http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units#GeForce_600_Series

I don't think that the table is correct. Flops are too high.

gdf

Far too high; unless the 256 cuda cores of a GTX650 for $179 really will outperform the 1024 cuda cores of a one year old GTX590 ($669). No chance; that would kill their existing market, and you know how NVidia likes to use the letter S.

My calculated guess is that a GTX680 will have a GFlops peak of around 3000 to 3200 - just over twice that of a GTX580, assuming most of the rest of the info is reasonably accurate.

When it comes to crunching, a doubling of the 500 generation performance would be a reasonable expectation, but 4.8 times seems too high.

I don't see how XDR2 would in itself double performance, and I doubt that architectural enhancements will squeeze out massive performance gains given that it's dropped in size from 520 to 334mm.sq.; transistor count will apparently remain the same.
Perhaps for some enhanced application that fully uses the performance of XDR2 you might see such silly numbers, but for crunching I wouldn't expect anything more than a 2.0 to 2.5 times increase in performance (generation on generation).
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Wow, that looks totally stupid! Looks like a boy's christmas wish list. Well, before every new GPU generation you'll find a rumor for practically every posible (and impossible) configuration floating around..

- XDR2 memory: seems like rumor guys have fallen in love with it. It's not on GCN, though. And if I were nVidia I'd consider it too risky to transition the entire new product lineup at once. There'd need to be serious production capacity ready by now.

- Traditionally nVidia rather goes with wider memory busses than higher clocks, which matches well with their huge chips. I don't see any reason for this to change.

- The core clocks are much higher than even on AMDs HD7970 (925 MHz on pre-release slides). Traditionally nVidias core clocks have been lower and I see no reason why this should change now.

- The shader clocks are totally through the roof. They hit 2.1 GHz on heavily OC'ed G92s, but the stock clocks have been hoovering around 1.5 GHz for a long time. Going anywhere higher hurts power efficiency.

- They introduced a fixed factor of 2 between base and shader clock with Fermi. Why, if they'd change it again with Kepler? I'd expect this to stay, for some time at least.

- 3.0 billion transistors for the flag ship would actually be lower than GF100 and GF110 at ~3.2 billion. At the same time the shader count is said to increase to 640. And the shaders support more advanced features (i.e. must become bigger). Unless Fermi was a totally inefficient chip (I'm talking about the design, not the GF100 chip!), I don't expect this to be possible.

- Just 190 W TDP for their flagship? They've designed power constrained monster chips since some time. If these specs were true, rest assured that Kepler would have gotten a lot more shaders.

- The proposed die size of 334 mm² actually looks reasonable for a 3.0 billion transsitor chip at 28 nm.

- The astronomic FLOPS are a direct result of the insane clocks speeds. Not going to happen.

Overall the proposed data looks more like a traditional ATI "mean & lean" design than a nVidia design.

They may be able to push clocks speeds way higher if they used even more hand crafted logic rather than synthesized one (like in a CPU). Count me in for a pleasant surprise if they actually pulled that off (it requires tons and megatons of work).

MrS
____________
Scanning for our furry friends since Jan 2002

Rumors say there's a too large bug in the Kepler A1 stepping in the PCIe 3 part, which means introduction will have to wait another stepping -> maybe April.

MrS
____________
Scanning for our furry friends since Jan 2002

Anther article with a table of cards/chip types:

here

Its a bit blurry. I can't claim credit for finding this one it was posted by one of the guys over at Seti. Interesting spec sheets though.
____________
BOINC blog

That's the same what's being posted here. Looks credible to me fore sure. Soft evolution of the current design, no more CC 2.1 style super sacalar shaders (all 32 shaders per SM). Even the expected performance compared to AMD fits.

However, in the comments people seems very sure that there's no "hot shader clock" in Kepler. That's strange and would represent a decisive redesign. I'd go as far to say: nVidia needs the "2 x performance per shader" from the hot clock. If they removed this they'd either have to increase the whole chip clock (unlikely) or perform a serious redesign of the shaders, make them more power efficient (easy at lower clocks) and either greatly improve their performance (not easy) or make them much smaller (this was not being done here, according to these specs).

