3-4 Tflops sounds a bit high to me. Keep in mind that next gen consoles will likely retail for around $400. If you look at the (rumored) first leaked dev kits for each next gen system : PS4 dev kit was at 1.25Tflops (GPUs), and the 720 dev kit was at 1.4 TFlops (GPU). These first dev kits are 6 times more powerful then the current hardware.
I think when these consoles are released they'll both sit around 2.5TFLOPS for total GPU power. Primarily because of statements made by Epic's Tim Sweeney, who stated that Samaritan Demo would require next gen consoles to have 2.5TFLOPS of graphics power to render the demo at 1080p. Epic had a big hand in the development of the 360, and this generation has seen Unreal Engine 3 become the go to engine of choice for developers, I'm sure Unreal Engine 4 will have a similar impact.
2.5TFLOPS is 10 times more powerful then what we have now, I can't wait to see what kind of things developers will do with that kind of power.
Rumors point to the next xbox have an 8 core CPU, while Sony would have 4. For Sony to add a second CPU would be extremly costly, and wouldn't be very power efficient. Don't forget that xbox 720 will not only have kinect 2, but also rumored body sensors, heads up display, and DVR functionality; these could easily chew up 2-4 cores. It's likely that the 720 will need the extra 4 cores just to stay even with the PS4.
Latest PSU headlines:
Results 76 to 100 of 159
-
12-02-2012 #76Newbie







- Join Date
- Dec 2012
- Location
- MN
- Posts
- 6
- Rep Power
- 0
- Points
- 182 (0 Banked)
-
itachi73378 wants to slowly undress this post.
-
12-02-2012 #77Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
Yeah the PS4 will be somewhere between 2-2.5TF (1.84TF GPU + 0.6GF APU) and we'll get to see some great visuals,physics etc
Latest devs kits are A10s with devs using anywhere from 7850 to 7970s to simulate the final hardware.
Thinking they will use a derivative of the 8850 or somethingLast edited by itachi73378; 12-02-2012 at 22:45.
-
12-05-2012 #78
-
12-05-2012 #79
I need POWAH for 400-500 dollar range.
Tha Fuck? I don't need no relationship with my tv, sick bastard. Why the hell should I pay you more money to use features I can get at the touch of a button on my remote? Oh, you want me to use these features with a camera and my voice instead? I'm not that damn lazy and I want to save some money FOR GAMES.
-
12-05-2012 #80
-
12-07-2012 #81Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
How do you figure that? I notice many people do not take the RSX into account.
I know it's hard to do a straight up compare, but you can't ignore the RSX in your calculations.
3.2 GHz CELL (1 PPE, 6 x SPU): ~180 GFLOPS
Note: CELL has 8 SPU, but 1 SPU is fused off during manfacturing, and 1 is available only to the OS, and not games.
500 MHz RSX: ~360 GFLOPS
Total: 540 GFLOPS
2.5 TFLOPS would only be about 5 times more powerful.Last edited by Alpha; 12-07-2012 at 19:17.
-
12-07-2012 #82Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
-
12-07-2012 #83Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
Of course you will never hit the maximums, but that applies to either generation. The next generation will be more efficient, cause now we are dealing with unified shaders (better load balancing per scene). Maybe a difference of 50% to 80%? But really, it depends greatly on the developer, and what they are trying to render.
-
12-07-2012 #84Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
Well the bottom line for me is that 2-2.5 teraflops is what is required for Samaritan graphics according to Epic's Time Sweeney and we'll definitely get Agni's philosophy visuals with that kinda power in the late stages of next gen.
I don't really care about the numbers as long as we get dem graphics
-
12-07-2012 #85Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
Samaritan was a Unreal 3 demo. I want hardware that can do the Unreal 4 Elemental demo. Tim Sweeney basically said that would would need an AMD Radeon 7970 or GeForce GTX 680 to pull that off.
1 GHz AMD Radeon 7970 - 4.1 TFLOPS
I think a 4 GHz (800 MHz GPU) AMD A10-5800K APU (0.7 TFLOPS) with a 1 GHz AMD Radeon 7950 (3.4 TFLOPS) should make for a good combination to achieve that.
An A10 with a 8950 would probably be more suitable.Last edited by Alpha; 12-07-2012 at 21:36.
-
12-08-2012 #86Newbie







