 things to consider.

Capacity which is developer dependent. So a developer can divide the main ram anyway they desire. Code size is generally much smaller than graphic size remember.

Bandwidth which is developer dependent. So a developer can put as many textures or geometry they want into main ram for RSX to use. With bandwidth been shared with Cell.

The point is PS3's memory and bandwidth system has been designed to allow RSX full access to both memory pools simultaneously during rendering.

[bluvi69]

i just want to ask to thread starter, since you said you dont have any console atm,
how suddenly you like about console anyway, are you pc gamer ?

just funny thing, i have xbox, ps2, dreamcast, and now x360
but sometimes other user think i am just xbox fanboy.

and you defended and hyping PS3, and dont have any console, especially ps1 or ps2 ??
woow...

[Scribe]

I believe the RSX will have at least one trick up its sleave, though i really don't know what. However in supporting HDMI 1.3, it would have to be able to supply a lot more speed to fully utilise the associated features.

I think i remember in a more modern announement from sony is that the RSX is "sorta" based on shader model 3.0, asin it's pretty much compatible, but because they don't have to support directx they've changed the spec somewhat.

[nick²]

I thought including a 256bit bus increases the number of pins on the chip. Therefore making it harder to make smaller.


more little lines in a smaller space is of course harder to do... but its not THAT hard. They do it all the time... The PCB is a very small cost of any GPU... Its just fiberglass and copper... Once its layed out, machines easily produce it... The real cost is the design of the GPU and the memory...




I'm getting tired of people saying XDR cant be used by RSX.

There's 2 things to consider.

Capacity which is developer dependent. So a developer can divide the main ram anyway they desire. Code size is generally much smaller than graphic size remember.

Bandwidth which is developer dependent. So a developer can put as many textures or geometry they want into main ram for RSX to use. With bandwidth been shared with Cell.

The point is PS3's memory and bandwidth system has been designed to allow RSX full access to both memory pools simultaneously during rendering.

Well capacity is the only real use... although I have to disagree. 256 megs of texture memory is a lot to use up, but 256 megs of system memory is not a lot to use up at all...

On the bandwidth note, there really is not much of an improvement, and a chance at a lower bandwidth.

This is easily illustrated by the following points:

1) GPU to XDR bandwidth is not 25 GB/s, I believes its about 18 GB/s. (real vs theoretical).

2) Cell will potentially be using a lot of that 25 GB/s of bandwidth. (real world scenarios)

3) The amount of usable XDR will probably be quite small, as the system memory is easily filled up... 200 megs is easily swallowed by a game app.

4) Depending on what you need, and where from, the processing may be slower or faster. Lets say the GPU is currently processing a scene, and currently this scene is using 60 megs out of the 250 MB of texture data for cached in the memory. This is very common, since in any given scene you only see a small portion of the map's textures. Say you are in a room, and you have the texture for two enemy models, a few wall and floor textures, and a few object textures. This is a very small portion of the total amount of textures cached in memory for the entire map. There is plenty you aren't seeing on the map that won't get rendered. The GPU would only need the textures of the objects in the scene, and not all of the other information.

So naturally, the GPU requests the data for this 60 MB of data. Where is the data? Well, if its all in the XDR, you will get the 60 MB/s, but at the best rate the XDR can give. We know, it is at maximum 18 GB/s. It could be much worse. I wouldn't be suprised if its at 14 GB/s or so in game, maybe as low as 8 GB/s. So now, in this case rendering the frame completely from XDR could be exceptionally slow.

Now, lets say that 20% is stored in the XDR, and 80% in the GDDR3 (a more fair representaiton of how the data would be structured). 80% would be retrieved at 22.4 GB/s, while the rest is retrieved simultaneously at 12 GB/s. Averaging the (total size / time) would yield something like 26-27 GB/s.

The key is that most should be coming from the GDDR3. If more than half comes from the XDR, you could end up with something slower.. much slower...

The obvious problem with this is that it becomes much more complicated if you do use the XDR. If you do happen to store ALL textures for one room in XDR, you get a very big potential bottleneck. If the CPU is doing a lot of work, and requesting a lot of data from memory, you could go down to a crawl in such a scene. The developer would have to make sure this isn't the case, and optimizie where each set of data goes.

So, in essence, when done right the XDR will give a small boost in bandwidth, but nowhere close to the touted 48 GB/s. It will certainly be under 30 GB/s in almost every scenario. In many cases though, it could provide absolutely no benefit, and in the worst cases it could actually bottleneck it.

This is not a solution to the bandwidth problems... Its no better off bandwidth wise than the 360, which is equally crippled in this respect...

Why is this important? Well simply because a lot of effects require memory bandwidth. 4x AA simply requires a lot of memory bandwidth. HDR simply requires a lot of memory bandwidth. You would have to do these things after rendering the rest of the scene. That extra bandwidth would have provided a lot of extra eye candy... I don't think either system can truly pull these things off well, without suffering poor FPS.

[Jugix]

Nick2, I would really love to see some technical specifiations to back up all those guesses and predictions. Saying things like:"XDR is no use to RSX because Cell takes all the bandwith in real world scenarios!" is not something I would call an technical explanation.

Why is it that high-end processors on PC need only 3GB/s memory bandwidth to feed athlon64 FX-60 with physics, game data etc in F.E.A.R, HL2, etc. etc. but on PS3 case XDR memory is no good to RSX altough it has +20GB/s bandwith to it! I don't see Cell doing THAT much physics work that it would take 20GB/s without a hickup!

Just give me your best shot! Thanks! :)

[Aleman]

The AGEIA PPU uses high bandwidth GDDR3 for physics calculations.

[xen]

I think you are well over estimating how much ram will be used for game code to be honest. You do realise how little ram consoles have had in the past. Only 32MB for PS2 and i just cant see game code going up that many orders of magnitude in a next gen console game.

But again how the memory is split is upto the developers wishes

it could be 80/20 in favour of graphics
it could be 50/50 split
it could be 20/80 in favour of code (highly unlikely)

The point is the main ram in PS3 is general purpose and can be used however the dev chooses

I agree most graphics should come from GDDR3 memory. Which is why you have a dedicated bus to it for RSX. But you also have a bus to XDR where you can store as many textures or geometry as can fit dependent on what devs feel the games needs.

And bandwidth to main memory would be divided between RSX and Cell

So if as you say you get say 30GB/sec or under total from both ram pools thats still far more than what you have going to Xenos for textures, shaders, geometry etc


Now, lets say that 20% is stored in the XDR, and 80% in the GDDR3 (a more fair representaiton of how the data would be structured). 80% would be retrieved at 22.4 GB/s, while the rest is retrieved simultaneously at 12 GB/s. Averaging the total size / time would yield something like 26-27 GB/s.

And again if we compare it to xbox360 and use your own example of how much bandwidth a cpu consumes and the sharing of bus bandwidth with the cpu.

You have 12 GB/s to Xenos for textures, shading, geometry (taking away bandwidth the cpu uses from GDDR3 as your example)
and have 26-27 GB/s to RSX for textures, shading, geometry

again i'm only talking about the part of rendering a scene before applying the AA etc where the EDRAM on xenos would be used.

[Guilty Bystander]

256 bit bus upgrade would be ideal but probably too expensive & difficult to create -Probably same reason why Microsoft also avoided it.

I also think that the eDram solution would work great but again on the very expensive side.

What about increasing Cache on RSX as suggested by some. Can that help & how ?


Archy


RSX already has a 256bit memory interface it's just that it's devided over the GDDR3 and XDRAM.
Xenos doesn't have a 256bit memory interface because it doesn't need it all bandwidth hungry effects go into the eDRAM's bandwidth which is 256GB/s.
By the way implementing a 256bit memory interface (quad 64bit) isn't that difficult or costly at all.

By the way what's is this cache you speak of?
GPU's don't have cache only the Flipper (Gamecube GPU) and Xenos have eDRAM for framebuffer if you want to call that cache.



To keep costs down and enable better die strinks down the line both console GPUs have been designed to use 128bit busses instead of the 256-512bit busses found in modern GPUs


There are no desktop videocards out with a 512bit memory interface.



This is not a solution to the bandwidth problems... Its no better off bandwidth wise than the 360, which is equally crippled in this respect...

Why is this important? Well simply because a lot of effects require memory bandwidth. 4x AA simply requires a lot of memory bandwidth. HDR simply requires a lot of memory bandwidth. You would have to do these things after rendering the rest of the scene. That extra bandwidth would have provided a lot of extra eye candy... I don't think either system can truly pull these things off well, without suffering poor FPS.


Well that's not true.
Xenos has the eDRAM to put all bandwidth hungry effects (HDR, MSAA, Z-testing, Alpha blending etc.) so the 128bit memory interface is more than sufficient for storing textures, geometry, CPU code, sound etc.

MSAA and HDR are no problem when using the eDRAM effectively.



Why is it that high-end processors on PC need only 3GB/s memory bandwidth to feed athlon64 FX-60 with physics, game data etc in F.E.A.R, HL2, etc. etc. but on PS3 case XDR memory is no good to RSX altough it has +20GB/s bandwith to it! I don't see Cell doing THAT much physics work that it would take 20GB/s without a hickup!

Just give me your best shot! Thanks!


Well it's 6,4GB/s with PC's doing only CPU code, geometry and sound can be easily put that 6,4GB/s.
The highend videocards nowadays have about 40GB/s which is where textures, HDR, MSAA etc is stored in.

Like I and Nick2 said the XDRAM bandwidth is NEVER fully accessible by RSX.
RSX can only communicate with the XDRAM with 16-18GB/s MAX.
You can read about that in the infamous Cell bandwidth to GDDR3 sheet.

[EbonySeraphim]

Oh goodness…why has this discussion been put out again? (more motivation to get 2nd revision for PS3: Explained thread done).


@GuiltyBystander
XBox360 has no bandwidth problems due to eDRAM? Really now? That’s amazing! I wasn’t aware that an entire 22.4GB/s or bandwidth was freed up because AA, HDR, and z-buffering can be done on eDRAM. How goes that performance hit when developers try to use AA and HDR at the same time that won’t fit the entire frame buffer in eDRAM forcing tiling to be used? Why don’t you go ask the Half Life 2 developers what it performance it costs to the system and what they had to do to get around it. Every solution in the computing world has a cost somewhere else unless there is just extra or unused hardware at the same time as the operation that needs to be done. Now I’m not saying eDRAM doesn’t alleviate any bandwidth problems, because it undoubtedly does make the situation better for some frame buffer operations. But to think that it makes the 22.4GB/s shared between the GPU and CPU scotch free from any issues is to be deluded of reality.

Speaking of bandwidth on the XBox360, how does sharing a bus with the CPU handle on the system? While the CPU may not be the higher bandwidth consumer in many games, it’s certainly going to be saturating the GDDR3 memory controller and bus with many small requests that degrades overall latency and bandwidth performance considerably. Maybe that’s why the XBox360 version of games notably are missing anistrophic filtering where the PS3 shows it…

The RSX may only have one explicit “video memory bus” that is 128 bits wide, offering 22.4GB/s to GDDR3; but the bus to the Cell processor(FlexIO) which is connected to XDR RAM is entirely separate. Can this bus be used to alleviate bandwidth? It sure as hell can. Somehow, I don’t see vertex data and graphics commands taking up 35GB/s bandwidth between the Cell and RSX. Last time I checked, Sony wasn’t stupid enough to put such a fat bus between two major components only to have it not be utilized. Last time I checked, PC GPUs could use system memory as graphics memory freely(at least for textures, or not frame buffer also) – and the only thing making it non-viable for heavy use for most PCs, was the bandwidth on that bus. That bus is typically PCI-E 16x or AGP 8x running at 4GB/s – much slower than FlexIO. Perhaps this why Sony, who put eDRAM on the PS2’s Graphics Synthesizer, didn’t see it fit to use eDRAM again in the RSX. NVidia seems to agree, who is no rookie in making GPU for embedded systems. They both saw the fat bus between the RSX and Cell and aren’t dumber than you. If they say “this bus can be used effectively and bring benefits to the table” I think I trust them over you and nick^2 saying “it can’t be used beneficially.”

eDRAM solutions aren’t new. If it was the golden answer to bandwidth problems, don’t you think you’d see it being used everywhere? Here’s a clue – Microsoft’s solution isn’t the best solution for alleviating bandwidth for all GPUs. Neither is Sony/nVidia’s solution to have a fat bus between CPU and GPU. The most efficient solution for the “ultimate gaming console” would have so many parts that the cost of the GPU would be sky high, its heat output would be sky high, and it would take up way too much space. You should be more than convinced eDRAM isn’t a cure-all when Sony used it with the PS2’s GS, the graphics chip they made themselves, and yet this time around, see no use for it. You think Sony is stupid enough to forget a consideration to do something they did before?

Your “Sony needs to do this” list is very impractical for the same reasons mentioned above.
-256bit buses are too big and costly to allow the price of the system to drop.
-The bandwidth “lost” to being on a 128 bit bus, and not having any eDRAM for frame buffer effects is addressed by putting a fast bus between the Cell and RSX. I say again: this bus can be used to alleviate bandwidth off of the main video memory bus for more than just vertex data and commands. Read up if you are still in denial: http://www.ps3forums.com/viewtopic.php?t=25846
-More video memory != higher resolution textures for free. When you bump up texture resolution, you are also increasing the amount of GPU execution resources needed to apply filters on those textures(AF), and execution resources needed to apply those textures to vertices(texel fill rates). Last time I checked, the Xenos only had 16 texture units and the RSX had 24, allowing it to support higher resolution textures without reaching this bottleneck. Considering textures, the RSX certainly has room to move into XDR RAM for texturing just like PC GPUs do. They can do this with less worry about bandwidth too since they have much more bandwidth than PCI-E and AGP buses.


Well that's not true.
Xenos has the eDRAM to put all bandwidth hungry effects (HDR, MSAA, Z-testing, Alpha blending etc.) so the 128bit memory interface is more than sufficient for storing textures, geometry, CPU code, sound etc.

MSAA and HDR are no problem when using the eDRAM effectively.
This is where you are mistaken Frame buffer operations constitute some of the bandwidth hungry effects. Not all.


Like I and Nick2 said the XDRAM bandwidth is NEVER fully accessible by RSX.
RSX can only communicate with the XDRAM with 16-18GB/s MAX.
You can read about that in the infamous Cell bandwidth to GDDR3 sheet.
Perhaps you should read my reply to nick^2 too then. It would be nice if either if you two could pull up some credible sources that show this to be true and is limited to the situation between the Cell and GDDR3. Also, you both might want to consider that what’s most important is the RSX being able to pull from XDR RAM. The only thing the Cell might push to the RSX or GDDR3 is vertex data. Textures will just be held in XDR, while the RSX makes requests for the data.

So go ahead and delude yourself in thinking that one or two aspects(bus width and eDRAM) of a system can make a whole system will reap obvious improvements. In the computing world, that is not how things work. Consider the whole system at once. GuiltyBystander, you’re guilty of thinking eDRAM is a golden solution. Get a clue as to why this isn’t the best solution when there aren’t any top performing GPUs on the market today that use it.

nick²:
Yes, the Cell will be using some of that XDR RAM. But won’t the XBox also be using some of that 512MB’s RAM too? Just like Xen has said, you should probably stop applying certain considerations to make one console look worse when it applies to the other.

Also what matters with the 256bit bus limitation is the cost. Of course it can be done. Microsoft could have done it on the Xenos, and Sony could have done it on the PS3. A lot can be done with technology today to vastly improve many limiting situations. It’s just that currently, the cost to put those solutions in products are too high for most consumers. It’s kind of like why PS2 slim was “not possible” when the PS2 first came out. If Sony had made the PS2 that small on release, it would have been over $1000 at launch probably. But Sony didn’t aim to make it that small, and thus waited to make it smaller later, when it could be more affordable.

1) Your assessment of real vs theoretical bandwidths apply to all memory busses and situations. Why are you singling out the bus between XDR and the RSX? Unless you apply the same “theoretical” nerf to all memory busses, all are assumed to be the reported maximums.
2) If the Cell is using up 25.6GB/s bandwidth, why isn’t the XBox360 CPU using this much bandwidth? If it were to be using that much bandwidth, then there would be -4.2 GB/s bandwidth left for the Xbox360 GPU or the XBox360 CPU would be bottlenecked for bandwidth to RAM. Did you pull that real world scenario out of your ass?
3) Yes yes, again with the imbalanced considerations. So wouldn’t the 512MBs RAM on the XBox360 CPU suck up 200MBs of RAM just as easily?

Nick, I suggest you actually learn about the Cell because some of what you say is completely false and I’m sure comes from your own speculation. You can’t speculate wildly on bandwidth limitations. Or else, I might as well speculate the XBox360 memory controller realistically can only sustain 10GB/s bandwidth. Then I could go off on a field day about how weak the XBox360 is couldn’t I? Do us all a favor, and let’s not involve speculation where it shouldn’t be placed to favor of one over the other. You have been called out for bsing your “facts” by me, Xen, and Judix. Time to get your facts straight.

I’m not even going to get in depth to discredit your more “in-depth” analysis since they were based off of your already flawed speculation. It’s very dangerous to more solve problems based on already inaccurate solutions. The errors propagate and result in an extremely inaccurate end result.

The Cell:
All memory controllers on all elements on the Cell processor have 25.6GB/s bandwidth on the EIB (Element Interconnect Bus). There are NO bottlenecks below this number between any two elements, and that includes the bandwidth between the FlexIO controller and XDR memory controller. Additionally, the memory controller for XDR RAM is independent of other elements in the Cell. It takes NO PPE or SPE cycles to access XDR RAM.

Understand the design of the thing before you completely butcher it like you just did.

[mrboo]

EbonySeraphim You should go easy on people :P

[nick²]

Nick2, I would really love to see some technical specifiations to back up all those guesses and predictions. Saying things like:"XDR is no use to RSX because Cell takes all the bandwith in real world scenarios!" is not something I would call an technical explanation.

Why is it that high-end processors on PC need only 3GB/s memory bandwidth to feed athlon64 FX-60 with physics, game data etc in F.E.A.R, HL2, etc. etc. but on PS3 case XDR memory is no good to RSX altough it has +20GB/s bandwith to it! I don't see Cell doing THAT much physics work that it would take 20GB/s without a hickup!

Just give me your best shot! Thanks! :)

This isn't a guess or prediction. I gave three paragraphs of why it would not be of much use. Its in plain text, and if you disagree, please say why.

As far as AMD chips, I believe they can pull over 8 GB/s now... I just saw a few benchmarks at tomshardware, where they actually overclocked an FX-62 and had it pulling 10 GB/s.

Cell does need a great deal of memory bandwidth if it is taking full potential of its 7 SPE's and PPE. Each one has to have access to memory, and 7 mouths to feed can require a lot of bandwidth.

[nick²]

I think you are well over estimating how much ram will be used for game code to be honest. You do realise how little ram consoles have had in the past. Only 32MB for PS2 and i just cant see game code going up that many orders of magnitude in a next gen console game.


Well, I'm not sure what you are basing this on. Sure they did ok with 32 MB of ram, but the games are constrained by this 32 MB of ram. In 2006, anything you play on the PS2 is vastly inferior to modern technology.

You can check the memory footprint of any game on the PC by hitting CTRL+ALT+DEL. There are games that exist, which can eat up to 600-800 MB of ram. Now people like to go and say there is "bloat" from running on a PC, but really most of the space is taken up by the data structures built by the game. You have map data, entity data, etc etc. There is far more storage space needed for all of this data, and most of that 600-800 MB is being used by legitimately needed data. Now, often 200-300MB is used for redundant storage of textures into system memory... but the rest of it, save about 30-50 megs, is needed data.