So overall.. let's wait for April then :D

MrS
____________
Scanning for our furry friends since Jan 2002

GK110 release in Q3 2012.. painful for nVidia, but quitre possible given they w´don't want to repeat Fermi and it's a huge chips. Which needs another stepping before final tests can be made (some other news 1 or 2 weeks ago).

And the other article: very interesting read. If Charlie is right (and he has been right in the past) Kepler is indeed a dramatic departure from the current designs.

MrS
____________
Scanning for our furry friends since Jan 2002

No more CC 2.1-like issues will mean choosing a GF600 NVidia GPU to contribute towards GPUGrid will be easier; basically down to what you can afford.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

They are probably going to be CC 3.0. What ever that will mean ;)

MrS
____________
Scanning for our furry friends since Jan 2002

I'm concerned about the 256-bit memory path and comments such as "The net result is that shader utilization is likely to fall dramatically". Suggestions are that unless your app uses physics 'for Kepler' performances will be poor, but if they do use physics 'for Kepler' performances will be good. Of course only games sponsored by Nvidia will be physically enhanced 'for Kepler', not research apps.

With NVidia (and AMD) going out of their way to have patches coded for games that tend to be used for benchmarking, the Internets Dubious Information On Technology will be even more salty and pinched. So wait for a Kepler app and then see the performances before buying.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

If the added speed is due to some new instructions, then we might be able to take advantage of them or not. We have no idea. Memory bandwidth should not be a big problem.

gdf

GK104 is supposed to be a small chip. With 256 bit bandwidth you can easily get HD6970 performance in games without running into limitations. Try to push performance considerably higher and your shaders will run dry. That's what Charlie suggests.

This is totally unrelated to GP-GPU performance: just take a look at how little bandwidth MW requires. It "depends on the code".. as always ;)

And, as GDF said, if nVidia made the shaders more flexible (they probably did) and more efficient for game physics, this could easily benefit real physics (the equations will be different, the general scheme rather similar).

MrS
____________
Scanning for our furry friends since Jan 2002

Some interesting info:
http://wccftech.com/alleged-nvidia-kepler-gk104-specs-exposed-gpu-feature-1536-cuda-cores-hotclocks-variants/

http://www.brightsideofnews.com/news/2012/2/10/real-nvidia-kepler2c-gk1042c-geforce-gtx-670680-specs-leak-out.aspx

If this is real, it seems that kepler multiprocessors are doubled GF104 MP. I hope that it works better than GF104 for compute.

gdf

If you can only use 32 of the 48 cuda cores on 104 then you could be looking at 32 from 96 with Kepler, which woud make them no better than existing hardware. Obviously they might have made changes that allow for easier access, so we don't know that will be the case, but the ~2.4 times performance over 104 should be read as 'maximum' performance, as in 'up to'. My impression is that Kepler will generally be OK cards with some exceptional performances here and there, where physics can be used to enhance performance. I think you will have some development to do before you get much out of the card, but hey that's what you do!
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

no, it should be at least 64/96, but still i hope they have improved the scheduling.
Anyway, with such changes there will be time for optimizations.
gdf

32 from 96 would mean going 3-way superscalar. They may be green, but they're not mad ;)
As GDF said, 64 of 96 would retain the current 1.5-way superscalar ratio. And seeing how this did OK, but not terribly good I'd also say they rather increase the the number of wave fronts in flight than this ratio. I wouldn't be surprised if they processes each of the 32 threads/warps/pixels/whatever in a wave front in one clock, rather than 2 times 16 in 2 clocks.

And don't forget that shader clock speeds are down, so don't expect a linear speed increase with shader number. Anyway, it's getting interesting!

MrS
____________
Scanning for our furry friends since Jan 2002

I wouldn't be surprised if they processes each of the 32 threads/warps/pixels/whatever in a wave front in one clock, rather than 2 times 16 in 2 clocks.
MrS

They've got the basic parameters of the HD7970 totally wrong, although it's been officially introduced 2 months ago. Performance is also wrong: it should be ~30% faster than HD6970 in games, but they're saying 10%. They could argue that their benchmark is not what you'd typically get in games.. but then what else is it?