- Join Date
- Dec 2012
- Location
- MN
- Posts
- 6
- Rep Power
- 0
- Points
- 182 (0 Banked)
The reason I figure 10 times more powerful is because I was focusing solely on the GPU. I was paraphrasing a quote from Tim Sweeney, who was saying GPUs needed to be 2.5 TFLOPs. RSX is 550mhz, not 500; and does 228.8GFLOPS, not 340. The xbox GPU is 240GFLOP for comparison.
http://www.ign.com/articles/2005/05/...ation-3?page=3
Samaritan was based on a heavily modified version of Unreal 3, which was close to Unreal 4 in architecture. I can't remember who at Epic, but it was stated that to get the full features of Unreal 4 in 1080p a GPU will need to do at least 1Tflop.
http://www.geforce.com/whats-new/art...-time-graphics
An A10 with an 8950 would be amazing, but will never happen. Sony is looking to save some money on a GPU by having a GPU built into the processor (A10), because of that they'll be able to opt for a less expensive, less power hungry gpu (less heat). I'd expect something more in line of the 7700/7800 or 8700/8800.
-
12-08-2012 #87Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
I think the recent rumors about a 1.84 teraflop GPU are correct. Sony and MS are both pushing for native 1080p and 60fps (60fps is not mandatory)
So they will probably go with a derivative of 7850. maybe 8850 at best
-
12-08-2012 #88
There were some information in a paper about UE4 elemental demo.
It ran with GTX680 ~30fps and 90% of the time at 1080p.. (most likely using dynamic framebuffer.)
GTX680 is ~3TF GPU.
If we get <2TF GPU we will not see 1080p with 60fps or even 30fps with the effects like we saw in Elemental demo.
I really hope we will get as new architecture as possible and with as much power as possible a 3-5TF would be nice.
Preferably with an interposer/chip stacking to get fast memory with wide I/O for a GPU and a CPU. (256MB-4GB with >~1024bit I/O.)
http://www.tezzaron.com/about/papers...712-dist-2.pdfLast edited by jlippone; 12-08-2012 at 17:45.
'no, no one in their sane mind uses OpenGL on PS3' - Repi
'nope, PS3 uses a wonderful low-level API called libgcm' - Repi
-
12-08-2012 #89Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
-
12-08-2012 #90Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
Actually the specs that Sony released for the PS3 on the RSX clock rate is not correct. Sony released the RSX at 500 MHz, and not 550 MHz, and I think that was to keep the heat down.
The wikipedia article on RSX lists 400.4 GFLOPS for RSX at 550 MHz.
From that Nvidia article:It's like the minimum specs that are listed for running a game on a PC. In other words, they quote the bare minimum to maximize sales, even if it looks and runs poorly on such a listed spec.Unreal Engine 4’s next-generation renderer targets DirectX 11 GPU’s and really starts to become interesting on hardware with 1+ TFLOPS of graphics performance,
I think A10 with 8850 (1000+ shaders?) is probably what Sony will go for, largely due to managing heat in a console size box. Unfortunately, that in my opinion is inadequate for the next generation. 8950 (around 2000 shaders?) is where it should be, even if it means a bigger console, with more fans.
MOAR CORES!!!
The 7950 (1792 shaders) has dropped a lot in price in the past year, from around $450 to $300, so a chip at that size (28nm, 4.3 Btransistors, 352mm˛) is getting quite affordable. Much more so in late 2013, and a die shrink to 20 or 22 nm, and a redesign with less transistor leakage/power saving modes would help a lot in the heat department.
The nice thing about the 7950, is that it's a 7970 with parts fused off, for greater manufacturing yields. That would be probably the same for the 8950.
Console developers can optimize their code for a console's specific architecture, over PC code, so the specs to pull of the Elemental demo might be much lower that a GTX680. But in the end, sometimes you just need lots of brute force to get the job done.
Another thing to keep in mind is that any 8000 series chip might not be ready in time for a late 2013 console introduction, because AMD has pushed back the release of the 8000 series to mid-2013. They indicated it's taking them longer to get the technology right for Heterogenous System Architecture (HSA).
HSA is important so it would be nice to have that in a next generation console.Last edited by Alpha; 12-08-2012 at 18:31.
-
12-08-2012 #91Apprentice