The point is, it is very difficult to get games like call of duty 2 down to a memory footprint of less than 256 MB... It takes a lot of trimming and a lot of tweaking.



But again how the memory is split is upto the developers wishes

it could be 80/20 in favour of graphics
it could be 50/50 split
it could be 20/80 in favour of code (highly unlikely)


So , are you suggesting that it likely that modern games will only need 60-80 megs of ram? Not a chance.



I agree most graphics should come from GDDR3 memory. Which is why you have a dedicated bus to it for RSX. But you also have a bus to XDR where you can store as many textures or geometry as can fit dependent on what devs feel the games needs.

And bandwidth to main memory would be divided between RSX and Cell

So if as you say you get say 30GB/sec or under total from both ram pools thats still far more than what you have going to Xenos for textures, shaders, geometry etc


Well there is much more to it than just texture memory. Most of the bandwidth is used for other processes (color and z-data). I'm not sure how effective the XBOX's EDRAM is at this... I can't really say, since they haven't released much information on it...

I'm not really sure which system makes the best use of bandwidth... All I know, is either way, both systems are relatively low in terms of bandwidth...



And again if we compare it to xbox360 and use your own example of how much bandwidth a cpu consumes and the sharing of bus bandwidth with the cpu.


Well, I think you are assuming I am arguing in favor of the XBOX 360. I am not. The 360 is probably in the exact same shape as the PS3 here.



You have 12 GB/s to Xenos for textures, shading, geometry (taking away bandwidth the cpu uses from GDDR3 as your example)
and have 26-27 GB/s to RSX for textures, shading, geometry


Yes, you have the 12 GB/s for textures.. that is probably right on the nose. However, the mistake is that the RSX doesn't have to perform the bandwidth intensive work on color and z-buffer data that is offloaded to the XBOX edram... The 27 GB/s is all it has to do everything...

Once again, no one is quite sure how effective the EDRAM is at preserving memory bandwidth. Less than half the bandwidth is used for things other than textures, so if it does a very good job, than that 12 GB/s is enough... It may or may not be.. I don't know.. The effective bandwidth is probably very close...

Either way, both systems could've used a bit more memory bandwidth..

[nick²]

nick²:
Yes, the Cell will be using some of that XDR RAM. But won’t the XBox also be using some of that 512MB’s RAM too? Just like Xen has said, you should probably stop applying certain considerations to make one console look worse when it applies to the other.


See, this is what pisses me off about fanboys. They always assume you are secretly conspiring to make one console look bad and the other one look bad... Maybe its hard to comprehend, but they BOTH lack memory bandwidth. EDRAM helps quite a bit with the XBOX 360, so it can afford to share memory bandwidth... Otherwise it would absolutely crawl... However, the EDRAM is there to offset the lack of dedicated memory bandwidth from the GDDR3... Nevertheless, its still not going to achieve much more than the RSX will... It probably isn't going to be much different IMO...



Also what matters with the 256bit bus limitation is the cost. Of course it can be done. Microsoft could have done it on the Xenos, and Sony could have done it on the PS3. A lot can be done with technology today to vastly improve many limiting situations. It’s just that currently, the cost to put those solutions in products are too high for most consumers. It’s kind of like why PS2 slim was “not possible” when the PS2 first came out. If Sony had made the PS2 that small on release, it would have been over $1000 at launch probably. But Sony didn’t aim to make it that small, and thus waited to make it smaller later, when it could be more affordable.


Right, both consoles saved about $50... I do think its funny where the skimping came down upon... Arguably, it was on one of the most important parts..



1) Your assessment of real vs theoretical bandwidths apply to all memory busses and situations. Why are you singling out the bus between XDR and the RSX? Unless you apply the same “theoretical” nerf to all memory busses, all are assumed to be the reported maximums.


Because in order to use the XDR, the RSX would need to go through a second memory controller (on the cell). Everything else can achieve close to its theoretical bandwidth.



2) If the Cell is using up 25.6GB/s bandwidth, why isn’t the XBox360 CPU using this much bandwidth? If it were to be using that much bandwidth, then there would be -4.2 GB/s bandwidth left for the Xbox360 GPU or the XBox360 CPU would be bottlenecked for bandwidth to RAM. Did you pull that real world scenario out of your *expletive deleted*?


Well, think about it. In one scenario you have 8 independent processing units, in the other you have 3. The more cores, the more bandwidth you need... period. You can't starve anyone, otherwise you will end up with major bottlenecks.



3) Yes yes, again with the imbalanced considerations. So wouldn’t the 512MBs RAM on the XBox360 CPU suck up 200MBs of RAM just as easily?


Yes.. and?



Nick, I suggest you actually learn about the Cell because some of what you say is completely false and I’m sure comes from your own speculation. You can’t speculate wildly on bandwidth limitations.


Its not speculation, it simple logic from using released specs. Its very basic reasoning. But hey, you don't have to believe what I say. You can believe you are right, and that the RSX can really achieve 48 GB/s of memory bandwidth... Ignorance is bliss, right?



Or else, I might as well speculate the XBox360 memory controller realistically can only sustain 10GB/s bandwidth. Then I could go off on a field day about how weak the XBox360 is couldn’t I?


Sure you could say that, but the difference between my statement and yours is that your statement doesn't make any sense.. The numbers I'm quoting are from my recollection and explanation of what was released at devstation...



Do us all a favor, and let’s not involve speculation where it shouldn’t be placed to favor of one over the other. You have been called out for bsing your “facts” by me, Xen, and Judix. Time to get your facts straight.


No, I just know more.



I’m not even going to get in depth to discredit your more “in-depth” analysis since they were based off of your already flawed speculation. It’s very dangerous to more solve problems based on already inaccurate solutions. The errors propagate and result in an extremely inaccurate end result.


In other words you don't understand it enough to talk about it.



The Cell:
All memory controllers on all elements on the Cell processor have 25.6GB/s bandwidth on the EIB (Element Interconnect Bus). There are NO bottlenecks below this number between any two elements, and that includes the bandwidth between the FlexIO controller and XDR memory controller. Additionally, the memory controller for XDR RAM is independent of other elements in the Cell. It takes NO PPE or SPE cycles to access XDR RAM.

Understand the design of the thing before you completely butcher it like you just did.

Right, but this is not an SPE or PPE.. its an RSX. An RSX using cell's memory controller to access the XDR memory. A little reading comprehension might help.

[Mr.Smith]

The 360's Xenos is equal to a an ATI x1900 and the price of those cards are just as much as an Xbox360. If anything the RSX is equal to the Geforce 7900 GT.

[mrboo]

nick² im surprised your posting here after Ebony pwned you hard.

[Kentatsu]

Hi everyone, im new here and umm yea i like this rsx conspiracy thread lol. I'd like to help contribute by supporting Archy's speculation with this:


Atari CEO Bruno Bonnell is a wishy-washy kind of guy. On the one hand, he says in an interview with GamesIndustry.biz that his company has no plans to release any games for the PS3 until 2007, citing a lack of information about the console as the main reason.

Fair enough, but he then goes on to say how fab the PS3 is going to be, "picking up the high-end of the market, the hardcore, passionate fans." Bonnell confirms that there will be Atari titles on both the Wii and the 360 this year.

http://www.1up.com/do/newsStory?cId=3152599


What do you all think about this?? ;o

[nick²]

nick² im surprised your posting here after Ebony pwned you hard.

I'm suprised there are so many children on these boards using words like "pwned".

[razorblade416]

nick² im surprised your posting here after Ebony pwned you hard.

I'm suprised there are so many children on these boards using words like "pwned".

You're suprised that there are kids on a game console forum? :suspect:

[nick²]

nick² im surprised your posting here after Ebony pwned you hard.

I'm suprised there are so many children on these boards using words like "pwned".

You're suprised that there are kids on a game console forum? :suspect:

No.. I was mostly ridiculing his grammar...

[Organic_Shadow]

It doesn't matter what the slight bottlenecks or flaws of each console's hardware setup is. The developers are being PAID to figure this junk out, and make the best use of it possible. If for some odd reason SquareEnix can't get around a certain hardware limitation and some guy in a forum CAN, then that's SE's problem, NOT SONY's.

Some developers are laughing at these doom and gloom nay sayers of the PS3 hardware, while other devs are grunting like cavemen and whacking the PS3 dev kits with sticks.

[nick²]

Some developers are laughing at these doom and gloom nay sayers of the PS3 hardware, while other devs are grunting like cavemen and whacking the PS3 dev kits with sticks.

Hehe, like that imagery...

Please abide by the 5 word minimum rule. - The_One

[EbonySeraphim]

My sincerest apologies nick². My assumption was that you were also in a strict console comparative mindset along with GuiltyBystander. Now that I have that removed, let’s move on.


Because in order to use the XDR, the RSX would need to go through a second memory controller (on the cell). Everything else can achieve close to its theoretical bandwidth.
I see where you are trying to get at with this, although the FlexIO bus is no where near capping the 25GB/s bandwidth as it has even more bandwidth than XDR RAM. When the transfer hits the EIB on the Cell, I don’t see the 200GB/s bandwidth there being a problem. And finally, when the transfer hits the XDR memory controller, it has a 25.6GB/s bandwidth cap there. No parts of the chain are bandwidth limited, and any of the busses and memory controllers(EIB and FlexIO) outside of that 25.6GB/s memory controller on XDR have over the required bandwidth to handle overhead. The biggest problem in the whole setup is latency as a transfer would go through the FlexIO memory controller and also XDR’s memory controller. But considering the FlexIO memory controller can handle more bandwidth than XDR’s memory controller, I’m not quite convinced where the capping to 18GB/s would take place, unless it applies to XDR’s memory controller – and in turn, apply to any element on the Cell sending and receiving data.


Its not speculation, it simple logic from using released specs. Its very basic reasoning. But hey, you don't have to believe what I say. You can believe you are right, and that the RSX can really achieve 48 GB/s of memory bandwidth... Ignorance is bliss, right?
I never claimed it would always sustain the max bandwidth did I? But if you’re going to presume to apply real world considerations and mention the real performance vs maximums, you should apply the same to all computing systems. Even if you can justify the added bandwidth limitations of the RSX <-> Cell communication, that 18GB/s number is probably not only considering the “extra” overhead, but it already has considered a base limitation that all buses go through. In other words, the bus maximum from XDR is 25.6GB/s. Real performance from XDR might be 21~22GB/s. From that already “normalized” performance number, the two memory controllers are factored in for an additional nerf down to 18GB/s.

I’m just clarifying this issue because between RSX and Cell alone, you are making the added bandwidth out to be very small compared to the whole system’s performance which do not have the added “real world” performance numbers applied to them. If the bus between the RSX and Cell would only offer 18GB/s with the overheads of the two memory controllers, how fast is the bus that the RSX has to GDDR3 memory in real world performance? 20GB/s? That would be much closer to the max 22.4GB/s compared to the RSX <-> Cell scenario going from 25.6GBs -> 18GB/s wouldn’t it? The point is that this added bus that gives something to the RSX on the scale of its performance.


Sure you could say that, but the difference between my statement and yours is that your statement doesn't make any sense.. The numbers I'm quoting are from my recollection and explanation of what was released at devstation...
It would help to provide a better explanation of your numbers. I think I see where you were coming from with them though. Keep in mind, its hard to get away with focusing down on this bandwidth numbers, saying the PS3 is bandwidth limited without the context of the console war, and even abstracted from the bandwidth of other busses in the PS3 itself. Otherwise, you could say that any computing system is limited simply by pointing out that the actual numbers are lower than the maximums reported. These numbers are given for simplicity’s sake.


In other words you don't understand it enough to talk about it.
No, I fully understood it and read it. But it was based on something I believed to be false to begin with. So why waste time traveling down the branches when I believed I cut off the root?


Right, but this is not an SPE or PPE.. its an RSX. An RSX using cell's memory controller to access the XDR memory. A little reading comprehension might help.
A section of your text was slightly ambiguous to me. I re-read and have it clarified now. It seemed to me that you implied CPU cycles were being taken by bandwidth demands between XDR and FlexIO.

Other things: (still mostly directly to you nick²)
The Cell’s massive computing ability doesn’t need to be fed by purely “new and different” data from system memory. If the PPE loads a single game object or entity – say an explosion - it can generate many other objects that need complex calculations on chip, without adding a memory footprint anywhere off the chip. Something like the particles from the explosion could be fed to an SPE directly, where it does some physics calculations, and then sends rendering calls to the RSX. From the same explosion on the PPE, sound information may be generated and passed off to an SPE handing sound processing, where that SPE directly sends the output directly to FlexIO for sound output. Storing all of that data in RAM where it doesn’t need to be is a waste of RAM. It is also why PC games use far more RAM than they need to.* Instead of putting all of the information that is ever needed and used in system memory to save on execution resources, the smallest amount should be placed in system memory to take advantage of much greater resources. Additionally, the Cell has 2.2meg on chip memory to allow for such expansion on chip, and the EIB runs at 200GB/s to allow for data to be passed around quickly between the elements and likely result in off-chip actions not limited to placing results in XDR. Ideally the EIB will be used to process massive amounts of data that aren’t purely spawned from system memory, and only read/write minimally from system memory.

*Since PCs have hit around 128MBs of RAM, the saying in the development industry and computer science world is “memory is cheap.” Solutions are implemented everyday where the trade off is made to use more memory in favor of saving execution resources with very little consideration.


nick² im surprised your posting here after Ebony pwned you hard.

I'm suprised there are so many children on these boards using words like "pwned".
LOL. mrboo no one is “pwned” until the dust clears. If anything, I got pwned harder last round, but both nick and I are still standing.

[nick²]

I'm sorry, I don't have time for a full post...

However

I think this would be settled if someone would just pull up the article with the picture from devstation, and the memory bandwidths. There was one very good article with quotes from developers... I believe one confirmed that the numbers added up, and another one said that "no one blinked" when they saw the slide, because its what they expected.

There was a big hoopla that the Cell couldn't write to the GPU memory, but this is the same as in the PS2... and no suprise to anyone..

[EbonySeraphim]

Instead of the Cell writing to GDDR3, I'm sure having the RSX read from XDR(or simply use the data as it) would be a faster solution.

Note that I'm not arguing any absolute numbers any more. It's all relative. Any bus only achieves some percentage of it's max theoretical bandwidth and I'm sure you know that. You basically told me that the bus between RSX and the Cell has added overhead beyond this already existing constraint because it goes through two memory controllers. Assuming I agree with there being an addition issue that drops performance significantly on this bus, I'll say the efficiency of the bus drops yet again by another 30%. Doing the math (operations sequentially applied), it comes out to 56% of the original maximum bandwidth of the bus, which from 25.6GB/s results in ~14.3GB/s bandwidth. That number is fine, although you compared this "minimal extra" to the maximum theoretical bandwidth that the RSX has to GDDR3. What is the point in comparing a realistic number to a theoretical maximum within the same console?

A better comparison to show how much the RSX <-> Cell bus brings to the table would be to add the realistic attainable bandwidths with and without the XDR bus, and then compare the two. I'm just going to assume 80% is the realistic bandwidth you'd get from the max bandwidth of any bus right off the bat:

22.4GB/s to GDDR3 drops to 17.92GB/s with standard bus limitation(80%).
25.6GB/s bandwidth to XDR, you drop to 20.48GB/s with standard bus limitation(80%).
This totals to: 38.4GB/s if we assume the bus between RSX and XDR RAM runs like any other bus.

But then you mention that this bus has an added overhead right? So drop that 20.48GB/s bandwidth yet another 30%(I'm being very generous here) to get to 14.34GBGB/s. So you have:

17.92GB/s to GDDR3 (80%).
14.34GB/s to XDR. (80% then apply another 70% of that result)
This totals: 32.3GB/s.

Removing the XDR link to the RSX leaves you with only 17.92GB/s realistic bandwidth to begin with. If you factor in the added bus's realistic bandwidth, there is 32GB/s bandwidth. That's almost double, and that was adding a number less than your 18GB/s number. So when you say something like:


Also, the main system XDR won't be used much, and will not provide much extra bandwidth. Firstly, it cannot provide 25 GB/s to the RSX Its clocked in a bit slower due to having to go through the cell's memory controller (I think it was 18 GB/s?), not to mention the cell will be using it as well. You also have to average the bandwidth, not add.. Most of the system memory will be used by applications, so if you have 50 MB of data to spare in XDR, you are lucky. In practice, it probably will not help the memory bandwidth situation a great deal, and in some situations could actually turn out slower to do this way. You can pretty much guarentee they will be filling up the GPU memory before they touch the system memory at all.
I do wonder where you are coming from in saying the part that I put in bold. If by "extra" you are referring to how much bandwidth this additional bus can provide on top of the GDDR3 bus; that "extra" amount is 100% - virtually doubling the bandwidth. How is this "not much"? Was it because you compared the 22.4GB/s theoretical added maximum to the 18GB/s realistic maximum on the single bus? Because that's unfair...I'm positive the 18GB/s considers both the limitations of a normal bus, plus the added issue of both memory controllers between the RSX and XDR RAM. If you're going to compare the two, apply the same standard bus limitation to the 22.4GB/s link.


This is not a solution to the bandwidth problems... Its no better off bandwidth wise than the 360, which is equally crippled in this respect...
Really now? If we are talking realistic maximums here, that 22.4GB/s bandwidth, is not 22.4GB/s at all. My rule of thumb could be off, but say 80% is the efficiency - that brings the total bandwidth to 17.92GB/s for both the GPU and CPU of the XBox360. And somehow, the RSX's dedicated 17.92GB/s, and the Cell's dedicated 20.48GB/s make it "not much better"? Please, do explain further. Because I am thoroughly confused. Make sure you aren't comparing realistic numbers to theoretical maximums while you're at it.

Of course using FlexIO for shared video memory isn't a cure all solution either. I was quite clear in addressing GuiltyBystander that no solution is a cure all. I also believe I mentioned to you that any system can be said to be limited relative to something else because things can always be "faster." The best Sony can do is offer easy improvements when they can easily and reasonably be provided. Such is what the FlexIO bus is providing in the case of the PS3. Such is what the eDRAM is doing for the XBox360 GPU. But I guess to you, anything short of a 256-bit memory controller running at 3Ghz+ is crippled?


So, in essence, when done right the XDR will give a small boost in bandwidth, but nowhere close to the touted 48 GB/s. It will certainly be under 30 GB/s in almost every scenario. In many cases though, it could provide absolutely no benefit, and in the worst cases it could actually bottleneck it.
Small boost....almost a 100% increase...small boost...almost a 100% increase...

What's a medium sized or big boost to you then?

[Solaris]

I'm sorry, I don't have time for a full post...

However

I think this would be settled if someone would just pull up the article with the picture from devstation, and the memory bandwidths. There was one very good article with quotes from developers... I believe one confirmed that the numbers added up, and another one said that "no one blinked" when they saw the slide, because its what they expected.

There was a big hoopla that the Cell couldn't write to the GPU memory, but this is the same as in the PS2... and no suprise to anyone..

Is this the image you're on about?

http://www.theinquirer.net/images/articles/PS3_memory_bandwidths.jpg

That chart does show that the cell can write to the GPU memory though, at 4GB/s.

[nick²]

Ok, very quick post.. I will get in more tommorrow.. but your error is in the math.

You don't add the bandwidth. Not everything in the memory is used at one time. You have to use this formula:

(amount of data transferred) / (time)

Let's go back to my example... we have 60 MB of data needed for this one frame (the same data will be used for multiple frames, slightly chaning as one progresses).