I'm not going to trust their data on unreleased hardware ;)

MrS
____________
Scanning for our furry friends since Jan 2002

It seems that we are close
http://semiaccurate.com/2012/03/05/nvidia-will-launch-gk104keplergtx680-in-a-week/

gdf

Better pictures, benchmarks and specifications.
http://www.tomshardware.com/news/Nvidia-Kepler-GeForce-GTX680-gpu,15012.html

It should be out 23th March, but by the time it gets to Barcelona is going to be May or June.
If somebody cand give one to the project we can start porting the code earlier. This seems to be an even bigger change than Fermi cards were.

I had the same sort of "bad feeling" - these cuda cores are not what they use to be, and the route to using them is different. Some things could be much faster if PhysX can be used, but if not who knows.
http://www.tomshardware.com/news/hpc-tesla-nvidia-GPU-compute,15001.html Might be worth a look.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Well at the very least they are seem to be like the fermi gpus with 48 cores per multiprocessor which we know that have a comparative poor performance.

I hope that they figured it out, otherwise without code changes it might well be on par with a gtx580.

There's a price going to be paid for increasing the shader count by a factor of 3 while even lowering TDP. 28 nm alone is by far not enough for this.

Seems like Kepler is more in line with AMDs vision: provide plenty of raw horse power and make "OK to use", but not as bad as with VLIW, and not as easy as previously. Could be the two teams are converging to rather similar architectures with Kepler and GCN. The devil's just in the details and software.
(I haven't seen anything but rumors on Kepler, though)

MrS
____________
Scanning for our furry friends since Jan 2002

Suggested price is $549, and suggested 'paper' launch date is 22nd March.

With the 1536 shaders being thinner than before, similar to AMD's approach, getting more work from the GPU and reaching the shaders might be the challenge this time.

The proposed ~195W TDP sits nicely between an HD 7950 and 7970, and noticeably lower than the 244W of the GTX580 (25% higher), so even if it can just match a GTX580 the energy savings are not insignificant. The price however is a bit daunting and until a working app is developed (which might take some time) we will have no idea of performances compared to the GTX500's.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

That's because Kepler fundamentally changes the shader design. How is not exactly clear yet.
@Retvari: and that's why comparisons to GTX560 are not relevant here. I'm saying it's going to be great, just that it'll be very different.

BTW: in the past nVidia chips got rather close to their TDP in "typical" loads, i.e. games. There an HD7970 hovers around the 200 W mark. 250 W is just the power tune limit.

Edit: further information for the brave.. original is chineese.

MrS
____________
Scanning for our furry friends since Jan 2002

They cannot afford to separate gaming and computing. The chips will still need to be the same for economy of scale and there is a higher and higher interest in computing within games.

Changes are good, after all there are more shaders, we just have to learn how to use them. As it is the flagship product we are prepared to invest a lot on it.

gdf

Well.. even if they perform "only" like a Fermi CC 2.0 with 1024 or even 768 Shaders: that would still be great, considering they accomplish it with just 3.5 billion transistors instead of 3.2 billion for 512 CC 2.0 shaders. That's significant progress anyway.

MrS
____________
Scanning for our furry friends since Jan 2002

Otherwise people will be disappointed the next you "only" make a medium step.. ;)

MrS
____________
Scanning for our furry friends since Jan 2002

I wouldn't expect things to work straight out of the box this time. I concur with Zoltan on the potential accessibility issue, or worsening of. I'm also concerned about potential loss of cuda core function; what did NVidia strip out of the shaders? Then there is a much speculated reliance of PhysX and potential movement onto the GPU of some functionality. So, looks like app development might keep Gianni away from mischief for some time :)
The memory bandwidth has not increased from the GTX580, leaving space for a GTX700 perhaps, and there is no mention of OpenCL 1.2, or DirectX 11.1 that I can see of. In many respects NVidia and AMD have either swapped positions or equilibration this time (TDP, die size, transistor count). Perhaps NVidia will revert to type in a future incarnation.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

The problem with CC 2.1 cards should have been the superscalar arrangement. It was nicely written down by Anandtech here. In short: one SM in CC 2.0 cards works on 2 warps in parallel. Each of these can issue on instruction per cycle for 16 "threads"/pixels/values. With CC 2.1 the design changed: there are still 2 warps with 16 threads each, but both can issue 2 instruction per clock if the next instruction is not dependent on the result of the current one.