- Join Date
- Jul 2008
- Location
- Michigan
- Posts
- 206
- Rep Power
- 0
- Points
- 2,134 (0 Banked)
the 8850 is looking like a great but if it's only $199. That's cheaper than the 7850 is and its been out for almost a year now and it started at $250. $400 for a crossfire 8850 setup is looking like my next setup if this is all true
-
12-08-2012 #92
I'm sure that you can have all features supported with 1TF of computing power.
Problem is that if one looks at information from Elemental demo, it means that the 1TF power gives you a 720p and around <~20fps.. (If scaling is linear to GFLOPS.. (which it isn't))
Combined power is also not as easy to utilize as in case of one GPU, it really depends on how well those can be used together. (2TF combined would be sore disappointment, 4 would be better..)
Stacking memory and interposer should ease the power consumption quite nicely and thus the generation of heat. (data transfer is the biggest problem on chip design.)
Why it wouldn't be ready?
They have had the base chip done a long time ago and the console version of the chip should be done separately from the PC parts. (from certain point forward.)
It's good to remember that they have people working on 9xxx and most likely 10xxx series GPUs at the moment.Last edited by jlippone; 12-08-2012 at 18:30. Reason: readability, coherency, the usual.. ;)
'no, no one in their sane mind uses OpenGL on PS3' - Repi
'nope, PS3 uses a wonderful low-level API called libgcm' - Repi
-
12-08-2012 #93Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
If they are sharing similar technology with Heterogeneous System Architecture (HSA) in the Graphics Core Next (GCN 2.0) architecture, and it's delayed on the PC side, because of that, it would be natural to conclude it's also delayed on any console part, unless you have information saying otherwise?
Keep in mind, AMD has said mid-2013 for the GCN 2.0 architecture, but they also may slip past that date.Last edited by Alpha; 12-08-2012 at 18:57.
-
12-08-2012 #94
If it's not same chip, mid 2013 is far enough to get console with GPU out within same year.
I would bet that it would be possible to release console in 2013 with early 2014 series GPU.
Especially so if the console uses variation of mid/high end GPU and not top of the line one.
We already have had presentations on 2014 GPUs and papers on research for 2017..
Yes, I'm optimistic for I know that more programmability & accessibility directly means better games from programmers.
Last edited by jlippone; 12-08-2012 at 19:29.
'no, no one in their sane mind uses OpenGL on PS3' - Repi
'nope, PS3 uses a wonderful low-level API called libgcm' - Repi
-
12-08-2012 #95Dedicated Member







- Join Date
- Jul 2006
- Posts
- 1,382
- Rep Power
- 54
- Points
- 2,773 (0 Banked)
Here's my ideal tech for PS4:
AMD "Fusion" APU
~200W TDP
1 ARM Cortex-A15 CPU Core @ 1 GHz (very low power standby/sleep mode processor)
4 "Steamroller" CPU Cores @ 4 GHz
24 Compute Units (CU), 1536 Graphics Core Next 2.0 (GCN 2.0) Shaders @ 1.2 GHz (3.7 TFLOPS)
Heterogenous Systems Architecture (HSA)
128 Texture Units
128 Z/Stencil ROP Units
48 Color ROP Units
12 MB Shared Cache (CPU and GPU access the same memory space, as per HSA, so cache is shared also)
28 nm, ~4 billion transistors, 300mm˛
8GB GDDR5 Memory, 1500 MHz Memory Clock (6 Gbps GDDR5), 512-bit (384GB/s) memory bandwidth (later can be on interposer, if tech not ready by late 2013)
750GB HD 7200 RPM SATA
10X Blu-ray Drive SATALast edited by Alpha; 12-08-2012 at 20:39.
-
12-08-2012 #96Master Guru







- Join Date
- Nov 2010
- Location
- Mordor
- Age
- 19
- Posts
- 6,532
- Rep Power
- 56
- Points
- 4,840 (130,110 Banked)
-
12-08-2012 #97
If there will be something on interposer, it will start as such and move on chip later.
This is the 'cheaper' way to do it and more memory efficient.Last edited by jlippone; 12-08-2012 at 21:40.
'no, no one in their sane mind uses OpenGL on PS3' - Repi
'nope, PS3 uses a wonderful low-level API called libgcm' - Repi
-
12-10-2012 #98Member







- Join Date
- Sep 2005
- Posts
- 138
- Rep Power
- 57
- Points
- 2,763 (0 Banked)
-
12-10-2012 #99
-
12-10-2012 #100
Thread Information
Users Browsing this Thread
There are currently 2 users browsing this thread. (0 members and 2 guests)





Reply With Quote







Bookmarks