Let's say the cell is using 40% of its available bandwidth, leaving 15 GB/s. Due to the inefficiencies of having to use a second hop (GPU memory controller to cell's memory controller) - I'll be conservative here - you lose 2 GB/s. That gives you 13 GB/s for the RSX. I think this is quite reasonable, that the CPU would be using this much memory bandwidth and that there is a 10% loss in bandwidth due to extra latencies.

Now like I said, let's take the worst case scenario - all of the data is in system memory. Now what? All 60 MB/s is transferred at 13 GB/s. This is slower than keeping it all on the GDDR3.

Now lets take a more balanced scenario... 50% on the RSX, 50% on the RSX. 30 MB gets transferred at ~23 GB/s. The other 30 MB gets transferred at 13 GB/s. We have to transfer the same amount of data from both memory banks, but the data in XDR takes twice as long... The way the math works out, its roughly equivalent to transferring the whole thing from the GDDR3 at ~23 GB/s...

The best balance would be somewhere from a 70-30 to a 80-20, which could give you up to 30 GB/s....

[nick²]

BTW, I believe that is the picture I was talking about.. and I was wrong, the cell can't really read from the GPU memory...

However the numbers for measured READ/WRITE from GPU to main memory are lower than what I remember... I guess it depends on how they were measured... but I think that fits what I was saying pretty well..

[Guilty Bystander]

XBox360 has no bandwidth problems due to eDRAM? Really now? That’s amazing! I wasn’t aware that an entire 22.4GB/s or bandwidth was freed up because AA, HDR, and z-buffering can be done on eDRAM. How goes that performance hit when developers try to use AA and HDR at the same time that won’t fit the entire frame buffer in eDRAM forcing tiling to be used? Why don’t you go ask the Half Life 2 developers what it performance it costs to the system and what they had to do to get around it. Every solution in the computing world has a cost somewhere else unless there is just extra or unused hardware at the same time as the operation that needs to be done. Now I’m not saying eDRAM doesn’t alleviate any bandwidth problems, because it undoubtedly does make the situation better for some frame buffer operations. But to think that it makes the 22.4GB/s shared between the GPU and CPU scotch free from any issues is to be deluded of reality.


Xbox 360 certainly has the least bandwidth problems of any console in history.
Well Valve are lazy programmers which is easily shown by the fact Half Life 2 has been on the market 2 years and still they didn't fix the audio stutterbug.
Using tiling with the eDRAM on the Xenos takes a lot of effort (same reason why none done it to date) so using just the GDDR3 would make it nice and simple for Gabe Fatneck and his team to lazy port Half Life 2 to the PS3 and Xbox 360.
It has nothing to do with the eDRAM not being good enough.



Speaking of bandwidth on the XBox360, how does sharing a bus with the CPU handle on the system? While the CPU may not be the higher bandwidth consumer in many games, it’s certainly going to be saturating the GDDR3 memory controller and bus with many small requests that degrades overall latency and bandwidth performance considerably. Maybe that’s why the XBox360 version of games notably are missing anistrophic filtering where the PS3 shows it…


Sure that's why games like Brothers in Arms for example doesn't have AF on both PS3 and Xbox 360.
Doing AF in a small closed enviroment like that arena in the Heavenly Sword and in the Xbox 1 graphicly like title Untold Legends doesn't say anything about other games.
Sonic, BiA, Resistance, Motorstorm, Virtua Tennis, Fatal Inertia and about every other game on the PS3 suffer the same Trilinear or 2-4x AF like most Xbox 360 titles.



The RSX may only have one explicit “video memory bus” that is 128 bits wide, offering 22.4GB/s to GDDR3; but the bus to the Cell processor(FlexIO) which is connected to XDR RAM is entirely separate. Can this bus be used to alleviate bandwidth? It sure as hell can. Somehow, I don’t see vertex data and graphics commands taking up 35GB/s bandwidth between the Cell and RSX. Last time I checked, Sony wasn’t stupid enough to put such a fat bus between two major components only to have it not be utilized. Last time I checked, PC GPUs could use system memory as graphics memory freely(at least for textures, or not frame buffer also) – and the only thing making it non-viable for heavy use for most PCs, was the bandwidth on that bus. That bus is typically PCI-E 16x or AGP 8x running at 4GB/s – much slower than FlexIO. Perhaps this why Sony, who put eDRAM on the PS2’s Graphics Synthesizer, didn’t see it fit to use eDRAM again in the RSX. NVidia seems to agree, who is no rookie in making GPU for embedded systems. They both saw the fat bus between the RSX and Cell and aren’t dumber than you. If they say “this bus can be used effectively and bring benefits to the table” I think I trust them over you and nick^2 saying “it can’t be used beneficially.”


So wait a second here!
nVidia knows best but Ati doesn't?
Sure that's why Ati has been on top since the Radeon 9700 series (R300).

Also that 35GB/s bus is highly theoretical and can never be sustained because SPE workload differences on the EiB.



eDRAM solutions aren’t new. If it was the golden answer to bandwidth problems, don’t you think you’d see it being used everywhere? Here’s a clue – Microsoft’s solution isn’t the best solution for alleviating bandwidth for all GPUs. Neither is Sony/nVidia’s solution to have a fat bus between CPU and GPU. The most efficient solution for the “ultimate gaming console” would have so many parts that the cost of the GPU would be sky high, its heat output would be sky high, and it would take up way too much space. You should be more than convinced eDRAM isn’t a cure-all when Sony used it with the PS2’s GS, the graphics chip they made themselves, and yet this time around, see no use for it. You think Sony is stupid enough to forget a consideration to do something they did before?


The 4MB eDRAM of the PS2 GS was used for all it's storing (textures, z-buffering etc.) while the 10MB eDRAM of the Xenos is only for frame buffer.
Textures, geometry etc will all be put into the GDDR3.
So that's a lame comparison as the eDRAM for the GS was VRAM while the eDRAM for the Xenos is just for frame buffer.



Your “Sony needs to do this” list is very impractical for the same reasons mentioned above.
-256bit buses are too big and costly to allow the price of the system to drop.
-The bandwidth “lost” to being on a 128 bit bus, and not having any eDRAM for frame buffer effects is addressed by putting a fast bus between the Cell and RSX. I say again: this bus can be used to alleviate bandwidth off of the main video memory bus for more than just vertex data and commands. Read up if you are still in denial: http://www.ps3forums.com/viewtopic.php?t=25846
-More video memory != higher resolution textures for free. When you bump up texture resolution,you are also increasing the amount of GPU execution resources needed to apply filters on those textures(AF), and execution resources needed to apply those textures to vertices(texel fill rates). Last time I checked, the Xenos only had 16 texture units and the RSX had 24, allowing it to support higher resolution textures without reaching this bottleneck. Considering textures, the RSX certainly has room to move into XDR RAM for texturing just like PC GPUs do. They can do this with less worry about bandwidth too since they have much more bandwidth than PCI-E and AGP buses.


For gods sake!
The RSX already has a 256bit memory interface it's just devided over the XDRAM and GDDR3.
I've said this in this very same thread for about 10 or more times.

You keep revering to this fast bus but it's not that fast at all.
It can only read 20GB/s and write 15GB/s to the XDRAM theoreticly.
Which in realworld performance comes to:

http://www.theinquirer.net/images/articles/PS3_memory_bandwidths.jpg

By doing a higher texture resolution you almost only need more VRAM capacity bandwidth needs can be quite sustainable.
Doing AF or any other filtering technique you're almost only eating from the GPU's it's VRAM capacity wise because bandwidth wise 8x AF only takes about 4GB/s of bandwidth.

The X1900XTX also has 16 texture units while the 7900GTX has 24 texture units both GPU's are clocked @650MHz.
Now you don't see the 7900GTX winning any battles do you the X1900XTX whoops it's *** in almost every game.
In heavy games like Oblivion and FEAR the differents in performance is actually pretty big.



This is where you are mistaken Frame buffer operations constitute some of the bandwidth hungry effects. Not all.


Okay enlighten me then!
HDR, MSAA, Alpha blending, Z-testing, Z-buffering, Z-culling.
What else is there?



Perhaps you should read my reply to nick^2 too then. It would be nice if either if you two could pull up some credible sources that show this to be true and is limited to the situation between the Cell and GDDR3. Also, you both might want to consider that what’s most important is the RSX being able to pull from XDR RAM. The only thing the Cell might push to the RSX or GDDR3 is vertex data. Textures will just be held in XDR, while the RSX makes requests for the data.


Yeps there you said it yourself.
They just use the XDRAM for textures but now how about HDR, MSAA, Z-testing etc. do they really want to fit that all into that limited 22,4GB/s?



So go ahead and delude yourself in thinking that one or two aspects(bus width and eDRAM) of a system can make a whole system will reap obvious improvements. In the computing world, that is not how things work. Consider the whole system at once. GuiltyBystander, you’re guilty of thinking eDRAM is a golden solution. Get a clue as to why this isn’t the best solution when there aren’t any top performing GPUs on the market today that use it.


I certainly don't think the eDRAM is gold as everything has their up and downsides.
However I know for a fact putting in that eDRAM next to the 512MB GDDR3 is a much better solution than using 256MB XDRAM of shared memory with a crippled amount of bandwidth with high latencies and 256MB GDDR3.

The reason why PC's don't use it is because of 2 reasons.
1. Being cost
2. Being PC games would need to be optimised a lot. Well have you ever seen an optimised PC game? So besides letting developers make a game that has to run on hundreds or thousands different system configurations they also need to optimise for each and everyone of them?



The 360's Xenos is equal to a an ATI x1900 and the price of those cards are just as much as an Xbox360. If anything the RSX is equal to the Geforce 7900 GT.


Well we ofcourse don't know for a fact the RSX is a G71 @550MHz but all information seem to point it is a 1:1 hardware duplicate.
Xenos however is a tricky thing though as it has more RAW power than the X1800XT while it has less than the X1900XTX and while in featureset, shader complexity, shader executions the Xenos is way beyond the X1900XTX.

[Solaris]

I'm not going to pretend i know more about the the guys in here. But for the guys claiming there is no beneficial advantages for the RSX's ability to read from the XDR ram. A developer working at Team Ninja who is working on Heavenly Sword, was actually asked a question that directly relates to this subject.

He was asked if the RSX will suffer in comparison to the Xenos when it comes to frame buffer effects, due to the lack of edram. He replied by saying if anything the RSX will have the advantage with frame buffer effects, then mentioned that this is because it is connected to 2 buses and not 1. Clearly he's indicating that it's something that's very beneficial. Considering he's developing one of the best looking games I've ever seen, I'll assume he knows what he's talking about.

[Knuckles126]

I'm not going to pretend i know more about the the guys in here. But for the guys claiming there is no beneficial advantages for the RSX's ability to read from the XDR ram. A developer working at Team Ninja who is working on Heavenly Sword, was actually asked a question that directly relates to this subject.

He was asked if the RSX will suffer in comparison to the Xenos when it comes to frame buffer effects, due to the lack of edram. He replied by saying if anything the RSX will have the advantage with frame buffer effects, then mentioned that this is because it is connected to 2 buses and not 1. Clearly he's indicating that it's something that's very beneficial. Considering he's developing one of the best looking games I've ever seen, I'll assume he knows what he's talking about.
And thats what we all will continue to do until Sony releases some more information pertaining to the subject.

[Bill_Gates]

Bah, the only reason they dont mention the RSX is there is nothing special about it, sucks monkey balls, and would rather bring more publicity to their failure which is better known as "CELL".

[Ero_senin]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines

[Knuckles126]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines
We still do not have the final specs on the RSX though.

[Ero_senin]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines
We still do not have the final specs on the RSX though.

but thats why it says in the official specs from this page.... :? :?

http://www.ps3land.com/ps3specs.php

[Crazy Phat]

It doesn't say final specs. Those are just the specs from E3 2005.

[HiX]

It's great to have people like Ebony and nick². They make the forum more interesting and civil. Yay for you guys.

[mynd]

Not quite in the same boat. The 360 uses edram to alleviate bus traffic.
When your not having to write everything back to the memory, the 128 bit bus becomes less of a problem. If anything, some embedded ram like the 360 would be a nice feature to add to the RSX.

To keep costs down and enable better die strinks down the line both console GPUs have been designed to use 128bit busses instead of the 256-512bit busses found in modern GPUs

To help compensate both GPUs are designed to use two busses for total working bandwidth.

On Xenos
1 bus to GDDR3 for textures, geometry, front buffer
1 bus to EDRAM for AA, stencil and Z tests, blending (this is the xenos backbuffer)

On PS3
1 bus to GDDR3 for textures, geometry, front/back buffer, AA, Z depth, blending
1 bus to XDRAM for textures, geometry, front/back buffer, AA, Z depth, blending

You can see a difference here. The bus to EDRAM on xenos is its back buffer. You cant hold textures or geometry here. This area is for a rendered frame only ready to be processed and copied back to GDDR3 to display. Meaning all textures, geometry, texture filtering, shaders etc has to come from GDDR3.

Whereas on PS3 both busses to RSX can provide it with textures, geometry, texture filtering, shaders etc

This would mean during rendering* PS3 would have upto twice as much bandwidth for shading and textures than 360.

upto 48GB from both busses to RSX
upto 22.4GB from the GDDR3 bus to xenon

Please remember this is during the shader fragment stage of a frame before applying AA etc. And that you cant hold textures in Xenos EDRAM

And before someone comes in saying
"yeah but XDR bandwidth is shared with Cell" or
"RSX has to go through Cell to get to XDR"

please refer to how Xenon has to go through the GPU to get to its Ram. And that its textures/shaders/geometry bandwidth is shaded with the CPU.

* the rendering of shader fragments, applying textures, shadows, texture filtering etc basically everything you see before the AA stage.

I'm going to ignore the numbers on the bandwisth issues, cause there is pleanty of people of arguing over that all ready.

Right lets get some facts sorted, geometry is very very little data in comparisim to textures, and texture mainpulation.

The EDRAM is totaly used for writing your backbuffer, I'm gong to ignoring the "free" image manipualtion that they can do with the back buffer because thats post render. However the final results of any render are sent to it. This means your not wriiting back to you main memory for your back buffer, and your back buffer is a constant drain on memory bandwidth, i.e. if your rendering, your writing back, regardless of what tectures of vertex buffers your grabbing. This is why overdraw has always been always a problem with games.
Hence, by removing this "write back" you've freed up the bus for use purley as a read only bus, youre not writing anything back.
Exceptions to this would of course be render to texture (I'm unsure if you could get the edram to do this, but most likely there is a "target" point on the setup of the render to texture command).

i.e.

RSX:

GDDR->RSX->GDDR

360

GDDR->RSX->EDRAM

So you simply dont need the bandwidth.

Furthermore, if you are contantly swapping betwen banks of memory, I see all mary hell breaking loose.
For starters, you probably have to implictly request witch memory you are getting the data buffers from. So you'd have to have some sort of tracking system as to where your data is kept. Ie. you cant just throw models and textures all over the place.
These chips need to know where its getting the data from.


You can see a difference here. The bus to EDRAM on xenos is its back buffer. You cant hold textures or geometry here. This area is for a rendered frame only ready to be processed and copied back to GDDR3 to display

Copied back to the GDDR3? Are you sure? I guess you only wirting once, but I would prefer to think the edram can be displayed imediatly (ok so it cuts you buffer in half but I would prefer it).
Still once is better than every draw call.

[The_One]

The Xbox 360 doesn't have any bandwidth problems because all bandwidth hungry effects (HDR, MSAA etc.) gets placed into the eDRAM while textures, sound, CPU processes get placed into the GDDR3 where 22,4GB/s is sufficient of bandwidth for. Stop making a puny 10MB of eDRAM sound like the Be All and End All memory pool! At resolution lower than 720p (or equal to it), yes, the eDRAM is going to be fine and dandy. Lets see the Xbox hit the theoretically 1080i max resolution. Your 10MB of eDRAM ain't going to be large enough to hold 2xFSAA plus the 1080i scene itself; I don't even need to factor in HDR at this point--It simply isn't enough. Now, don't take it the wrong way. The eDRAM is definitely a smart idea (I have commended ATi and Microsoft numerous times about this), I simply hate it when people put it on a elevated pedestal like the Be All End All memory.

Effects like MSAA and HDR also will be much better to do with less of a beating on performance. Yeah... It's a huge performance hit, that's why Heavenly Sword has FP16 equivalent HDR, 4X MSAA and 8X tri-linear anistropic filtering all being displayed at 720p, right :roll:? Seriously, it's not that big of a problem.
@Guilty Bystander: a 256bit bus is not costly or difficult? I might agree on the cost issue, but definitely not the difficulty issue. I'd like to see a PSThree 4 years down the line with a 256-bit GDDR3 Bus. Yeah, didn't think so, won't be happening in our life time--It takes too much darn physical area! This is when drawing a diagram would really help, but heck, I'm too lazy. On top of that, my wings can't hold a pencil properly.

To keep costs down and enable better die strinks down the line both console GPUs have been designed to use 128bit busses instead of the 256-512bit busses found in modern GPUs Thank you for pointing that out, at least I'm not the only sane person that's trying to point this major point out :?.

Maybe its hard to comprehend, but they BOTH lack memory bandwidth. Another sane person... Join the club, bro!

Everyone keeps saying how difficult it is to make a console with a 256-bit graphics bus... All it is, is copper on a PCB. Sure there is a little extra cost to it, but people overblow it... It's not saving much money, it really comes down to not having enough space. When you're producing a console as expensive as the PS3, every darn CM² on that PCB is going to count. I doubt there's enough space to route 2 more 64bit memory busses to the RSX; not saying that it couldn't be done if it was really thought out in the first place... But just look at the PS3, it already got a size increase from E3 2005, and that's WITH extra hardware removed! After this post, I think I'm going to stop arguing the 128bit bus position since apparently no one really listens (Everyone can just go believe what they want 8)). Oh well, c'est la vie.


nick² im surprised your posting here after Ebony pwned you hard.

I'm suprised there are so many children on these boards using words like "pwned".You'd be surprised at what people say here...

RSX already has a 256bit memory interface it's just that it's devided over the GDDR3 and XDRAM. Wait, am I missing something here? 3x64=256? Hmm, when did that happen? As far as I know, there's two 64bit bus to the GDDR3, which euals 128bit (I don't think I have to say that... do I :??) and ONE 64bit bus to the XDR which equates to... Gee willikers, don't need a calculator for this one: 192bits. Where did the extra 64bit pop up from?

Well, think about it. In one scenario you have 8 independent processing units, in the other you have 3. The more cores, the more bandwidth you need... period. You can't starve anyone, otherwise you will end up with major bottlenecks. Oi, that's flawed logic. Each SPE reports to the PPE via the EIB, and the SPE's have their own local memory which cuts down on XDR "fetching". The PPE is probably the one that's going to use the XDR memory bandwidth the most, and even then, not all that much. I'd really be surprised to see more than 50% of the XDR bandwidth being used at any given time for any 1st or 2nd gen PS3 game. So really, there's only one core using the XDR on a consistent basis, all the rest are... Doing homework, most of the time--hopefully. Uh... Yeah. I have no idea what I just said.

Furthermore, if you are contantly swapping betwen banks of memory, I see all mary hell breaking loose. Why would anyone want to do that, though? It's highly inefficient, put something where it's supposed to go, and leave it; until it's no longer necessary. I don't see why any devs would need to constantly swap data between the two memory banks (I'm assuming you are refering to the XDR and the GDDR3 in the PS3 with that statement?).

They just use the XDRAM for textures but now how about HDR, MSAA, Z-testing etc. do they really want to fit that all into that limited 22,4GB/s? Uh... DUH? Unless you got a better idea, that's where it's going to HAVE to fit. Besides, not all games are going to use the ultra fugly HDR, ya know :snigger?

