It would barely be faster than the hd7970, a larger chip, and wouldn't be out until some time after the hd7970. It would hardly be exciting, unless it was priced ridiculously low.
Seriously if the highest end Kepler part only has gtx590 performance, it will be a fail in my book.
It would barely be faster than the hd7970, a larger chip, and wouldn't be out until some time after the hd7970. It would hardly be exciting, unless it was priced ridiculously low.
I would hardly call outperforming your competition by 20-30% and nearly doubling the performance of your previous generations single GPU a failure.
Then there is the matter of TSMC ~1.6x (optimistic) gain.
This is from a TSMC spokes person, giveing a "optimistic" outlook on gains.
TSMC’s senior vice president of R&D, Shang-yi Chiang
“There are misunderstandings that TSMC introduced RDR mainly because of the gate-last [structure]. I’d like to share with you the reason we use RDR had little to do with the choice of gate-last. The [reason] is more for the performance and for the control which will result in [better] yield and cost.”
There is a penalty to introducing this kind of restriction: gate density plummets. Chiang said a 70 per cent linear shrink generally leads to a two-fold improvement in areal density from process generation to generation. Factor in the restricted design rules, and the areal improvement is just 1.6x. “So there’s a 40 per cent layout penalty and that’s very, very significant. However, this can be recovered if we adopt a different layout style, and it’s not difficult to do,” Chiang claimed.
It’s not clear what this style looks like and how it affects standard-cell design in particular. But Chiang added: “TSMC has demonstrated that it will be very willing to share with our customers on how we did that and I’m sure there is more than one way to do that.”
520mm^2 on 40nm -----> 520/1.6 = 325mm^2 on 28nm.
Since a 520, has 512cores, a 1024 core, would likely be twice as big.
Okay.... 325mm^2 doubled up = 650mm^2.
Then you add in new features, that might take up more space ect..... yeah...
1024 core fermi for 680 is definately out of the picture.
That said Nvidia doesnt need that big a increase to beat the 7970.
Im guessing 768 cores or so..... which should put it around 20-25% faster than the 7970.
I agree, that design goals have to target /compelling reasons to upgrade current Fermi workstation cards.
..............below is a copy and paste................
Nvidia Kepler and Maxwell will Focus Mainly on Performance per Watt
Nvidia Kepler to arrive by 2012
Nvidia in the event of GPU Technology Conference (GTC) 2011, has revealed some details on upcoming GPUs based on Kepler and Maxwell. It is expected that Kepler will be released sometime in 2012, and will be manufactured on a 28nm process.
Nvidias Maxwell to Offer 16 Times Higher Performance than Fermi
Nvidia claimed that, Fermi architecture would achieve typical double precision (DP) performance of 1.5GFLOPS per watt. Also in Kepler architecture the performance per watt is increased to about 3 or 4 times, hence, we can expect 1.125TFLOPS 1.50TFLOPS of DP performance by a chip with 250W thermal design power.
Whats new with Nvidia Kepler?
Last month, Nvidias CEO, Jen-Hsum Huang, revealed that Kepler will have lower power consumption than current GPUs based on Fermi architecture and is also expected to provide high energy efficiency. Kepler has promised to improve the capability for GPU computing and enhance programming features that simply double the number of stream processors. Kepler, ultimately, is expected to have a slightly different design from the current Fermi architecture in order to meet all their objectives.
Future of Nvidia Kepler and Maxwell
The specifications and performance of Kepler is not yet revealed yet, but it is expected to have features similar to Fermi. Echelon will be its future chip with 1024 cores and 10 teraflops power which will mainly focus on high performance computing. Also it will run on 128 mini-Streaming Multiprocessors (SM) each consisting of 8 cores
http://lenzfire.com/2011/12/nvidia-...l-focus-mainly-on-performance-per-watt-81669/
notty22, great slides there. I've never seen those before. I am seeing a lot of emphasis on Performance/Watt and optimization for lower power consumption & efficiency. Perhaps NV doesn't want a 300W monster card after all.
The 1.125TFLOPS – 1.50TFLOPS of DP performance by a chip with 250W thermal design power sounds impressive for GPGPU compute applications, but consumer GPUs will for sure be capped at 1/4-1/8th of their SP. That's how NV separates their Tesla line from consumer GeForce line.
Also, we know that Fermi was a monster for GPGPU compute tasks but compared to even Cypress (HD5870), its dual precision performance was very weak:
Techreport
GTX480 Single Precision = 1.35 Tflops
GTX480 Double Precision = 168 Gflops
vs.
HD5870 Single Precision = 2.72 Tflops
HD5870 Double Precision = 544 Gflops
I think if we just start comparing dual-precision performance, that by itself won't tell us a lot about real world improvements in GPGPU computation. Afterall, Fermi was superior for GPGPU tasks vs. HD5870 with only 30% of AMDs DP performance!
one thing i did read about Kepler, is that nVidia will stop listing 3 different clock speeds (Core, Shaders, Memory, when Shaders = Core X 2) and will tie down the Shader clockspeed to the Core clock and only list 2 speeds.
it's been on a major hardware site i'll find the link.
(source pending)
I reckon Kepler will demolish 7970 performance this time. Nvidia have far more money and more time than last to deliver something serious. Also the fallout from the 480 cycle should have settled ( it was a bit hot and not vastly superior to the 5870). I reckon 50% more performance than 7970 at 2560/1600.
Yeah, the 50% more performance sounds about right... but everything points to it being even more late than Fermi. If Nvidia expected Kepler to come out ~6 months after SI, they wouldn't shrink Fermi and would just ride it out on 40nm - a mature process with cheaper wafers.
Both AMD and Nvidia "rebagded" their mobile parts, to the 7xxx and 6xx card names.Kepler in some form or another is coming out this quarter, otherwise I do not think Apple would have signed Nvidia on to deliver GPU's to their various 2012 mac models.
ARE you spamming?!?!wow havent seen spy vs spy in a long time
[redacted]
Anybody know if Kepler can do at least 5760x1080 without the need for an adapter? Or if it does require one, what kind? I'm guessing mDP?
Nvidia hasn't released any 28nm Fermi products - not on notebooks and not on desktops. The only rumors that existed regarding 28nm Fermi was for notebooks, and that rumor pointed to a December launch, but that didn't happen. Shrinking an existing GF114, GF116, or GF118 GPU would likely have been a much easier task, and quicker to market option, than AMD engineering and releasing it's new Tahiti chip. I think it's safe to say that rumor was probably BS.
Kepler in some form or another is coming out this quarter, otherwise I do not think Apple would have signed Nvidia on to deliver GPU's to their various 2012 mac models.
With Apple moving ever away from desktop/workstation units, high end Kepler is going to be largely irrelevant to them, their decision to move to nVidia will likely have nothing to do with GK100. So unless we see die shrunk Fermi re-badged as Kepler, it really won't mean much.
HA HA! I knew I made the right choice!