Load/Store units could also be an issue, but I think this is much more severe.

MrS
____________
Scanning for our furry friends since Jan 2002

Glad to hear it was the right information for you :)

I think there's more going on. Note that in CC 2.1 they had 3 blocks of 16 shaders, which are arranged in 6 columns with 8 shaders each in the diagram. In the GK104 diagram, however, there are columns of 16 shaders. If these were still blocks of 16 shaders, there would be 12 of the blocks, which in turn would require 12 dispatch units - much more than available.

This wouldn't make sense. What I suppose they did instead is to arrange the shaders in blocks of 32, so that all threads within a warp can be scheduled at once (instead of taking 2 consecutive clocks). In this case there'd be "only" 6 of these blocks to distribute among 4 warps with 8 dispatch units.

Worst case we should stil see 2/3 of the shaders not utilized. However, there are 4 warps instead of 2 now. Still (as in CC 2.1) every 2nd warp needs to provide some instruction suitable for parallel execution, but load balancing should be improved.

And there's still the chance they increased the "out of order window", which is the amount of instructions that the hardware can look ahead to find instructions suitable for superscalar execution. As far as I understand this had only been the next instruction in CC 2.1.

I too suppose it's not going to be a DP monster - and it doesn't have to be as a mainly consumer / graphics oriented card. Leave that for GK100/GK110 (whatever the flag ship will be).

MrS
____________
Scanning for our furry friends since Jan 2002

Release date is supposed to be today!
I expect Europe has to wait for the US to wake up, before the Official reviews start. Until then tweaktown's unofficial review might be worth a look, but no CUDA testing, just games.

There is an NVidia Video here.
The card introduces GPU Boost (Dynamic Clock Speed), and 'fur' fans will be pleased!

LegitReviews posted suggested GK110 details, including release date.

____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

it appears that they are actually available for real at least in the UK.
So it is not a paper launch.
gdf

nvidia-geforce-gtx-680-review by Ryan Smith of AnandTech.

Compute performance certainly isn't great and FP64 is terrible (1/24)!

They can be purchased online from around £400 to £440 in the UK, though the only ones I can see in stock are £439.99! Some are 'on order'. So yeah, real launch, but somewhat limited and expensive stock. Also, they are the same price as an HD 7970. While AMD launched both the HD 7970 and HD 7950, NVidia had but one, as yet... This is different from the GTX480/GTX470 and the GTX580/GTX570 launches.
We will have to wait and see how they perform when GPUGrid get's hold of one, but my expectations are not high.

Other Reviews:

Tom’s Hardware

Guru 3D

TechSpot

HardOCP

Hardware Heaven

Hardware Canucks

TechPowerUp

Legit Reviews

LAN OC

Xbit Labs

TweakTown

Phoronix

Tbreak

Hot Hardware
Link Ref, http://news.techeye.net/hardware/nvidia-gtx-680-retakes-performance-crown-barely#ixzz1psRj9zuD

____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

It's build for gaming, and that's what it does best. We'll have to wait a few more months for their new compute monster (GK110).

MrS
____________
Scanning for our furry friends since Jan 2002

So far we have no idea of how the performance will be here.
I don't expect anything super at start (gtx580 like performance), but we are willing to spend time optimizing for it.

gdf

If somebody can run here on a gtx680, let us know.

thanks,
gdf

Almost nothing, this should not matter.
gdf

http://www.tomshardware.com/reviews/geforce-gtx-680-review-benchmark,3161-14.html

Moreover, Nvidia limits 64-bit double-precision math to 1/24 of single-precision, protecting its more compute-oriented cards from being displaced by purpose-built gamer boards. The result is that GeForce GTX 680 underperforms GeForce GTX 590, 580 and to a much direr degree, the three competing boards from AMD.