Okay enlighten me then!
HDR, MSAA, Alpha blending, Z-testing, Z-buffering, Z-culling.
What else is there? What is Z-testing? Are you refering to the process of testing the depth (Z values) during either Z buffering or Z culling? I don't think Z testing is a standalone process/algorithm. It has to be used in conjunction with either Z buffering or Z culling... Otherwise, what are you going to use the returned Z data for? I'd like to emphasis "I think", so correct me if I'm wrong.

Note: Yes, I realize that my replies aren't in order, and they are crappily phrased at times. This is what I get for posting a reply at 3:00AM in the morning (I have no time at any other time. Been busy, anyone noticed the fact that I haven't be dishing out bans lately? :lol:). Okay, good night to you all, gotta sleep.

[GunTeng]

Im actually quite enjoying this debate, whether factual or theoretical.
I won't pretend I have an understanding of what processes are involved, figures or designs etc, but I am starting to see a picture from both sides of the debate.

Put it this way. While you guys are explaining in 3D conceptual terms, I am able to understand it in 2D terms. So please ( for my sake), don't do a Ken and start talking all 4D..;)

To be honest, both sides are relying on speculation ( and I think all involved would agree) & until you manage to actually get your hands on the hardware itself, you won't be able to state absolute facts. But from what I can understand, everyone makes sense, although I can't really say who I would believe as I don't have the knowledge to understand the design criteria.

Keep this debate going...Im learning all the time :)

[mynd]

Why would anyone want to do that, though? It's highly inefficient, put something where it's supposed to go, and leave it; until it's no longer necessary. I don't see why any devs would need to constantly swap data between the two memory banks (I'm assuming you are refering to the XDR and the GDDR3 in the PS3 with that statement?).


EXACTLY!!! Yet there are still people talking about it.
And despite the 3:00am grammer, you make perfect sense.

To clarify the whole z thing...
Z-Testing is the process of using your Z-Buffer to do Z-culling.
Hence Z-testing is the comparisim of you current poly results to the z-buffer results, if the entire triangle fails, it is culled.
The only thing being held by the edram is the z-buffer. Everthing else is worked on the fly folks. The 3 polys z values are compared to the eqivlant screen space z values, if all 3 are futher away (higher value) the triagle is not render by the pixel shader.

This all happens traditionally between the vertex shader and pixel shader.

[nick²]

Everyone keeps saying how difficult it is to make a console with a 256-bit graphics bus... All it is, is copper on a PCB. Sure there is a little extra cost to it, but people overblow it... It's not saving much money, it really comes down to not having enough space. When you're producing a console as expensive as the PS3, every darn CM² on that PCB is going to count. I doubt there's enough space to route 2 more 64bit memory busses to the RSX; not saying that it couldn't be done if it was really thought out in the first place... But just look at the PS3, it already got a size increase from E3 2005, and that's WITH extra hardware removed! After this post, I think I'm going to stop arguing the 128bit bus position since apparently no one really listens (Everyone can just go believe what they want 8)). Oh well, c'est la vie.


Well, maybe you are right, but I'm not sure how big of a space difference it does make.. I'm looking at a 6800 graphics card, with a 256 bit memory bus, and a 6600 GT with a 128 bit bus.. Aside from the heatsink size, they take up roughly the same amount of space..




Well, think about it. In one scenario you have 8 independent processing units, in the other you have 3. The more cores, the more bandwidth you need... period. You can't starve anyone, otherwise you will end up with major bottlenecks.

Oi, that's flawed logic. Each SPE reports to the PPE via the EIB, and the SPE's have their own local memory which cuts down on XDR "fetching". The PPE is probably the one that's going to use the XDR memory bandwidth the most, and even then, not all that much. I'd really be surprised to see more than 50% of the XDR bandwidth being used at any given time for any 1st or 2nd gen PS3 game. So really, there's only one core using the XDR on a consistent basis, all the rest are... Doing homework, most of the time--hopefully. Uh... Yeah. I have no idea what I just said.



Well, you have to remember there is no shared cache, and very small amounts of local memory on the SPE's. This means
A) they will be fetching data redundantly - different SPE's requesting the same information.
B) They will have to do it more often than a normal CPU because of the 256 kb local store.

As far as going through the PPE for memory, I don't know if thats correct or not, but as far as each SPE is concerned the memory controller activities should be transparent... It is literally like 8 cores requesting memory on their own.

The worst thing you can do to a processor is make it wait for data. Now granted, the processor will rarely use even close to 100% of its memory resources... However, due to the design, its not hard to see why the PS3 needs a lot of memory bandwidth.

What everyone also needs to remember is that memory bandwidth doesn't just come down to amount of data that can be transferred in a second. It also has to do with time... If you have 8 processing units, all requesting data at the same time, you can't have some SPE's waiting for the data. Having a high speed bus saves time. Instead of the data reaching the SPE in .5 ms, it might get there in .2ms. This really helps in times where an SPE might have to wait "on line" to get what it requested in memory.

If you look at multiprocessor boards on PC's, one of the reason for the very steep diminishing returns per processor comes from the memory access issues. It gets very complicated for each additional processor you use, and there can be a lot of bottlenecks when they all have to fight for the same reasources. Thats why 4 processors won't yield anything close to 4 times the processing power in a PC environement (even for very multi-threaded programs).

While its unlikely that most games will be using over say 60% of the available memory bandwidth on a consistent basis, you can be pretty sure that it will fully utilize the bandwidth if there are many simultaneous requests in a short period of time (which of course would not be constant, but instead happen occasionally)... This is exactly why the bandwidth is there. On these intermittent occasions, if the response time was too slow, everyone would have to wait for their data much longer... This could potentially slow down the system a great deal. You'd end up in a situation where there'd be a ton of requests, and an inability to deliver them from the memory to an SPE in a timely fashion..

[Guilty Bystander]

Stop making a puny 10MB of eDRAM sound like the Be All and End All memory pool! At resolution lower than 720p (or equal to it), yes, the eDRAM is going to be fine and dandy. Lets see the Xbox hit the theoretically 1080i max resolution. Your 10MB of eDRAM ain't going to be large enough to hold 2xFSAA plus the 1080i scene itself; I don't even need to factor in HDR at this point--It simply isn't enough. Now, don't take it the wrong way. The eDRAM is definitely a smart idea (I have commended ATi and Microsoft numerous times about this), I simply hate it when people put it on a elevated pedestal like the Be All End All memory.


Using tiling putting a 720p/1080i with HDR FP10 Blend, 4x MSAA etc. CAN FIT INTO THE 10MB eDRAM.



Yeah... It's a huge performance hit, that's why Heavenly Sword has FP16 equivalent HDR, 4X MSAA and 8X tri-linear anistropic filtering all being displayed at 720p, right ? Seriously, it's not that big of a problem.


nAo32 only uses FP16 which is not blended why else do you think MSAA can be enabled.
By the way non blended FP16 isn't real HDR.
It might look similar but it isn't real HDR just like the INT10/16 HDR done within the Source engine.



@Guilty Bystander: a 256bit bus is not costly or difficult? I might agree on the cost issue, but definitely not the difficulty issue. I'd like to see a PSThree 4 years down the line with a 256-bit GDDR3 Bus. Yeah, didn't think so, won't be happening in our life time--It takes too much darn physical area! This is when drawing a diagram would really help, but heck, I'm too lazy. On top of that, my wings can't hold a pencil properly.


A 256bit memory interface might have been difficult and costly in the time when it was used with the Matrox Parhelia, Ati Radeon 9700/9800 series, GeForce FX5900/5950 series and maybe even with the Ati Radeon X800/850 series and GeForce 6800 series but nowadays it's easy.
Sure a 256bit memory interface costs more than a 128bit memory interface but it's not a costly as you might think.
Using quad channel memory controllers (256bit=4x64bit) isn't new tech anymore and has been mass produced for some time now on highend GPU's.



Wait, am I missing something here? 3x64=256? Hmm, when did that happen? As far as I know, there's two 64bit bus to the GDDR3, which euals 128bit (I don't think I have to say that... do I ?) and ONE 64bit bus to the XDR which equates to... Gee willikers, don't need a calculator for this one: 192bits. Where did the extra 64bit pop up from?


Ever thought of the fact 1 channel might be switched off for either a higher yield rate or because it's pointless to get the RSX memory controllers to adress the XDRAM with 51,2GB/s (3200MHz x 128bit : 8bit) while the Flex I/O can only deliver a theoretical MAX of 15GB/s write and 20GB/s read.

[Scribe]

If that yield rate concept was true, it just wouldn't have the extra channel in the first place cos that would affect the yield.

Also, sony OFFICIALLY stated that the RSX will be a modified shader model 3.0, that is that they've made some nice(why else make them) changes to it that don't perfectly conform to dx9.0c specs. Meaning for better or worse, the chip is NOT a 1:1 copy of anything nVidia has on the market.

A note on the CELL, the SPUs don't go through the PPU and each have their own memory manager than functions independently and uses DMA, even independantly to its own SPU and is clocked at something stupidly high like 6ghz (rest of the chip is of course still 3.2ghz). It is capable of queing memory access commands without slowing down SPEs, and is capable of making sure memory access commands are schedualed with all the other processors to hide memory latency.

[EbonySeraphim]

nick²:
Bandwidths are added when they can be used simultaneously due to being on separate buses. If you’re trying to tell me that the RSX will never use ~20GB/s bandwidth from XDR at the same time as using 20GB/s from GDDR3, then you might as well talk about how any bus isn’t really giving you their max bandwidth unless they are being used at the same time. Stop trying to partially apply “real world” usage scenarios or limitation in an imbalanced manner. Almost any of the points you are making in the PS3, apply to all other hardware, and all other parts contained in the PS3 itself. So really, what is your purpose in saying all of that?


Let's go back to my example... we have 60 MB of data needed for this one frame (the same data will be used for multiple frames, slightly chaning as one progresses).

Let's say the cell is using 40% of its available bandwidth, leaving 15 GB/s. Due to the inefficiencies of having to use a second hop (GPU memory controller to cell's memory controller) - I'll be conservative here - you lose 2 GB/s. That gives you 13 GB/s for the RSX. I think this is quite reasonable, that the CPU would be using this much memory bandwidth and that there is a 10% loss in bandwidth due to extra latencies.

Now like I said, let's take the worst case scenario - all of the data is in system memory. Now what? All 60 MB/s is transferred at 13 GB/s. This is slower than keeping it all on the GDDR3.

Now lets take a more balanced scenario... 50% on the RSX, 50% on the RSX. [b]30 MB gets transferred at ~23 GB/s. The other 30 MB gets transferred at 13 GB/s. We have to transfer the same amount of data from both memory banks, but the data in XDR takes twice as long... The way the math works out, its roughly equivalent to transferring the whole thing from the GDDR3 at ~23 GB/s...

The best balance would be somewhere from a 70-30 to a 80-20, which could give you up to 30 GB/s....
Where the hell did 15GB/s come from? What do you mean by “available bandwidth”? Why pick 40% XDR memory usage? How the hell does the Cell usage of XDR RAM make the FlexIO <-> XDR Memory controller any more inefficient? Why are you still applying imbalanced and unfair comparison on components inside the same system? It’s beyond mind boggling. Where did the added drop to 13GB/s come from? You need to make your starting numbers clear, and every “nerf” you throw in after that fact, you had better explain it. When you do, apply the same applicable nerfs to whatever else you’re comparing to.

Why even pull the “worst case scenario” out of thin air? Why would any developer put all of their textures in video memory unless they already saw it reasonable?

Look at the bolded portion. Why did you use the maximum theoretical bandwidth of GDDR3 and put that next to the “realistic” performance of the RSX <-> Cell bandwidth? There is no error in my math. There isn’t even necessarily an error in yours. You’re just comparing numbers that make no sense.

@GuiltyBystander:


Xbox 360 certainly has the least bandwidth problems of any console in history.
Still in your dream world eh?


Well Valve are lazy programmers which is easily shown by the fact Half Life 2 has been on the market 2 years and still they didn't fix the audio stutterbug.
Using tiling with the eDRAM on the Xenos takes a lot of effort (same reason why none done it to date) so using just the GDDR3 would make it nice and simple for Gabe Fatneck and his team to lazy port Half Life 2 to the PS3 and Xbox 360.
It has nothing to do with the eDRAM not being good enough.
Before you go on and make yourself look even dumber, I think you should realize that it is Valve who came up with a solution to avoid using tiling to eDRAM, while still taking advantage of HDR and AA. This “lazy” developer you speak of is doing something others aren’t.

http://www.beyond3d.com/forum/showthread.php?t=32152


Sure that's why games like Brothers in Arms for example doesn't have AF on both PS3 and Xbox 360.
Doing AF in a small closed enviroment like that arena in the Heavenly Sword and in the Xbox 1 graphicly like title Untold Legends doesn't say anything about other games.
Sonic, BiA, Resistance, Motorstorm, Virtua Tennis, Fatal Inertia and about every other game on the PS3 suffer the same Trilinear or 2-4x AF like most Xbox 360 titles.
http://www.beyond3d.com/forum/showthread.php?t=31221

That seems to be kicking at the texture bandwidth issue for XBox360 doesn’t it? And by the way, next time, reply to everything I say because you haven’t even begun to address the basis of my claim to begin with:

“Speaking of bandwidth on the XBox360, how does sharing a bus with the CPU handle on the system? While the CPU may not be the higher bandwidth consumer in many games, it’s certainly going to be saturating the GDDR3 memory controller and bus with many small requests that degrades overall latency and bandwidth performance considerably. Maybe that’s why the XBox360 version of games notably are missing anistrophic filtering where the PS3 shows it…”

Even if the XBox360 can run AF to some level, it would be taking a hit somewhere else on texture bandwidth. Either way, the 22.4GB/s bus has it’s own restrictions that you are acting as if it doesn’t exist.


So wait a second here!
nVidia knows best but Ati doesn't?
Sure that's why Ati has been on top since the Radeon 9700 series (R300).

Also that 35GB/s bus is highly theoretical and can never be sustained because SPE workload differences on the EiB.
I pretty much said that Sony and nVidia know what’s up with their GPU, and their console. I didn’t say that ATI was stupid or wrong in putting eDRAM in the Xenos did I? I just said that Sony and nVidia don’t see that option fit for the RSX in the Playstation 3’s situation, and bringing along eDRAM into the RSX would make it suffer elsewhere in performance more than it would bring any benefit. Think before you write.

Also, read up on what I’m discussion with nick^2 – every bus gives a theoretical maximum number. Your justification or whatever you were talking about relating to SPE workloads (probably weakly trying to run off of what nick said) makes absolutely no sense.


The 4MB eDRAM of the PS2 GS was used for all it's storing (textures, z-buffering etc.) while the 10MB eDRAM of the Xenos is only for frame buffer.
Textures, geometry etc will all be put into the GDDR3.
So that's a lame comparison as the eDRAM for the GS was VRAM while the eDRAM for the Xenos is just for frame buffer.
You really believe that horse sh*t you just said? 4MBs for all textures? Damn that’s amazing! Why didn’t the XBox original show off 8x better textures with up to 64MBs of video memory?


For gods sake!
The RSX already has a 256bit memory interface it's just devided over the XDRAM and GDDR3.
I've said this in this very same thread for about 10 or more times.

You keep revering to this fast bus but it's not that fast at all.
It can only read 20GB/s and write 15GB/s to the XDRAM theoreticly.
Which in realworld performance comes to:
You are going down a horribly cloudy path. It was brought up because you were completely ignoring the other bus.

Your screenshot, while seemingly pinning to the PS3’s performance, would you care to show me what the real measured performance of that 22.4GB/s bus is? Oh yeah, that’s right! Microsoft most wants to hide this because they want to make their console seem more powerful and compare their theoretical maximums to the PS3’s real world performance. Sure sure, you can go and believe this is the difference. But just like nick, you are forgetting that if you’re going to compare two numbers, you better be comparing theoretical vs theoretical, or realistic measured vs realistic measured. Otherwise, there is no validity in the comparison.

By doing a higher texture resolution you almost only need more VRAM capacity bandwidth needs can be quite sustainable.
Doing AF or any other filtering technique you're almost only eating from the GPU's it's VRAM capacity wise because bandwidth wise 8x AF only takes about 4GB/s of bandwidth.
So then why don’t you add the bandwidth required to apply that texture after it’s been filtered? How much bandwidth does it take to use a feature like vertex texture fetch in vertex shading programs? That bandwidth adds up, and is the other major bandwidth consumer in GPUs that you have some naïve belief, does not exist. Just a heads up, I crunched the numbers a while back, and at 1080p@60fps@32bpp(really should be 24bpp) 4xMSAA only takes up ~2GB/s bandwidth.


The X1900XTX also has 16 texture units while the 7900GTX has 24 texture units both GPU's are clocked @650MHz.
Now you don't see the 7900GTX winning any battles do you the X1900XTX whoops it's *** in almost every game.
In heavy games like Oblivion and FEAR the differents in performance is actually pretty big.
http://www.gamespot.com/features/6145814/index.html

That’s far from whopping it in every game. In fact, you blatantly lied as the 7900 whoops ATI most significant in FEAR. And before you go down a useless path - I don’t give a rat’s ass about which of those two cards is better.


Okay enlighten me then!
HDR, MSAA, Alpha blending, Z-testing, Z-buffering, Z-culling.
What else is there?
I already did. Anything related to textures – filtering them, applying them, or using them for vertex processing such as is done with vertex texture fetch. Get learnt. Btw, good job on restating the same general algorithm three times.


The reason why PC's don't use it is because of 2 reasons.
1. Being cost
2. Being PC games would need to be optimised a lot. Well have you ever seen an optimised PC game? So besides letting developers make a game that has to run on hundreds or thousands different system configurations they also need to optimise for each and everyone of them?
I think #2 is the most reasonable thing you’ve said….ever. I’m almost ashamed to agree. #1 I still disagree with because if it is a clear benefit, ATI and nVidia would offer it on their top end cards which are very expensive anyways. As they implement it in their higher end cards, when they get older, producing them get’s cheaper. Soon everyone would have eDRAM on their GPUs and developers would be perfectly willing to optimize for it.


Well we ofcourse don't know for a fact the RSX is a G71 @550MHz but all information seem to point it is a 1:1 hardware duplicate.
Xenos however is a tricky thing though as it has more RAW power than the X1800XT while it has less than the X1900XTX and while in featureset, shader complexity, shader executions the Xenos is way beyond the X1900XTX.
That’s a very nice software centric mind you have there. Here’s a proposition for you –

“OpenGL, while being less used, is FAR superior than Direct3D9 and even DirectX 10. The RSX supports stuff BEYOND GLSL and OpenGLwhich makes it even more bad ass.”

If your bullsh*t detector didn’t just break due to picking high reading from what I just said, then maybe that’s why it didn’t break due to what you just said.

Figure out the difference between hardware, feature sets, and language implementations. You clearly lack the understanding of the differences between these things because of what you just said.

@nick:

Well, you have to remember there is no shared cache, and very small amounts of local memory on the SPE's. This means
A) they will be fetching data redundantly - different SPE's requesting the same information.
B) They will have to do it more often than a normal CPU because of the 256 kb local store.
Have you read the PS3 Explained thread? Have you understood what the optimal programming paradigms for the Cell processor are? If you did, then you would know that you are explicitly targeting software approaches that are stupid to implement on the Cell and can easily be avoided/mitigated by features on the Cell. Before you continue to assume anything else about the processor, I seriously suggest you check out this site and read all of the training presentations:

http://www.power.org/resources/devcorner/cellcorner

Do not insult other members by calling them a "dummy". - The_One

[shadow26]

well, im glad someone knows what they are talking about....
I also like watching other people get owned by ebonyseraphim....
but dont ask me for any input, i have no idea whats going on...

[WASHIMUL]

yea yea u are right EB

dont forget each SPE has a local storage of 256K and IBM clearly siad that it would be "more than enough " in any real world applications.

another thing that we should look into is that the CELL processor can handle some /most graphical tasks better than even the HIGH-END graphics card.

we could take the STENCIL performance (found by UNIVERSITY of CLIFORNIA @berkeley) for instance where the CELL zaps the rivals by a magnitude of 82:1

in TRE CELL beats a duel core G5 2.7ghz by a margin of 35:1.

it is a request to most members to make their posts out of real world performance numbers rather than mere assumption.

the recent interview on FANTASY LAB definitely answers all questions about the CELL's capabilities to handle graphical tasks

[janenba352]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines

Ill go all the way back to this....

Even on playstation.com there is no specs under the RSX or Cell meaning that the specs for both could change without Sony telling us about it. They want to keep a low profile and hit us with a huge surprise at TGS and probably a week or so before launch. That way they will drag in all the customers they want. If they do increase the specs for just the RSX that would be some godlike thing of itself but increase the Cell to 3.4GHz core and SPE that would be sicker. But who knows Sony has the NDA up so we have no clue even whats in the Final Dev Kits other then what they figured they would have at E3'05 which was over a year ago which things could have easily changed.

[HighPs3]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines

Ill go all the way back to this....

Even on playstation.com there is no specs under the RSX or Cell meaning that the specs for both could change without Sony telling us about it. They want to keep a low profile and hit us with a huge surprise at TGS and probably a week or so before launch. That way they will drag in all the customers they want. If they do increase the specs for just the RSX that would be some godlike thing of itself but increase the Cell to 3.4GHz core and SPE that would be sicker. But who knows Sony has the NDA up so we have no clue even whats in the Final Dev Kits other then what they figured they would have at E3'05 which was over a year ago which things could have easily changed.


This is by far the stupidiest marketing I have ever heard of.. Have something so great but dont tell anyone about it until 2 months before launch or a week?

Do you know how many fans they have lost from E3 til now alone? For the sake of flaming mabe not many but you cant say not one has been swayed by basically poor marketing.

TGS is the last shot they got to make an impression thats atleast how I feel. If they dont want to release spec info, online info, game release dates, launch titles (all of them) ill just get a 360, or upgrade my PC and wait until you all review it and I can read about what it does.

After all Assassins Creed, Killzone, MGS4, Final Fantasy, etc arent coming out until next year anyhow...

Just how im feeling lately. Im hoping they release all that at more at TGS but I just dont think its going to happen. There going to show game trailers and thats it. Mabe have warhawk and resistance playable..again...
I hope im wrong though!

[archy141]

Great stuff from nick² & EbonySeraphim ! Cant wait to see each of your replies.

EbonySeraphim, I love your style of debate -KEEPIN IT R E A L!.
Well balanced & comprehensive replies backed with solid reasoning that pack a punch. Unfortunately my tech knowledge on this subject is just nowhere with you guys & I don’t try to BS otherwise.

Anyhow here are my 2 cents on the matter – for the non technical

Neither Sony’s nor Microsoft bus designs are perfect.

Each was developed to best suit there individual consoles architecture. And of course each could have been better. It’s not the first or the last time developers will have to work with constraints. Current technology is not just limited by what a manufacturer can create but more by what the consumer is prepared to pay -today. Their prime objectives are not fulfilling the whims of Fanboys at the risk of bankrupting themselves. At the end it’s all about the Money.

Sony have previously used the eDRAM with the PS2 & there is no reason they couldn't have done it again if they felt it benefited there PS3 architecture. The XBOX360 is selling the eDRAM solution as something unique when in fact it is nothing revolutionary –Even the Gamecube used it before. Let’s not forget Sony has been in the console making market for a while & we should give them credit for knowing what they are doing. Each manufacturer does what suits its architecture best.

When people make remarks like 256bit bus could have been easily implemented by Sony i.e. Sony were too stupid to come up with this unique idea & they are the bright geniuses - I think they should just stick to counting paper statistics. Being able to understand a product technically is one thing & being able to make business decision based on economic feasibility & technical viability is another. Some people are just being plain ignorant & trying to justify business logic with personal whims.

Let the games speak :wink:



Now for more fuel on the RSX & G80 Link (getting rather scarce)


The other thing we heard is that the G80 is going to be mostly G70/G71 with DirectX 10.0 stapled on. Meaning, that since the architecture is basically G70, it has not been built from the ground up, keeping DX10 in mind.
http://www.cooltechzone.com/Departments/Featured_Story/NVIDIA_G80_Delayed_200604092276/"


I came across this article suggesting the G80 is closer to G70/71 than we possibly think. If that is the case the whole idea of RSX being G80 like would seem even more possible.

G80 is looking to be the first true dual-core GPU, as opposed to multiple cores per PCB and SLI. This will obviously make it bigger unless the RSX is just a smaller subset of the G80. Recall PS3 did get fatter even though we lost from the original feature list.

Majority of the community is set in concrete about the RSX being NV47 core based (G70/71). The idea that it could be an altogether new G80 core cannot be accepted by many. If this article is true than the G80 has a lot more in common with the NV47 core than people realise. I have posed the question on Beyond3D & no one has said outright that it’s a new breed of core altogether.

If the G80 is a NV47 core 'stapled' with extra features (SS4, GDDR4 etc) than there is even more possibility that the RSX could be like G80.


To Recap my RSX <--> G80 Links

1. Most powerful GPU on PS3 release
2. Built in HDCP
3. Built In Audio support
4. G80 90nm production at IBM FishKill - same place as Cell
5. G80 Technology was there when Sony approached NVIDIA (2003 patents )
6. 4D Texture tables
7. Both based on NV47 Core (Speculation from quoted article)



According to Harrison, "It's all going well. We are in full swing, from a developer point of view, and over 10,000 developer kits have been shipped. Developers now have final PS3 hardware in their hands, though there will be some upgrades to the operating system - there's nothing unusual about that. The new controller is now in developers' hands, so all the pieces of the puzzle are there."

http://www.computerandvideogames.com/article.php?id=143840


Okay who knows a PS3 developer close by that we can jump ?

I wonder if someone will leak out Final spec details before Sony do..

Come on you can do it fella ! ! !


Archy

Do NOT double post! Use the EDIT button instead! - The_One

[razorblade416]

Please no double posting, use the edit button.

When you say the PS3 got fatter does that mean from E305? Because it also didn't have holes for cooling it then either, so that's not a good reason. As for the final hardware, it can still change at any time if they want it to. But if the devs do a special treat in their kits, NoUseMercinary could see if they would let somehting slip in one of the PS3Land interviews :wink:

[Itaintrite]

Yes, ask a trick question.

"Do you think the G80-based RSX is more powerful than the Xenos?"

:P

[The_One]

@Guilty Bystander: Do you even read what the heck I wrote? When did I say that a 1080i scene CANNOT fit into a 10MB of eDRAM while using tiling? I was even stating that tiling HAS to be used at that resolution (and therefore it will fit), and it's not something that every dev wants to use (it effectively disables the usage of the oh-so-useful MEMEXPORT).

And on the 256bit bus issue. I will say this again: PHYSICAL SIZE!!! It has nothing to do with technical limitations! Think of it this way: You have a ship that can carry 10 tons of goods, but you have a package of 15 tons. It doesn't matter how technologically advanced your ship is, it simply ain't going to carry the goods without capsizing. If Sony wants to be able to shrink the PS3 down the a PSThree (which I'm sure will happen), it's going to have to utilize a 128bit bus. Other components (CPU/GPU) can be shrunk by lowering the wafer size (smaller nm fabrication process), but one thing that will never shrink are the physical bus sizes on the PCB (and pin sizes, I think), and this is the point I'm trying to point out...


Ever thought of the fact 1 channel might be switched off for either a higher yield rate or because it's pointless to get the RSX memory controllers to adress the XDRAM with 51,2GB/s (3200MHz x 128bit : 8bit) while the Flex I/O can only deliver a theoretical MAX of 15GB/s write and 20GB/s read. If they are switched off, then it doesn't exist... Unless you're going to counter my point by saying that the CELL in the PS3 has 8 SPEs... Same deal here. If it's disabled, it's nonexistent.

dont forget each SPE has a local storage of 256K and IBM clearly siad that it would be "more than enough " in any real world applications. Since it's not cache, the developers have to explicitly control how the memory is going to be used. It may be a good thing to have, but it's definitely not an easy toy to play with.

I also like watching other people get owned by ebonyseraphim....
but dont ask me for any input, i have no idea whats going on...
Way to add to the community involvement there... :roll:.

@Archy: Seriously dude, DON'T DOUBLE POST! Don't make me remind you again; it's getting rather tedious.

"Do you think the G80-based RSX is more powerful than the Xenos?" No one has even confirmed the RSX to be G80 based, so the answer to your "trick" question is "No".

[Itaintrite]

The_One, the whole purpose of a "trick question" is to trick the developer into revealing more than he has to.

To make it clearer since you obviously didn't get it, if the developer answers either yes/no, it will confirm that the RSX is actually G80-based. Get it? :roll:

Of course, I was only (half-)joking..

[Lefein]

Ask what advantages there are in running an OpenGL based graphics engine on an NVidia designed GPU and what kind of effects they look forward to utilizing on the RSX architecture.

[Guilty Bystander]

Still in your dream world eh?


Least doesn't mean non now does it!



Before you go on and make yourself look even dumber, I think you should realize that it is Valve who came up with a solution to avoid using tiling to eDRAM, while still taking advantage of HDR and AA. This “lazy” developer you speak of is doing something others aren’t.


Well you're wrong there.
No Xbox 360 game to date uses tiling but still use MSAA and HDR so what was Valve bringing again?
I'd like your explaination as to why Valve still didn't fix the audio stutterbug because all I can up with is that they're lazy as hell even if it's an engine problem they could easily alter the engine a bit to fix and just upload it on steam.


I pretty much said that Sony and nVidia know what’s up with their GPU, and their console. I didn’t say that ATI was stupid or wrong in putting eDRAM in the Xenos did I? I just said that Sony and nVidia don’t see that option fit for the RSX in the Playstation 3’s situation, and bringing along eDRAM into the RSX would make it suffer elsewhere in performance more than it would bring any benefit. Think before you write.

Well does this sound familiar?


The RSX may only have one explicit “video memory bus” that is 128 bits wide, offering 22.4GB/s to GDDR3; but the bus to the Cell processor(FlexIO) which is connected to XDR RAM is entirely separate. Can this bus be used to alleviate bandwidth? It sure as hell can. Somehow, I don’t see vertex data and graphics commands taking up 35GB/s bandwidth between the Cell and RSX. Last time I checked, Sony wasn’t stupid enough to put such a fat bus between two major components only to have it not be utilized. Last time I checked, PC GPUs could use system memory as graphics memory freely(at least for textures, or not frame buffer also) – and the only thing making it non-viable for heavy use for most PCs, was the bandwidth on that bus. That bus is typically PCI-E 16x or AGP 8x running at 4GB/s – much slower than FlexIO. Perhaps this why Sony, who put eDRAM on the PS2’s Graphics Synthesizer, didn’t see it fit to use eDRAM again in the RSX. NVidia seems to agree, who is no rookie in making GPU for embedded systems. They both saw the fat bus between the RSX and Cell and aren’t dumber than you. If they say “this bus can be used effectively and bring benefits to the table” I think I trust them over you and nick^2 saying “it can’t be used beneficially.”

So if you agrea nVidia and Sony thought the eDRAM would be about useless then you think nVidia knows better than Ati don't you?
Ati thought using eDRAM would be better with the Xenos than using a 256bit memory interface and no eDRAM.
Oh and by the way Ati has much more experience on every part of the market than nVidia.
Seeing their low-end, mid-end and high-end all being faster eventhough their cards theoretical have less RAW power since the Radeon 9700 series than nVidia's cards I think Ati would problably be better in their judgements than nVidia.


Also, read up on what I’m discussion with nick^2 – every bus gives a theoretical maximum number. Your justification or whatever you were talking about relating to SPE workloads (probably weakly trying to run off of what nick said) makes absolutely no sense.

Well the Flex I/O is connected to the EiB.
If all 7 SPE's require 25,6GB/s of workload of the 204,8GB/s the EiB can deliver then the Flex I/O would be bandwidth limited.
Now I don't think this is going happen ever since developers just could optimise SPE's their workload but it could be possible.




You are going down a horribly cloudy path. It was brought up because you were completely ignoring the other bus.

Your screenshot, while seemingly pinning to the PS3’s performance, would you care to show me what the real measured performance of that 22.4GB/s bus is? Oh yeah, that’s right! Microsoft most wants to hide this because they want to make their console seem more powerful and compare their theoretical maximums to the PS3’s real world performance. Sure sure, you can go and believe this is the difference. But just like nick, you are forgetting that if you’re going to compare two numbers, you better be comparing theoretical vs theoretical, or realistic measured vs realistic measured. Otherwise, there is no validity in the comparison.


The GDDR3 in both PS3 and Xbox 360 can be adressed with 22,4GB/s by both RSX and Xenos respectivicly.
With the XDRAM this doesn't work that way because XDRAM bandwidth has to go from Flex I/O to the RSX memory controller resulting in loss of bandwidth.
You can see RSX can adress the GDDR3 with full bandwidth in this picture.

http://www.theinquirer.net/images/articles/PS3_memory_bandwidths.jpg



You really believe that horse sh*t you just said? 4MBs for all textures? Damn that’s amazing! Why didn’t the XBox original show off 8x better textures with up to 64MBs of video memory?


The 64MB DDR RAM was shared for the whole system.
Game code, sound, the Windows 2000 OS all ate a lot of the system memory and it's bandwidth.
Leaving maybe 40MB MAX for VRAM (but problably more like 20-40MB) with 3-4GB/s of bandwidth.
Every game on the Xbox have a lot higher and more textures than the PS2 could ever dream off.
Just check Ninja Gaiden Black as an example compared to GoW2.
The only way the PS2 could keep up with the Xbox was because games had about or the same polygoncounts.
On textures however the PS2 could never compete with the Xbox from day 1. (fixed)



That’s far from whopping it in every game. In fact, you blatantly lied as the 7900 whoops ATI most significant in FEAR. And before you go down a useless path - I don’t give a rat’s *expletive deleted* about which of those two cards is better.


Pfff that article clearly states SLI vs Crossfire.
How about making a single GPU comparison.

http://images.tomshardware.com/2006/06/05/geforce_7950_gx2/image011.gif

http://images.tomshardware.com/2006/06/05/geforce_7950_gx2/image012.gif

http://images.tomshardware.com/2006/06/05/geforce_7950_gx2/image013.gif



Well whatever by the way why would I even bother anymore?
Oh and by the way I would appreciate it when you crank your acting like a mofo attitude down a lot while I and others are discussing in a normally manner.
Keep it civil ffs otherwise go ruin some other forum with your attitude!


If the G80 is a NV47 core 'stapled' with extra features (SS4, GDDR4 etc) than there is even more possibility that the RSX could be like G80.


To Recap my RSX <--> G80 Links

1. Most powerful GPU on PS3 release
2. Built in HDCP
3. Built In Audio support
4. G80 90nm production at IBM FishKill - same place as Cell
5. G80 Technology was there when Sony approached NVIDIA (2003 patents )
6. 4D Texture tables
7. Both based on NV47 Core (Speculation from quoted article

The G80 core will have a lot more transistors than 300 million more like 400-500 million.

1. The G70 was the best GPU in the world @E3 2005 also I've never heard nVidia or Sony officials say the RSX would be the best nVidia GPU on the market with the PS3 release. If nVidia and Sony said so they were talking about the May 2006 release because half a year down the road technology has progressed a lot with the new GPU (G80).
2. I've never ever heard RSX will have built in HDCP support.
3. The G80 will only pass through audio from another audio source because of it's HDMI connection. The G80 won't do audio atleast I've never heard about it.
4. I thought I heard G80 would be 80nm.
5. Patents doesn't mean the technology is available. Some Cell patents originate from 1998/1999 or so does that mean Cell was available in 1998/1999?
6. I never ever heard of the RSX doing 4D textures. All I've seen was with the E3 2005 Kutaragi saying PS3 games would be 4D meaning SPACE+TIME.
7. Based on doesn't mean both cores are the same because the X1800/1900 and X800/850 series are still R300 based cores. Would you say the X1900XTX is the same as the 9700Pro?

By the way the G80 only supports 4D texture lookups.

[mr.brightside]

The only way the PS2 could keep up with the Xbox was because games had about or the same polygoncounts on textures however the PS2 could never compete with the Xbox from day 1


You can't get polygons in textures. Polygons are used in character models and most other models in games, to build the maps etc.

Im assuming you mean texture resolution, or pixels.

[Guilty Bystander]

You can't get polygons in textures. Polygons are used in character models and most other models in games, to build the maps etc.

Im assuming you mean texture resolution, or pixels.


Whoops I ofcourse mean:



The only way the PS2 could keep up with the Xbox was because games had about or the same polygoncounts (a comma here to devide the sentence ofcourse) on textures however the PS2 could never compete with the Xbox from day 1


Just read it as two different sentences otherwise like this:

The only way the PS2 could keep up with the Xbox was because games had about or the same polygoncount.
On textures however the PS2 could never compete with the Xbox from day 1.

[mr.brightside]

Ahhhh, i retract my statement :!:

[Guilty Bystander]

Ahhhh, i retract my statement


No problem man I could understand why you got that idea because I got the same reading it myself that's why I fixed it.

[The_One]

Oh and by the way I would appreciate it when you crank your acting like a mofo attitude down a lot while I and others are discussing in a normally manner.
Keep it civil ffs otherwise go ruin some other forum with your attitude! I just gave him a warning on his previous post for calling other members a dummy, lets hope he listens :roll:.

The_One, the whole purpose of a "trick question" is to trick the developer into revealing more than he has to.

To make it clearer since you obviously didn't get it, if the developer answers either yes/no, it will confirm that the RSX is actually G80-based. Get it? :roll:

Of course, I was only (half-)joking.. Yeah... I get it now XD. I feel so stupid :lol:.

[archy141]

Guilty Bystander - there was a reason why i commended EbonySeraphim & Nick². They seem to talk more sense, more factual & don't sound aggressive in trying to get there points across. You should try reading all of the thread before making replies to it - else you will come across as being ignorant.


1. The G70 was the best GPU in the world @E3 2005 also I've never heard NVIDIA or Sony officials say the RSX would be the best NVIDIA GPU on the market with the PS3 release. If NVIDIA and Sony said so they were talking about the May 2006 release because half a year down the road technology has progressed a lot with the new GPU (G80).
2. I've never ever heard RSX will have built in HDCP support.
3. The G80 will only pass through audio from another audio source because of it's HDMI connection. The G80 won't do audio at least I've never heard about it.
4. I thought I heard G80 would be 80nm.
5. Patents doesn't mean the technology is available. Some Cell patents originate from 1998/1999 or so does that mean Cell was available in 1998/1999?
6. I never ever heard of the RSX doing 4D textures. All I've seen was with the E3 2005 Kutaragi saying PS3 games would be 4D meaning SPACE+TIME.
7. Based on doesn't mean both cores are the same because the X1800/1900 and X800/850 series are still R300 based cores. Would you say the X1900XTX is the same as the 9700Pro?

By the way the G80 only supports 4D texture lookups.