Does GPUGRID use 64-bit double-precision math?

It would be nice if you also reported here what you find for other projects - thanks!

MrS
____________
Scanning for our furry friends since Jan 2002

We have a small one, good enough for testing. The code works on Windows with some bugs. We are assessing the performance.

gdf

It would seem NVidia have stopped support for XP; there are no XP drivers for the GTX 680!
http://www.geforce.com/drivers/results/42929 I think I posted about this a few months ago.

Suggestions are that the 301.1 driver is needed (probably Win).
http://www.geforce.com/drivers/beta-legacy
A Linux 295.33 driver was also released on the 22dn, and NVidia's driver support for Linux is >>better than AMD's.

The cards fan profile is such that the fans don't make much noise; so it might get hot. This isn't a good situation. If we can't use with WinXP, then we are looking at W7 (and presumably an 11% or more hit in performance)? If we use Linux we could be faced with cooling issues.

The 301.1 driver might work on a 2008R2 server, but probably not on earlier servers.

Good luck,
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

The GPUGRID application doesn't support the GTX680 yet. We'll have test units soon and - if there are no problems - we'll update over the weekend or early next week.

MJH

Good idea; no point returning lots of failed tasks!

I expect you will see an announcement when there is a working/beta app.

Thanks,
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Well, if we make the rather speculative presumption that a GF680 would work with Coolbits straight out of the box, then yes we can cool a card on Linux, but AFAIK it only works for one GPU and not for overclocking/downclocking. I think Coolbits was more useful in the distant past, but perhaps it will still work for GF600's.
Anyway, when the manufacturer variants appear, with better default cooling profiles, GPU temps won't be something to worry about on any OS.
Cheers for the tip/recap, it's been ~1year since I put it in an FAQ.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Thanks Mowskwoz, we have taken this thread a bit off target, so I might move our fan control on linux posts to a linux thread later. I will look into NVidia CoolBits again.

I see Zotac intend to release a GTX 680 chip clocked at 2GHz!
An EVGA card has already OC'ed to 1.8GHz, so the markets should see some sweet bespoke GTX680's in the future.

So much for PCIE 3.0

I see NVidia are listing a GT 620 in their drivers section...
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

For the 1.8 GHz LN2 was neccessary. That's extreme and usually yields clock speds ~25% higher than achievable with water cooling. Reportedly the voltage was only 1.2 V, which sounds unbelievable.

2 GHz is a far stretch from this. I doubt it's possible even with triple stage phase change cooling (by far not as cold as LN2, but sustainable). And the article says "probably only for the chinese market". Hello? If you go all the way to produce such a monster you'll want to sell them on Ebay, worldwide. You'd earn thousands of bucks a piece.

And something like "poor OpenCL performance" can not be said. It all depends on the software you're running. And mind you, Kepler offloads some scheduling work to the compiler rather than doing it in hardware. This will take some time to mature.

Anyway, as others have said, double precision performance is downright ugly. Don't buy these for Milkyway.

MrS
____________
Scanning for our furry friends since Jan 2002

Hey where's that from? Is there more of the good stuff? Did Anandtech update their launch article?

MrS
____________
Scanning for our furry friends since Jan 2002

Yes, looks like Ryan added some more info to the article. He tends to do this - it's good reporting, makes their reviews worth revisiting.
http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/17

Any app requiring doubles is likely to struggle, as seen with PG's.

Gianni said that the GTX 680 is as fast as a GTX580 on a CUDA 4.2 app here.
When released the new CUDA4.2 app is also supposed to be 15% faster for Fermi cards, which is more important at this stage.
The app is still designed for Fermi, but can't be redesigned for the GTX680 until the dev tools are less buggy.
In the long run it's likely that there will be several app improvement steps for the GF600.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help