1. Why would they be referring to the current most powerful GPU back in e3 2005 when the console was said to be coming out till spring 2006. I speculated it was more likely fall 2006 all along & hence they would have had enough time to produce a GPU with features like G80. READ THE THREAD

2. READ THE THREAD. Links were provided

3. What you are saying is speculation which I have read also on another forum
READ THE THREAD. Links were to NVIDIA 2003 patents describing a GPU with Audio

4. READ THE THREAD. Links were provided

5. It also doesn’t mean the technology has not been implemented. That’s why this is being discussed in a thread that is based partially on SPECULATION.

6. As well as Ken’s 4D, I gave two other links that pointed (i) A patent (thought to relate to G80) showing it UNIQUELY does 4D Textures in Hardware (ii). Naughty Dog running a Siggrapgh 2006 Course on a gaming technique that uses 4D Texture mapping techniques. READ THE THREAD. Links were provided

7. This was discussed a while back in beyond3D without a firm conclusion. I bought up the question again & same result. This combined with the article allowed me SPECULATE on the core being considered as NV47.

You really give the impression that you have not read anything to date in this thread even though you are one of the more frequent posters. I expect this of a new forum member but not you.

Archy

[mynd]

You know what, the proof is in the pudding and at the moment, NONE of the games look like they are pushing the G70 yet alone a g80 chipset.
Show me some "proof" in video form to confirm your theory.
Given the shear amount of time PS3 has had to get its house in order, it realistically should have a launch line up somewhere along the lines of the original Xbox, i.e. a halo like game, 2 months after launch the Xbox got what is still one of the most graphically impressive games ever on it in Rallisport Challenge. I see nothing , nothing that even looks impressive at launch on the PS3. And this is the problem you see.
You don’t get a 2nd chance, you’ve got to lead with your best out of the gate when you are playing catch up.
If it was not for Halo, the Xbox would have failed, plain and simple, it was up against a established console, with games coming out that were solid.

People will say things like "oh but you can't compare second gen xbox360 games to PS3 launch title, absolute bollocks, you HAVE to , people did it with the Xbox at launch, and they will do the same with the PS3. People will be comparing the 2 from day one. It's natural, because they are the competition, and the PS3 will need to justify the extra $$$.

Where am I going with this? How does it tie in with your "G80" theory?

Well that’s quite simple, you've see for SONY to have secretly shipped and upgraded the graphics chip and then given only developers months to improve the games graphics to take advantage even slightly of the new features will prove 2 things.

1) Sony are so completely arrogant about the selling power of the PS3 that they really don’t care what shape the games are in at launch (now given the video proof we have seen lately who can honestly argue with this)

2) Sony are no longer focusing the PS3 as a games machine (so why the hell upgrade the dam thing in the first place one might ask).

You know Assassins Creed probably was originally exclusive for the PS3, but Ubisoft either didn’t get an exclusive deal with Sony that they wanted, or they have decided they will not sell enough copies to recoup costs on this platform alone, given UBISOFT have had their best ever sales for ANY platform on the 360 with GRAW, they know this machine has viable sales. As for Atari, they have stayed completely clear of the PS3 for now. That’s bad news.

I would really like the idea of the G80 being included, and that it has infact been the target spec all along, but if it had, why the secrecy.

You know it's not like they have to worry about the 360 one upping them, the 360 gpu isn’t going to be magically upgraded to a newer spec is it now? They could have announced it the day after the 360 went on sale, and it would have made no difference would it?

They only reason for lack of info, if it has been upgraded, is because it a last minute thing, and that smacks of panicking.

Show me games SONY, Roll on TGS.

[GunTeng]

I dont know why being so noisy, everyone knows that the official RSX specs are this:

GPU
RSX @550MHz
1.8 TFLOPS floating point performance
Full HD (up to 1080p) x 2 channels
Multi-way programmable parallel floating point shader pipelines

Ill go all the way back to this....

Even on playstation.com there is no specs under the RSX or Cell meaning that the specs for both could change without Sony telling us about it. They want to keep a low profile and hit us with a huge surprise at TGS and probably a week or so before launch. That way they will drag in all the customers they want. If they do increase the specs for just the RSX that would be some godlike thing of itself but increase the Cell to 3.4GHz core and SPE that would be sicker. But who knows Sony has the NDA up so we have no clue even whats in the Final Dev Kits other then what they figured they would have at E3'05 which was over a year ago which things could have easily changed.


This is by far the stupidiest marketing I have ever heard of.. Have something so great but dont tell anyone about it until 2 months before launch or a week?

Do you know how many fans they have lost from E3 til now alone? For the sake of flaming mabe not many but you cant say not one has been swayed by basically poor marketing.

TGS is the last shot they got to make an impression thats atleast how I feel. If they dont want to release spec info, online info, game release dates, launch titles (all of them) ill just get a 360, or upgrade my PC and wait until you all review it and I can read about what it does.

After all Assassins Creed, Killzone, MGS4, Final Fantasy, etc arent coming out until next year anyhow...

Just how im feeling lately. Im hoping they release all that at more at TGS but I just dont think its going to happen. There going to show game trailers and thats it. Mabe have warhawk and resistance playable..again...
I hope im wrong though!

If you think about it, its not really a bad marketing ploy.
If all info was released at E306, then the anticipation factor would evaporate into nothing by the launch.
This way, they give you a taste have you anticipate TGS where you believe you will get all the juice on the console. After that you have only 2 months to wait for release & by that time all stores will have their final launch numbers for the SKU's.

So, after TGS ( if they build on the anticipation & excitement) people will be able to flock to their local stores & pre-order. It minimises confusion & builds anticipation to its highest levels. They don't want a XboX360 style mess at launch & this is one way of avoiding that.

Its like climbing a ladder ( sorry if this analogy is inappropriate or makes you feel nausious). If you go straight to the top, you can only go down from there. If you fall then then the landing will be much harder.
But if you go part of the way up & you fall, then its not such a harsh landing. If you don't fall, you have all that room left to climb to the top.

This is just my way of explaining my thoughts on SONY's marketing ploy. E306 was not the crucial point int its campaign. E3 was but the 1st step & TGS is the top of the ladder. :)

[archy141]

They only reason for lack of info, if it has been upgraded, is because it a last minute thing, and that smacks of panicking.

Show me games SONY, Roll on TGS.


I guess we will know soon enough.

I agree with you in the sense that Sony are coming from the rear & they will be in serious trouble & loose a big chunk of the console market unless they have:

1. Super games RIGHT at launch & onwards OR
2. Super gaming hardware such as improved RSX

Both Microsoft & Nintendo have things that make them uniquely attractive. Nintendo & its unique games because of its controller. Microsoft with its spend power to buy out best exclusives, -starting with GeOW on PS3 launch. Games to date have been poor for my liking on the XBXO360 but I feel things will change as they have a good selection of future releases. Anything they don't have Microsoft seems to buy. Ultimately they have more money than Sony & they are using a clever & effective ploy to win the console war. Sony on the other hand are relying more so on there brand name & the fact they will be offering a True ‘NextGen’ console. If the PS3 can show to be a lot more powerful than the XBOX360 than it can win against the XBXO360. If its marginal than I’m don’t think the average ‘mass’ consumer will be shopping for it. Having nextgen components such as BD, Linux, HD or even having a super CPU will NOT sell the console to the masses unless the games show the difference. The average consumer will be buying the console with the best looking & playing games available. They will not care to consider that PS3 games are first generation & XBOX360 are second generation.

It’s for the possibility of reason 2. that this thread exists.

If consumers knew they were buying something that was clearly superior for gaming than they would consider buying it even if the games were not there yet.
Right now I believe it’s going to be the attraction of PS3’s hardware that will determine how well it does. On the games & price front it already has enough competition. It needs to clearly differentiate itself on graphics capability against competitors to justify its price point & late arrival. It needs to win as a console foremost rather than a BD player or a Linux computer.

For this reason I have felt that Sony would have to make the RSX (maybe Cell ) more than it was spec’d out to be if the PS3 was to maintain Sony’s lead. Early comments we have heard from developers have indicated that the difference in the consoles is marginal. I attributed that to them possibly not having the final hardware in there developer kit.

All these reason have lead myself & others to speculate that the RSX has to be more than we are presently lead to believe. Otherwise Sony have really messed up by spending too much money developing an exotic console system that is more expensive than the competition, arrived later on the scene, does not hold the power of exclusive games & seems to be marginally better graphically (RSX GPU outdated ). Sure it has BD, HDMI & HD but these things will not be attractive at its price to the mass consumer. Games will be foremost important for them - & you need the best hardware for the best games visuals. BD, HDMI & HD will instead be seen as a big bonus.

A change in the RSX’s specs is one area that we can speculate would bring performance gains to PS3 games. It would also enable Sony to live up to there words that the PS3 is Truly NextGen . If the RSX was like a G80 than it would certainly help achieve this. A change such as that would UNDOUBDTLY make the PS3 a lot more attractive to consumers as it would help to clearly differentiate its graphical powers from XBXO360. If consumers knew they had a clear technology edge over the competition it would give them more security in there investment & restore belief in the Sony name brand. G80 is just one possibility. We could instead see other smaller changes to the original RSX specs such as memory upgrades, increased frame buffer cache, clock speed changes etc. that could provide the increase in performance.
Also there are other theories that have been floating about for longer which imply what little specs we have been given are correct but lot of the truth is still hidden. Such as the design resembling the block diagram structure with “Synergistic Media Elements” (see previous post).

The only positive reason I can think for Sony not giving out full RSX specs is that it is of high marketing value & they want to time this as part of the PS3 release marketing propaganda which has not kicked off yet.

I for one really hope RSX or some other change in Sony’s PS3 gives it a clearer gaming edge over its main competitor -XBOX360. It is the only area that I feel the PS3 is not showing Truly NextGen. Otherwise I feel the PS3 will suffer heavily due to lack of interest by the average consumer that just wants to play the best looking games at the cheapest price.




ARchy

[nick²]

nick²:
Bandwidths are added when they can be used simultaneously due to being on separate buses. If you’re trying to tell me that the RSX will never use ~20GB/s bandwidth from XDR at the same time as using 20GB/s from GDDR3, then you might as well talk about how any bus isn’t really giving you their max bandwidth unless they are being used at the same time. Stop trying to partially apply “real world” usage scenarios or limitation in an imbalanced manner. Almost any of the points you are making in the PS3, apply to all other hardware, and all other parts contained in the PS3 itself. So really, what is your purpose in saying all of that?


I'm not sure I see the problem with my scenario. I want to know what you disagree with..

They posted the pic up from devstation... That showed the measured bandwidths from RSX to local memory. Its actually lower than what I am using for my numbers. You can see that there is a 25% performance hit from theoretical to measured bandwidth.

Then there is the case of cell using 40% of its memory bandwidth. I don't get it.. What is so unreasonable about this scenario? 7 separate processing units only using 10 GB/s ? That sounds quite reasonable to me.





Where the hell did 15GB/s come from? What do you mean by “available bandwidth”? Why pick 40% XDR memory usage? How the hell does the Cell usage of XDR RAM make the FlexIO <-> XDR Memory controller any more inefficient? Why are you still applying imbalanced and unfair comparison on components inside the same system? It’s beyond mind boggling. Where did the added drop to 13GB/s come from? You need to make your starting numbers clear, and every “nerf” you throw in after that fact, you had better explain it. When you do, apply the same applicable nerfs to whatever else you’re comparing to.


I thought I did make my starting numbers clear. As for the reasons why... well, its supported by the numbers in the photo, and its what commonly happens in that scenario. When you have a graphics chip like the RSX, it directly accesses the local memory pool. There is no waiting for its turn in the queue, no latency filled middle connection it has to go through. 22.4 GB/s is pretty much 22.4 GB/s. Yet, when you have to use an external memory bank, through an interim connection, there is always going to be bandwidth loss. Always. This is why an ondie memory controller is so valuable nowadays. The CPU then gets direct access to the memory, and all the other devices fetch memory through the CPU. When the CPU has to use an external controller, bandwidth suffers greatly.

So a 10% loss in bandwidth for going through a memory controller is rather conservative.



Why even pull the “worst case scenario” out of thin air? Why would any developer put all of their textures in video memory unless they already saw it reasonable?


To illustrate a point... The usage of two memory banks doesn't equate to improvement, even if they can operate simultaneously. Using mostly XDR at such speeds would equate to SLOWER transfers than if it was all done in the GDDR. I'm trying to show you why you don't add bandwidths. The extremes are the best way to illustrate why.

If we transfer 60MB from XDR and 0 from GDDR, we could very well get something like 13 MB/s. Now lets slowly change that to 59MB in XDR and 1 from GDDR. What happens? Do we start adding the bandwidth? No. (Data transferred)/ (time) still comes out to something slightly over 13 MB/s. Then, try 58MB and 2MB... 57MB and 3MB.... At about 50:50 it starts to NOT become a liability.. At 50:50 it would roughly be equivalent to just having it all come from GDDR.


I didn't just cover the worst case scenario, but I covered other scenarios as well.. and the way the math works out... at those speeds, 70%:30% -80%:20% mix of GDDR3:XDR works out the best, and its going to top out around 30 GB/s, except in exceptional circumstances (low XDR usage by cell).

Its very difficult to optimize this as well. Textures are used repeatedly in different settings. This would mean different combinations of different textures. It would be very hard to keep it so that the GPU always uses a 70% 30% mix at all times. There will be situations where you get 40:60 or 90:10... its unavoidable.

[Guilty Bystander]

1. Why would they be referring to the current most powerful GPU back in e3 2005 when the console was said to be coming out till spring 2006. I speculated it was more likely fall 2006 all along & hence they would have had enough time to produce a GPU with features like G80. READ THE THREAD

2. READ THE THREAD. Links were provided

3. What you are saying is speculation which I have read also on another forum
READ THE THREAD. Links were to NVIDIA 2003 patents describing a GPU with Audio

4. READ THE THREAD. Links were provided

5. It also doesn’t mean the technology has not been implemented. That’s why this is being discussed in a thread that is based partially on SPECULATION.

6. As well as Ken’s 4D, I gave two other links that pointed (i) A patent (thought to relate to G80) showing it UNIQUELY does 4D Textures in Hardware (ii). Naughty Dog running a Siggrapgh 2006 Course on a gaming technique that uses 4D Texture mapping techniques. READ THE THREAD. Links were provided

7. This was discussed a while back in beyond3D without a firm conclusion. I bought up the question again & same result. This combined with the article allowed me SPECULATE on the core being considered as NV47.

You really give the impression that you have not read anything to date in this thread even though you are one of the more frequent posters. I expect this of a new forum member but not you.

Archy


Well listen up here Archy.

You are only speculating on things that haven't even been hinted towards the RSX.
Just because the G80 and RSX are both 90nm (eventhough I remember reading something about 80nm) doesn't mean it's the same chip.
The Xenon is also 90nm still it's no way like the Cell eventhough they're both made by IBM.
Videocards support 3D textures which by the way have almost never (if at all) been used.
G80 will only support 4D texture lookups and not 4D textures.

The reason why I ignore some of the links you profided is because it's 100% based on speculations and ignores all the facts we've been given thus far.


Where the hell did 15GB/s come from? What do you mean by “available bandwidth”? Why pick 40% XDR memory usage? How the hell does the Cell usage of XDR RAM make the FlexIO <-> XDR Memory controller any more inefficient? Why are you still applying imbalanced and unfair comparison on components inside the same system? It’s beyond mind boggling. Where did the added drop to 13GB/s come from? You need to make your starting numbers clear, and every “nerf” you throw in after that fact, you had better explain it. When you do, apply the same applicable nerfs to whatever else you’re comparing to.

RSX can read and write with 20GB/s and 15GB/s from the Flex I/O theoreticaly and Sony told at DevStation (which you can see by the picture) that would end up in a read and write of 15,5GB/s and 10,6GB/s in realworld performance.
Considering the pretty big drop admitted by Sony themselfs this proves Flex I/O is pretty inefficiënt and the Flex I/O is depending on how much workload the SPE's need from the EiB or how much bandwidth the 7 SPE's will leave for the Flex I/O.
Also because the XDRAM first has to go from the EiB to Flex I/O to the RSX memory controller there will be bandwidth loss and latency would be quite high aswell.

[archy141]

Well listen up here Archy.

You are only speculating on things that haven't even been hinted towards the RSX.
Just because the G80 and RSX are both 90nm (eventhough I remember reading something about 80nm) doesn't mean it's the same chip.
The Xenon is also 90nm still it's no way like the Cell eventhough they're both made by IBM.
Videocards support 3D textures which by the way have almost never (if at all) been used.
G80 will only support 4D texture lookups and not 4D textures.

The reason why I ignore some of the links you profided is because it's 100% based on speculations and ignores all the facts we've been given thus far.






I don't think you still have got what this thread is about.
I have said countless times it is based on SPECULATIONS that i provided to fill idle time whilst we wait for more PS3 news from Sony.

Throughout this thread you have not given any reasoning to why you disagree or agree with my specualtions & the links i make.
Instead you come on here to shout paper statistics on the Xenos & RSX in attempt to turn it into a Versus thread. What i have speculated at is not completly meaningless to everyone -look at poll.
If you feel that I'm just speculating -its becasue i am & please get that in your head for once & for all. If your statistics don't fit in with the nature of this thread than maybe you should save them for elsewhere.

Speaking of G80 & 4D textures. If you had actually been reading this thread you would know that it was I that bought that link to this thread by pointing to the relevant patent.



Originally Posted by version


4D !

"A discrete BRDF is naturally stored in a 4D lookup table.
While commercially available graphics hardware has long had
2D texture mapping capability, no hardware currently in use
has the 4D texture mapping capability necessary to implement
this natural representation. Even if 4D texture mapping were
implemented, the tables required to hold a measured BRDF would
be quite large. At four bytes per sample (single-precision
floating point) and a modest sampling rate of approximately
twenty samples per radian (or a parabolic parameterization of
64.times.64 samples), a BRDF requires 64 MB of memory. Since
a modern graphics accelerator board has only 64 MB of total
texture memory, this storage requirement is prohibitively large.
Conventional texture compression algorithms can reduce the space
requirement, but not by more than an order of magnitude.
"
http://patft.uspto.gov/netacgi/nph-P...s&RS=Papakipos

The difference between nick² , EbonySeraphim & you is that they talk sense. On the other hand you don't make sense half of the time & give the distinct impression you are just parrot waffling.



Archy

[JimRune]

Eh, I'd really like to see it happen, and I'll keep my fingers crossed, but I doubt it's going to happen.

[Ramza]

I don't really have the patience or interest to do anything more than skim EbonySeraphim and nick²'s bickering about bandwidth, but I figure this article might provide them with some fuel.

http://www.spectrum.ieee.org/aug06/4256

I also don't have time to post this article to its own thread so if anyone wants to, go for it. I'm sure there are plenty of people who won't find it here.

[DarkChild]

that was a very interesting article, here it is: <br />
<br />
page 1: <br />
<br />
<br />
page 2: <br />
<br />
<br />
page 3:

[archy141]

*EDIT* just realised after posting mine DarkChild Already pasted it : ( <br />
I have decided to leave this as well as it highlights key points &amp; makes lighter reading. <br />
<br />
Thanks Ramza. That was a superb...

[Shred0r]

check this out:

http://ps3forums.com/viewtopic.php?t=29260


[...]The cheapest model starts at $17,500, and it just goes up from there! When they're released in September,[...]

WAIT, didn't NVIDIA said sometime ago that when the PS3 comes out the RSX is gonna be the most powerful GFX card available on the market?

YEEEESSSS, then the RSX is gonna be more powerful than the "GOD" card and you get the full package (PS3) for only $600. Now that i call value for money!

And whereas you would have to run 20 and more CPUs to avoid a bottleneck when using this GOD gard, the PS3 only has to utilize one Cell to work with the RSX without getting into some trouble...

!!! WOW !!! What a great architecture!


Or, according to Nvidia, you can have 12 mega pixel hi-def video, and a card (actually it's more like a series of cards) that will run every PC game due for release in the next century.

Sony already announced that the PS3 will last for at least 10 years on the gaming market and to make sure that the PS3 will last that long NVIDIA just had to build a GFX card into the PS3 that is at least as strong as their GOD card!

Therefore the one and only plausible conclusion has to be that the RSX is nothing less than a ultra-modified GOD card!

Now, take that, Xenos!!!

(p.s.: i hope you don't take my post too serious....)

[Guilty Bystander]

nick² & Guilty Bystander take note 256Bit bus is not feasible as you make out.


Sure a 256bit memory interface is more expensive but you always state it as impossible which it certainly isn't.



didn't realise this until reading the article:

XBOX360 only does roughly just over (assuming its 21GB/s evenly split)
10.5GB/s CPU-->GPU & 10.5GB/s GPU-->CPU

In contrast the PS3 does
20GB/s CPU-->GPU & 15GB/s CPU<--CPU


The Xenon to Xenos bus is 21,6GB/s (10,8GB/s read and write).

[Jabjabs]

No Xbox 360 game to date uses tiling but still use MSAA and HDR so what was Valve bringing again?

Where did you hear that?

They have been using Tiling since the start of the year, it was only the launch titles that didn't use it at all. Just because they are using MSAA doesn't mean they are';t using tiling, all the EDRAM does is take all the blending work out of the main memory pipeline. MSAA is still being used only now only on the EDRAM's current tile that it is working on.

The advantage that it gives is the fact that you aren't storing all the buffers on the main RAM. The trade off is a lot less bandwidth usage. a lot less memory storage required and the only draw back is that you have to figure out how to sort your geometry into several pools. The easiest way to do this is to simply render all the geometry once for each tile, not very efficient but it gets the job done.

[PlaystationArt]

Performance


An NVIDIA® GeForce™ Go 7900 GTX calculates the global illumination solution (light bouncing to convergence) for the scene in these clips above in about 3.3 milliseconds (300 frames per second) per frame treating all surfaces as dynamic. Static geometry can be handled much faster.


http://www.fantasylab.com/newPages/rtgi.html

This is why I believe the RSX is more than what it seems.

[Jabjabs]

Just because it is running on a Geforce doesn't mean that other cards can't do it, that seams to just be the test setup they had to work with.

Also they have one very bare website. They need to get someone onto that. ;)

[archy141]

[quote]

Sure a 256bit memory interface is more expensive but you always state it as impossible which it certainly isn't.





You are just digging your grave even deeper.


Its possible to put a DUAL Cell processor in the PS3 but we know clearly that would be IMPOSSIBLE to in the current PS3 due constraints such as costs. For the same logical reasoning people have been saying the likelyhood of seeing a 256Bit bus in PS3 is not possible, but you just continued to be ignorant of that.



Archy

[Crazy Phat]

How the fuck is this thread still going? :|

[Guilty Bystander]

You are just digging your grave even deeper.


Its possible to put a DUAL Cell processor in the PS3 but we know clearly that would be IMPOSSIBLE to in the current PS3 due constraints such as costs. For the same logical reasoning people have been saying the likelyhood of seeing a 256Bit bus in PS3 is not possible, but you just continued to be ignorant of that.



Archy


Sure I am! :lol:
The RSX already has the 256bit memory interface devided into XDRAM and GDDR3 (possibly with 1 64bit bus switched off).
So it does have quad memory controller so cost, difficulty or whatever excuse you want to bring to the table won't work because the RSX will have a 256bit memory interface it's just not totally dedicated towards the GDDR3.

[archy141]

@Guilyt Bystander
Your age is showing in the mentality of your replies.
Differention between you & the likes of EbonyS & Nick could not be clearer.




How the *expletive deleted* is this thread still going? :|

Becasue I haven't given up on it yet.

It aint over till the FAT LADY SINGS

[laststar]

I came across this guy on a forum that is dead set on the fact that RSX HAS TO BE LOT more than we were led to believe.
The amount of passion he has poured into pages of his thread just made me take notice & gave me something to think about -especially when Sony news is so dry (not including bored media doom stories)

Anyhow i reponded to him & i wanted to share it with you all.
Please don't flame as its based on speculations & possibilities that may or may never happen & probably errors- certainly will give you something to think & discuss about..





Keller,

I totally agree with you. It would be simply crazy for Sony to pay NVIDIA that large of a sum of money for a GPU that according to the official specifications that were released at E3 of 2005 is nothing more than a slightly modified N47 or 7800/7900.

Sony and NVIDIA claimed they spent tons of money customizing this chip, bragged on the manhours used, and called it a customized GPU. If it's simply a N47 with a few ROPS taken out, a bus to it's GDDR3 RAM reduced in size to 128bits, the FlexIO bus attached, and perhaps a little more texture cache here and there then Sony's frigging lost it!

However, that's why I think the RSX *has* to be something more significant than a dumbed down slightly tweaked N47. It doesn't make sense for them to put all that work into making such minor changes, it doesn't make sense for Sony to pay all that money, and it doesn't make sense that they would say so much about the RSX being *custom* if it's really nothing significant at all.

For the kind of money Sony paid NVIDIA they could have just done those small modifications themselves! Sony has scientists and engineers working for them that know enough about GPU design to make some basic changes and small tweaks. The only thing that makes sense is that the RSX is something truly new and unique and *heavily* modified from the N47.

NVIDIA has lots of interesting patents with interesting ideas about graphical processing and so does Sony and their other companies (that file patents in such a way they can't be traced back to Sony for secrecy purposes). It just makes sense to me that with all this knowledge, skill, and ideas Sony would want to put something *special* into the RSX instead of barely fiddling around a little with an ordinary PC part.

The secrecy is also astounding. I feel that probably Sony has put stricter NDA's on the RSX than *any* of us realize. In the corporate world NDA's can be severely tight and severely inforced. This excessive secrecy is pretty obvious because in almost every interview with developers so far (even in recent ones) the Cell and it's SPE's are mentioned almost *every* time in *every* interview.

BUT THE RSX IS NOT MENTIONED AT ALL!

If the RSX was just an ordinary N47 like what Sony released at E3 of 2005 I would not understand this severe secrecy. However, I have an idea and a guess I would like to share with all of you.

The RSX is unique, special, and very powerful. It utilizes some new technologies significantly beyond the currently released NVIDIA GPU's. This could be some sort of multicore feature or perhaps something we have no idea about at this time.

Sony could have even kept these new features secret from many or all of their developers until the last Dev-Kits were released *after* E3 of 2006.

Currently, the developers could be under a strict NDA not to mention these features, not to mention anything about the new specs of the RSX, and not to even mention the specs of the RSX that Sony had officially released in 2005. They also could have been told to avoid answering any question or making any comment about the RSX other than completely vague generic statements.

Sony could very well want to release the news about these new features of the RSX at TGS or some other event just before the launch.

Right now, the current consensus of *most* gamers out there is that the Cell of the PS3 is truly amazing but the GPU is pretty mundane. It's going to be powerful but nothing really impressive and by the time the PS3 launches it will be even less so.

By announcing these new features Sony could easily double the hype of the PS3's power by proving to the world that not only does the Cell have an amazing CPU but also a cutting edge truly unique GPU that's also quite stunning.

I could be wrong and the RSX might not be anything more than a slightly modified 7800. If that turns out to be true then that's the case and there is nothing we can do about it. But something inside of me just wants to know one way or the other.

This secrecy is becoming so annoying because we have heard so much about the Cell and it's potential. We have read information about how it can perform vertex work, ray casting, physics, animation, shading, and even texturing! It has been proven that the Cell (even if it's proven that it's not very practical) can do things that CPU's are really NOT supposed to be able to do!

But then we have the RSX.

Let me ask you a question. Please, consider this seriously.

What's neat about the RSX? What is truly *cool* about the Reality Synthesizer?

Can you think of anything truly amazing or even really neat you have heard about it so far? Even one clearly unique feature?

NOPE! Nothing really significant has officially been revealed about the RSX (even the stuff at E3 of 2005) that makes it anything more than a pretty powerful but at the same time fairly mundane GPU.

There is honestly nothing to really get excited about when considering the RSX!

To sum up:

The Cell: Super Powerful, Way Cool, Can do LOTS of stuff!

The RSX: Umm... Pretty good, fairly powerful, standard GPU?

Sony MUST have planned to produce a RSX more significant than the above!




Makeitlookreal - I just had to logon to say HOW MUCH I ADMIRE THE AMOUNT PASSION YOU HAVE IN YOUR BELIEF !

I also wish that its more than a custom (maybe cut down rather than beefed up) 7800/7900 as many are speculating it to be..

I also think it does not make sense to pair such a fine, well researched & developed NEXTGEN CPU with what is practically last years PC GPU with trimmings.

But industry does not always follow our wishful logic.. they have business logic & financial constraints. Maybe the RSX was a 7800/7900 derivative that should have shipped in a console back in spring 2006 & than it would have been as sold - MOST POWERFUL GPU OUT at that time. But hang on.. according to developers on beyond3d the XBOX360 GPU would still have been slightly superior to the RSX specs.

To add fuel to your conspiracy i have a few thoughts of my own..

What if Sony never had actually planned to release in spring 2006.. judging by where the console is now it would seem there was no way at all that it was ever going to be out for Spring. It was just marketing hype to keep the fans waiting rather than have them rush off to by the competitors console. The console was always planned for fall of 2006 & all project scheduling was done to that release date right from E2005's announcement. This scenario is all very possible to this point.

Between the release of the XBOX360 & now Sony would have had plenty of time to study the inner workings of its competitors console & know what they are up against. Remember they are telling us this console is for 10 years.
I believe there are three possibilities that can take place in regards to the final RSX Spec's:


1. They have kept RSX as spec'ed in E2005. As a result they have decided to no longer market the GPU spec's as they do not show any advantage over competitors GPU. With all the heavy investment Sony has already made & taken on loans, maybe it just was no longer financially feasible to upgrade the RSX even if they wanted to meet there initial goals - MOST POWERFUL GPU OUT AT RELEASE.

2. They have upped the RSX spec's. Same 7800/7900 with possibly more GDDR3 or XDR memory & slightly higher clock speeds. Maybe it explains the extra money given to NVIDIA for the undisclosed project.

3. They have completely changed the RSX so that when it is out in November as always intended it will clearly boast - MOST POWERFUL GPU OUT AT RELEASE.
But question is will it be a custom multicore 7900/7800 ?
or even better .. a new chip altogether based on the G80.

[quote]
NVIDIA’s code-named G80 graphics processing unit (GPU) will incorporate 48 pixel shader processors and an unknown number of vertex shader processors, some unofficial sources said. The chip is still expected to support feature-set of DirectX 10 along with Shader Model 4.0, even though it will not take advantage of the unified processors that can compute both pixel and vertex shaders.[ 02/20/2006]

http://www.xbitlabs.com/news/video/display/20060220100915.html

Now that would really be worthy of pairing with CELL & making a console that had the potential to get anywhere close to 10 years life span.

Just reading that description makes you feel that only this would fit the description that Sony seemed to have been boasting about. Maybe it explains the extra money given to NVIDIA for a undisclosed project.

It would definitely be the MOST POWERFUL GPU on market in November & for some time to come. With DX10 & shader4... we can see it lasting quite a long while.. It is also NON Unified processor like RSX is meant to be.

This too me is the glove that fits the best if Sony were to live up to the Hype on RSX & a console for 10 years .. and very wishful thinking of course.

Not to get hopes too high as these are all just equally possible - including no changes at all & RSX as Spec'd at E2005.

Unfortunately all we can do is wait for TGS & hope Sony let off some real BIG FIREWORKS.


Archy



UPDATE Taken From This Thread:
***************************



The RSX being a G80 Derivative is very possible.

I know it may sound a little far fetched but let me explain how & why I believe that the RSX is a lot more than what was on the E3 2005 spec sheet.

Sony has repeatedly said that they want to build True Nextgen System that will last 10 years – i.e. high long term investment plan.

Sony spent silly amounts of money to create a NextGen CPU
Sony spent loadsa money to create a NextGen BluRay Media
Sony spent bags of money to get NextGen HDMI1.3 Connectivity

Sony did not spent wads of money to get yesterdays NV47 core glued to Flexio.

The N47 GPU part does not fit at all with the selection of other truly NextGen components. There is nothing NextGen about it today & hardly back in spring.

Having followed every piece of PS3 news for over a year I took a step back to look at the whole history of events that unfolded pre E3 2005 to post E3 2006.

If you do this you will realise something very apparent that we fail to notice when we look at the daily picture - Sony NEVER had a Spring 06 release in mind.
It was impossible even if they had wanted to consider it as they were clearly way behind on there PS3 roadmap with too many milestones still to cover.

Sony have made so many changes between spring & now that it makes you realise spring was never meant to be. The PS3 has evolved & taken so much more shape since its expected release date, that it really does look like Fall 06 was always the intended release date. Even the dev kits seem to be in good supply with a few months to go.

Sony did not fudge up there project management by this much.
A reschedule to fall 06 for release was not just because of BluRay or HDMI as otherwise everything else could have been polished & ready to go apart from these components. Instead lots of things were & are still being changed or completed e.g. controller, case, OS (XMB/Internet browser/Live), RSX chips apparently rolling out just recently & not pre spring if they were to make it into a console.

I believe Sony has been masterfully applying Sun Tzu - The Art of War in there business.

They always intended to release PS3 around Fall 06.
They bought time with loyal fans by selling them not so far Spring 06 release date & they did a great job – I’m still here !

I believe the RSX figures they gave from NVIDIA were arbitrary for a top GPU that was going to hit around Spring 06 (7800). This was a strategic move by Sony to keep the fans satisfied, media in frenzied & MOST importantly to force there competitors to respond & bare there console secrets - XBOX360 GPU specs etc.
It was executed brilliantly!

It would have been utterly stupid business sense for Sony to reveal the true RSX spec’s & give the competition the heads up. Especially knowing the XBOX360 had not been released yet & was still possible to changes. Better to lull the competition into false security.

Noticed how RSX specs disappeared from all Sony presentations?
The values they gave at E3 05 served there temporary purpose. The real values will probably be divulged when they are good & ready to strike a blow.

Sony have since had plenty of time to weigh up the competition & make changes where necessary.. Motion controller.. Free online offering.. Standard HD :D etc
To date they have addressed individual strengths in there rivals console & responded by making the necessary changes to better if not equal them.
Sony is well aware of the XBOX360’s 2 strengths – Live & Xenos & they will surely counter attack them also.

It’s clear that Sony has lot riding on the success of PS3 due to the massive investment made. They seem to be deadly serious to win this war & they will not Microsoft’s XBOX360 get away with it.

The RSX CANNOT be what was on paper at E3 05.
That is plainly illogical if you apply Sony’s though of mind to date.

I believe they will deliver some kind of GPU that will deliver all they spoke of at E3 05 & more. Sony will definitely try to better Xenos without a shadow of doubt. They will have been listening to all the positives & negatives by developers on Xenos so that they can weigh it up against the RSX & make changes if need be.

First of all I was thinking like many hopefuls that it’s a custom built NVIDIA 7800/7900 derivative. But than the other day I came across details of a new NVIDIA GPU – The G80.

If we had a user requirements document for the RSX - the G80 would fit it like a glove.


-G80 will be most powerful GPU out when released & agree with what Sony/NVIDIA proclaimed of RSX

-G80 is DX10 & Shader Model 4 compliant which will increase its life span & make Sony’s ambitious 10 years for PS3 more realistic

-G80 does not use a Unified Processor like RSX even though it is DX10.All DX10 cards are supposed to use Unified Processors according to MS spec. Seems they were developing more for a system like PS3 (non unified processor reuirement) than Vista.

-G80 first NVIDIA GPU to support in built HDCP (Must for HD BluRay Movies over HDMI) as well as HDMI (wish I could confirm it was HDMI 1.3)

-G80 uses GDDR4 memory

Another possible way to look at the G80 is that it is the end product of NVIDIA's customisation that went into a G70 core to make the RSX. G80 is the RSX + more for the PC market & RSX is the custom G80 for the PS3

You have to admit that really does tally up with what we were sold as RSX.
In fact it is a more similar to the Xenos in design but with full DX10 + Shader model 4 support and no Unified Processor. Sony new back at E305 that the Xenos was part DX10 compliant. They knew than what they were up against & would not have settled for anything inferior. I’m sure they are aware of he value of investing in a DX10 Shader 4 card if they wanted it to last in there console for a decade. Would they really settle for DX9 GPU knowing what was around the corner (MS Vista pushing for DX10) ?

Maybe MS did go to NVIDIA first for Xenos but they were not able to deliver a Unified Processor GPU in time & cost terms. I believe MS have personal interests in Unified Processor architecture as it is part of Vista’s DX10 GPU requirement. It makes sense that MS would have wanted to have similar GPU’s in Vista & XBOX360 to make PC port easier & cross develop with XNA.

Sony on the other hand did not have requirements for Unified Processor & were able to contract NVIDIA to build RSX with DX10 + Shader 4.

The G80 will most definitely SMOKE the Xenos & outlast it with full its full DX10 & shader model 4 support. This truly is a GPU that you can see worthy of being in the PS3 with the CELL, BluRay & HDMI1.3. Not very hard to picture at all.

Most likely we will not end up with the G80 but instead some kind of custom derivative that shares its feature set & integrates well into the PS3.

Sony could put all of my logic & reasoning down the pan & the RSX turn out to be nothing like the G80 but I’m 100% convinced at worst case scenario it will not have the E305 spec’s. It will most definitely be more than we were told.

NVIDIA CEO : Postponing the PS3 was a master stroke



You may want to go the Ninja THeory's private board. Very interesting stuff to dig out there.

[archy141]

You may want to go the Ninja THeory's private board. Very interesting stuff to dig out there.

Thats where i started off from. I than expanded upon the idea & introduced lot more direction & reasoning not bought up before. Please take the time to flick through the whole thread as there are lot more interesting posts from me spread through it.
I will consolidate some of my postings onto the first page soon.


Archy

[The_One]

So it does have quad memory controller so cost, difficulty or whatever excuse you want to bring to the table won't work because the RSX will have a 256bit memory interface it's just not totally dedicated towards the GDDR3. If we went by YOUR logic (and what everyone else has been trying to say), the RSX would have a 384bit bus if there was a dedicated 256bit bus to the GDDR3 only.

I don't care about the bus to the XDR, because that's not an exclusive bus to the XDR; it's to connect to the CELL and the EiB via the FlexIO. On top of that, there's no proof that the bus to the FlexIO (remember, it's not a direct connection the XDR, otherwise there would be NO bandwidth and latency issue at all for the RSX using the XDR) is 128bits considering how the connection to the XDR itself from the CELL is 64bits.

Two ways of boosting memory bandwidth are to widen the data bus between processor and memory and to increase the rate at which bits are conveyed. It’s like adding lanes to a superhighway while also raising the speed limit. But widening the data bus requires adding more pins to the chips and more wires—the copper traces on a printed circuit board—to connect the chips. That extra circuitry pushes up costs, a highly undesirable outcome in today’s price-sensitive market for PCs and game consoles. Wow, people only listen when some sort of "official" articles points out the obvious, eh? Isn't this what I've been saying for ages :roll:? Thanks for the article, Archy.

Sure a 256bit memory interface is more expensive but you always state it as impossible which it certainly isn't. C'mon, I know you, and that's definately not something you'll say; it's simply silly. I mean, nothing is truly "impossible". Heck, it's probably only a matter of time before humans figure out the secret to clinical immortality :lol:. By "impossible" or "improbable", we're pretty much saying that it can't happen because of the constraints in reality, be it physical factors or economic factors.

How the *expletive deleted* is this thread still going? :| You'd be surprised at the kind of threads that are still going :lol:. Just look at the "What are you listening to right now?" thread in the General Chit Chat section. 40-something pages and still going strong :shock:!

Just because it is running on a Geforce doesn't mean that other cards can't do it, that seams to just be the test setup they had to work with. I agree. They probably used that test setup because (from the benchmarks I've read), the GeForce7900-series graphics cards seems to be the better GPU (as compared to ATi's equivalent, of course) on the market right now (When I say "series", it includes the new 7950 ;)).

With all that said, I still have to say: It's highly improbably (heck, I'd even say "impossible") that the RSX is a G80 derivative; but hey, crazier things have happened :shock:!

[CaptHavok]

You know what I am quite speculative at this point for in 2005 what was said by the CEO of Nvidia and what is in this article http://www.penstarsys.com/editor/company/nvidia/g70_spec/index.html

it seems clear yet unclear as to which chip Nvidia has handed Sony being that it has been handed over only a few months ago.

I myself cannot be sure for certain reasons.

1. G70 was asstonishing in 2004 and a good compliment to the cell chip but to be the best this chip is becoming outdated.

2. G71 is prone to overheating problems and would not be a good companion to the Cell chip in a tight case like the PS3.

3. and finnally it has been recently that Sony has stated that the HDMI will be 1.3 and have support for True Color a feat not specifically stated as a function of a G70 or a G71.

perhaps we could see something else maybe a EVGA e-GForce 7900 chip but I would not go as far as stating it would be a G80 but heck if Sony and Nvidia do this thing I will be so happy to pull $599 plus tax out of my wallet for a PS3.

[Hendo..2006]

Ummmmmm exuse me but the poll says that its gonna be a G80 chip 38% vs. 35%

G80 it is then !!!!!! :wink: :lol:

[Grym]

This poll shows that a lot of people will be disappointed with RSX if it isn't a G80.

IMO it is very unlikely that it is a G80. Modified G71 is more likely.

[HiX]

This poll shows that a lot of people will be disappointed with RSX if it isn't a G80.

IMO it is very unlikely that it is a G80. Modified G71 is more likely.

I personally think it's a hybrid of G80 and G71 tailored for CELL but heh, that's just me.

[Hezz]

Considering that the G80 and the RSX are coming out at about the same time period it seems almost certain that one or the other would not be based on the others technology. Both would be based on earlier technology and yet it is still possible, even perhaps somewhat likely that the RSX and G80 share some technology.

My feeling is that the Sony/Nvidia/RSX project may have had a deep impact on Nvidia's view of what a future GPU for the PC should be like.

I am still holding out hope that the RSX will be something special and in all honesty the circumstancial evidence seems to tilt in that direction.

[The_One]

Ummmmmm exuse me but the poll says that its gonna be a G80 chip 38% vs. 35%

G80 it is then !!!!!! :wink: :lol: Uh... Right. A little forum poll is out to change the world, right :lol:?

[Ramza]

What??? This is a forum poll... I thought we were electing the RSX... Darn. :lol:

[Guilty Bystander]

Well Admins I think it's time to close this one as all lifeforce has drained out of it.
Whatever the RSX will be I'm quite graphics will be awesome either being a G71, G71/80 hybrid.

[mrboo]

Well Admins I think it's time to close this one as all lifeforce has drained out of it.
Whatever the RSX will be I'm quite graphics will be awesome either being a G71, G71/80 hybrid.

Agreed, how can one fanboys silly imagination and dreams wheel in so many people?

LOCK PLEASE ADMINS

[archy141]

There are NO valid reasons for this thread to be LOCKED

The still changing poll only goes to reflect the interest people have in this thread.
It should be left open till we are able to confirm the RSX spec's or silently allowed to drift to the bottom of the forum.

There are lot more pitiful threads that are irrelevant to the PS3 being created on daily basis that instead need to be considered for locking. This thread definitely has substance & contributions from the best of members on these forums.

I feel the people requesting this thread to be LOCKED are the ones that gave no valid contribution towards it from the start & instead tried to derail it all along.

I hope to make updates to this thread if & when new relevant information is found. Until valid reasons for & against what has been SPECULATED in this thread ar most welcome.


"Without speculation there is no good and original observation"
Charles Darwin



Archy

[Itaintrite]

Well Admins I think it's time to close this one as all lifeforce has drained out of it.
Whatever the RSX will be I'm quite graphics will be awesome either being a G71, G71/80 hybrid.

Agreed, how can one fanboys silly imagination and dreams wheel in so many people?

LOCK PLEASE ADMINS
One can say the same about Da Vinci.

[xen]

Well Admins I think it's time to close this one as all lifeforce has drained out of it.
Whatever the RSX will be I'm quite graphics will be awesome either being a G71, G71/80 hybrid.

Agreed, how can one fanboys silly imagination and dreams wheel in so many people?

LOCK PLEASE ADMINS

If you dont agree with the speculation put forward then fair enough but why should the thread be locked?

I'm personally itching for Sony to finally start revealing some nuggets of info on PS3 the silence is killing me.

In regard to RSX wouldn't it suit both parties to share development costs across both the RSX and G80 graphics chips where possible. Like the core graphics pipeline or improved shader technologies. Then have seperate teams for PC/console specific features.

[nick²]

The last thing this forum needs is another locked thread.

[arfi-gorgona-O]

Damn 69 people believes that RSX its a G80...Well ok,but don't blame Sony after ok? :shock:


It is pointless to do with more what can be done with fewer.

[nick²]

To me, its very funny.. people get their hopes up so high, and set this impossible standard.. I don't get how they are not dissappointed when they find out they are wrong. Oddly enough, they never seem to be dissappointed by bad news when it strikes...

I guess they are just optimists?

[GunTeng]

I see this thread as a bit of fun & an interesting read.
Its hardly the law of the universe & there is absolutely no reason to lock it.
If you have no imput to add to the thread, then don't say anything & it will slowly disappear.

If some idiot wants to believe someone elses guess at a possible outcome, then thats their problem.
You don't get bookies being closed down for a favourite to win horse coming last.

Maybe its is a G80.
Maybe it isn't a G80.
At the end of the day, it doesn't matter as long as the PS3 is what it is supposed to be. A next-gen console that is fully HD, backwards compatible...blahblahblah the list goes on ( of course with superb graphical prowess ;)).
But who in gods name ever said that the G80 was the only graphics chip capable for performing superb graphical prowess on the PS3?:)

[Crazy Phat]

I don't really care at this point as the PS3 already has games that put the other next gen consoles to shame.

Most graphically beautfil RPG: Final Fantasy XIII Trilogy
Most beautiful flying game: WarHawk IMO
Most beautiful next gen tactical action espionage: Metal Gear Solid 4
Most beautiful mystical game to date: Lair
Most beautiful action adventure game to date: Naughty Dog's New Game
Most beautiful co op two player game: 8 Days(if it was gameplay)

There are alot more games that I could mention but meh. These games look better than alot of the other console's games in my opinion. These games already show me that the PS3 is the most powerful machinery in terms of CPU power, and GPU power. :D

[HighPs3]

The G80 spec has the lead in the poll so far!

Thats crazy!

I hope you guy/girls are right!

[Graham]

This whole 10 years thing is beginning to annoy me. This technology will not last 10 years. It's plain and simple. In 10 years, technology will have improved so much that the PS3 will be outdated and long forgotten. The whole 10 years thing is just a marketing gimmick.
PSOne lasted 10 years, so why can't PS3? "lasting" 10 years does not mean it will be the primary PlayStation platform for the next 10 years. In 5 or 6 years time we'll see PS4, at which point the PS2 will have "lasted" 10 years...

[Chaotic_King]

This is ridiculous. The RSX is the same spec as last year, or slightly decreased for yield reasons. It is NOT a G80. It's based on the NV47, which is a G70.

[Lozt_again]

This is a long running thread eh!?

Somehow I doubt it'll be little more than a Tweaked original spec RSX.

[Lefein]

This is ridiculous. The RSX is the same spec as last year, or slightly decreased for yield reasons. It is NOT a G80. It's based on the NV47, which is a G70.

7900 is also NV47.. Just in case you were wondering..

[archy141]

I get the distinct impression you have not really read through this thread as otherwise your post would not have such a tone. I hope your avatar is not an indication of tainted opinions that come with no reasonable arguments.

No one is saying outright the RSX IS G80. Under the discussed reasonings we are merely speculating possibilities that may fill missing gaps in RSX's removed spec's. If the first gen PS3 games are anything to go by we are definitely in a WIN WIN situation whatever form RSX takes !



The RSX is the same spec as last year, or slightly decreased for yield reasons

Can you provide some evidence or reasoning for you to say this ?


Archy

[HighPs3]

This is ridiculous. The RSX is the same spec as last year, or slightly decreased for yield reasons. It is NOT a G80. It's based on the NV47, which is a G70.

7900 is also NV47.. Just in case you were wondering..

He's got links and facts though, I mean a much better post then the rest, clearly!

[Lefein]

He's dead wrong on his assumptions. Back at TGS 05 many developers had some concerns about the kits they were getting and EA even piped up saying that the "PS3 is fill-rate bound".. Sony ponied up immediately after TGS 05 and NVidia extended development of RSX for at least another three months.

Question:
Why would Sony be concerned about yields on a GPU they hadn't even taped out? ... exactly ;)

And about ths G80 mess... When NVidias chief says "We have a lot more mileage to go with this architecture" what exactly makes you think that a unified shader proc will be in the PS3? They have flatly stated that the RSX will be NV47 based and have stuck to that.

[archy141]

This is ridiculous. The RSX is the same spec as last year, or slightly decreased for yield reasons. It is NOT a G80. It's based on the NV47, which is a G70.

7900 is also NV47.. Just in case you were wondering..

And in G naming convention it is called the - G71

It can do 650Mhz clock speeds & is more efficient than the G70 as it has a lower transistor count (278 Million versus 300 million transistors) enabling the higher clock speeds.

Also discovered that there is a NEW version of the 7900GT/GTX shipping that will offer integrated HDCP support. Previously i was only aware of the G80 natively supporting HDCP.These revamped 7900's will be denoted as G73-B1.


In general, HDMI connectors used by the HDCP graphics cards incorporate audio and video into a single connector and and synchronize the audio and video signals. Nvidia's high-end 7900 (G73-B1) incorporates the audio mixing
http://www.guru3d.com/newsitem.php?id=4116

This news makes the 7900 fit the RSX bill little better than before to me.



*EDIT*


And about ths G80 mess... When NVidias chief says "We have a lot more mileage to go with this architecture" what exactly makes you think that a unified shader proc will be in the PS3? They have flatly stated that the RSX will be NV47 based and have stuck to that.

G80 is NOT a unified shader proc.


Nvidia’s code-named G80 graphics processing unit (GPU) will incorporate 48 pixel shader processors and an unknown number of vertex shader processors, some unofficial sources said. The chip is still expected to support feature-set of DirectX 10 along with Shader Model 4.0, even though it will not take advantage of the unified shader processors that can compute both pixel and vertex shaders.
http://www.xbitlabs.com/news/video/display/20060220100915.html



Archy

[Lefein]

Meh, I dont keep up with future stuff.

Oh, hey! You're right about the HDCP revision. HDCP is in the 6150 and the...


7950 ;)

[Redrider]

Nice thing about this thread is it has brought a lot to the table with minimal flaming, something that has been a rarity in recent threads. And I seriously hope that those who think the RSX will be G80 based are not disappointed if Sony says otherwise. I will re-affirm my position (about 14 pages ago). My vote goes to a modified version of the G71 (G73.5??).

[Hezz]

Locking this thread would be a crime. Since when have we had such a fun thread with so much good quality speculation. The majority of the threads started are uninteresting but this one has some meat to it. And best of all it will be a continuing drama until we all know what the RSX is really about. Hopefully we will find out at TGS or soon thereafter.

[Aleman]

They said "G70x" at GDC. Maybe that means the RSX core is between a G70 and a G71 (between 7800 and 7900).

Hey, whadya know, the transistor count and clock speeds agree with that. I think I already came to this conclusion like ten pages ago though.

[Reaktorleak89]

That would make sense. I'd be more than happy with that kind of power considering how powerful 7900s are. And in a closed console environment even the shoddy looking games will still end up with pretty good graphics.

[Lefein]

They said "G70x" at GDC. Maybe that means the RSX core is between a G70 and a G71 (between 7800 and 7900).

Hey, whadya know, the transistor count and clock speeds agree with that. I think I already came to this conclusion like ten pages ago though.

Haha, yeah, I could see it being something like a 7850 or so.. It's just too bad most people getting their hopes up are going to look at that like it's a bad thing.. The 7800 rocks! Hell, if it's 7900 based I'll be ecstatic! Like, I might spontaneously poop kittens and bunnies happy!

[Chaotic_King]

You are all setting yourself up for a fall.
The 7900 is based on the 7800, but the RSX has many differences, more relating it to a 7600GT.

[The_One]

You are all setting yourself up for a fall.
The 7900 is based on the 7800, but the RSX has many differences, more relating it to a 7600GT. You do know that what your saying makes no sense whatsoever, right? The 7600 series is based on the G70, it's actually newer; simply down scaled (lower clock and less pipes) as compared to its older brother, the 7800. Unless you're trying to say that an old design like the RSX is somehow the same as a NEW GPU design; which wouldn't make any sense either since it would have as much chance of being a 7900 series card as a 7600 series :?.

[gsss124]

hello,
THEREALNEXTGEN,
I just wanted to *thank you* for improving my knowledge.
I learned what is Valedictorian and who was Theodore Roosevelt!
This forum is more than games, it improves your knowledge!

[wotter]

Meh, I dont keep up with future stuff.

Oh, hey! You're right about the HDCP revision. HDCP is in the 6150 and the...


7950 ;)

I think there is also an MSI 7900 GT with HDCP :wink: .

[RavenLord]

why isn't there any explanation for this ? If they have better GPU, they could use it for a huge adv. boost..

[archy141]

They said "G70x" at GDC. Maybe that means the RSX core is between a G70 and a G71 (between 7800 and 7900).

Hey, whadya know, the transistor count and clock speeds agree with that. I think I already came to this conclusion like ten pages ago though.

Haha, yeah, I could see it being something like a 7850 or so.. It's just too bad most people getting their hopes up are going to look at that like it's a bad thing.. The 7800 rocks! Hell, if it's 7900 based I'll be ecstatic! Like, I might spontaneously poop kittens and bunnies happy!

From what I have researched the main difference between the 7800 & 7900 is the latter has more effecient design with a lower transistor count. This allows the 7900 GPU to be clocked higher whilst keeping temperature lower than you could with 7800.

It would therfore make more sense to have a 7900 like RSX @550MHz than a 7800 where heat dissipation in a console is going to be of high concern.


Product Name/# pixel processors/# vertex processors/Bus width/Memory/Type/Amount/GPU Speed/RAM Speed


GeForce 7900 GTX 24 8 256-bit GDDR3/512MB 650-700MHz 1600 MHz
GeForce 7900 GT_ 24 8 256-bit GDDR3/256MB 450MHz 1320 MHz
GeForce 7800 GTX 24 8 256-bit GDDR3/512MB 560MHz 1600 MHz
GeForce 7800 GTX 24 8 256-bit GDDR3/256MB 430MHz 1200MHz
GeForce 7800 GT_ 20 7 256-bit GDDR3/256MB 400MHz 1000MHz

G80____Geforce 32 16 256-bit GDDR4/256MB 700MHz 2500Mhz*

*Figures guessed at as spec's not released.

Before we all start swinging RSX is more of 7900 design we must remember the RSX will be a tailored custom solution based from PC GPU. It will not EXACTLY be any of the above but have trimmings in some areas & tweaks in others to suit the PS3 rather than the PC architecture.
Also not to let G80 out of picture so soon:



Now for more fuel on the RSX & G80 Link (getting rather scarce)


The other thing we heard is that the G80 is going to be mostly G70/G71 with DirectX 10.0 stapled on. Meaning, that since the architecture is basically G70, it has not been built from the ground up, keeping DX10 in mind.
http://www.cooltechzone.com/Departments/Featured_Story/NVIDIA_G80_Delayed_200604092276/"


I came across this article suggesting the G80 is closer to G70/71 than we possibly think. If that is the case the whole idea of RSX being G80 like would seem even more possible.

G80 is looking to be the first true dual-core GPU, as opposed to multiple cores per PCB and SLI. This will obviously make it bigger unless the RSX is just a smaller subset of the G80. Recall PS3 did get fatter even though we lost from the original feature list.

Majority of the community is set in concrete about the RSX being NV47 core based (G70/71). The idea that it could be an altogether new G80 core cannot be accepted by many. If this article is true than the G80 has a lot more in common with the NV47 core than people realise. I have posed the question on Beyond3D & no one has said outright that it’s a new breed of core altogether.

If the G80 is a NV47 core 'stapled' with extra features (SS4, GDDR4 etc) than there is even more possibility that the RSX could be like G80.



BTW


Meh, I dont keep up with future stuff.

I don't believe you for a second :wink:



Archy

[Cellman200]

RSX not based on G80. RSX based on G71.

Graphics processing unit (GPU)
Custom RSX or "Reality Synthesizer" design co-developed[citation needed] by NVIDIA and Sony:

• Based on NVIDIA G71 architecture, otherwise known as NV47
• Clocked at 550 MHz
• 128-bit DDR memory interface
• 211.2 GFLOPS programmable (384 FLOPS per clock)
• 1.8 TFLOPS
• Multi-way programmable parallel floating point shader pipelines
• 136 shader operations per clock
• 74.8 billion shader operations per second (100 billion with CPU)
• 24 2D texture lookups per clock
• 1.1 billion vertices per second
• 128-bit pixel precision offers rendering of scenes with high dynamic range imaging
• Full high definition output (up to and including 1080p)



http://home.btconnect.com/hgi/ps3/

Prove in link

[archy141]

RSX not based on G80. RSX based on G71.

Graphics processing unit (GPU)
Custom RSX or "Reality Synthesizer" design co-developed[citation needed] by NVIDIA and Sony:

• Based on NVIDIA G71 architecture, otherwise known as NV47
• Clocked at 550 MHz
• 128-bit DDR memory interface
• 211.2 GFLOPS programmable (384 FLOPS per clock)
• 1.8 TFLOPS
• Multi-way programmable parallel floating point shader pipelines
• 136 shader operations per clock
• 74.8 billion shader operations per second (100 billion with CPU)
• 24 2D texture lookups per clock
• 1.1 billion vertices per second
• 128-bit pixel precision offers rendering of scenes with high dynamic range imaging
• Full high definition output (up to and including 1080p)



http://home.btconnect.com/hgi/ps3/

Prove in link


Sorry no proof, arguments or reasoning to be found there.

It's just a paste of Sony's spec's released back in e3 2005.

Maybe you should read through this thread & don't just assume it is implying RSX is G80. There are plenty of reasoning & arguments given that you are welcome to challenge.


Archy

[Deldwen]

will they release the specs at TGS?

[spazchicken]

nobody knows if they will or not, it's all up to Sony...

[turdhat]

This is ridiculous. The RSX is the same spec as last year, or slightly decreased for yield reasons. It is NOT a G80. It's based on the NV47, which is a G70.

7900 is also NV47.. Just in case you were wondering..
Yes but all the 7900 is over the 7800 series is increased memory and core clockspeeds and a shrunk die to 90nm.

[proof]

How did this speculation get to 18 pages long? :?: :shock:

[Aleman]

How did this speculation get to 18 pages long? :?: :shock:

A couple weeks of slow news will do that ;)