Some leaked Nvidia Kepler info:
http://lenzfire.com/2012/02/entire-nvidia-kepler-series-specifications-price-release-date-43823/

Highlights:

Nvidia Kepler GTX690
750MHz Core clock
2×1.75 GB 4.5GHz GDDR5 Memory
2×1024 Stream Processors
2x448bit Bus Width
Priced at $999
Both AMD and Nvidia are following the same trend in maintaining the order of the series in GPUs. Like HD 7990, GTX 690 has a core clock lesser than GTX 680 and Bus width similar to GTX 670. Compared to AMD HD 7990′s price point of $ 849, GTX 690 is priced at $ 999. But comparing to the performance the GTX 690 can offer, HD 7990 is definitely over priced, and sure AMD has to reduce their prices drastically after the release of GTX 600 kepler series.

Nvidia Kepler GTX680
850MHz Core clock
2 GB 5.5GHz GDDR5 Memory
1024 Stream Processors
512bit Bus Width
Priced at $649
45% faster than HD 7970
Nvidia has a clear winner in its hands and that’s why they have commented that they had expected much more from AMD. But still we already know that AMD is preparing its revised GCN HD 8000 series graphic cards, which may be release in the latter half of this year.

Nvidia Kepler GTX670
850MHz Core clock
1.75 GB 5GHz GDDR5 Memory
896 Stream Processors
448bit Bus Width
Priced at $499
20% faster than HD 7970

Nvidia Kepler GTX660Ti
850MHz Core clock
1.5 GB 5GHz GDDR5 Memory
768 Stream Processors
384bit Bus Width
Priced at $399
10% faster than HD 7950

Nvidia Kepler GTX660
900MHz Core clock
2 GB 5.8GHz GDDR5 Memory
512 Stream Processors
256bit Bus Width
Priced at $319
Performance similar to GTX580



Looks like they are one-upping the 7970 on performance but more expensive, and still lagging on VRAM capacity for triple monitor support...lame. No mention of whether SLI is still needed for triple monitor either.

Continuing to watch and wait for the 7970 E6 6GB and hopefully a 680 Zotac with at least 4GB VRAM.

Then again maybe the leaked info is not real? Hopefully something official soon.

Update: some further research and posts on OCN seem to be of the opinion that the above article is fake specs... I sure hope so was hoping for at least 3GB VRAM and triple monitor support on the GTX 680.

Don't know if its fake info or not. To be sure, it could be something just to keep folks from buying AMD "while they wait"

I too have read that Kepler will be faster than comparable HD7xxx cards. However, not always by a lot. Rather, Kepler will show the big advantage on a handful of games optimized for Kepler architecture (particularly games with heavy PhysX), is the rumor.

Hopefully, prices will become more "sane" when Kepler hits the streets.

We'll see in a couple months

BTW, the most interesting rumor (to me) is that Kepler will NOT have "up-clocked" shaders.

For a few generations, Nvidia did better per shader than AMD partly because they ran their shaders twice as fast as the card base speed (faster than AMD shaders). Now the shaders are rumored to run at the card base speed. That alone could "slow down" FPS substantially per shader. We'll see.

I can't believe they're running prices like that. Obviously they feel that people will pay whatever for the next slightly faster card.

At that price I would hesitate to buy a 660Ti, even though traditionally the 670 is what I would go for.

Well, I'll buy when the prices are what I think they should be. My 570 is doing just fine right now. I could probably upgrade to a 580 for cheap once these are out.



The Jedi Master

And no word yet on their "surround gaming" on a single card?

Yeah exactly, all we can hope for is that this is fake specs and pricing as some say...

I was hoping the 680 would have at least 3GB VRAM and triple monitor "surround" on a single card, if Nvidia misses the Eyefinity boat again on this generation of cards, ATI will eat their lunch.

Hopefully NV will put out some official info soon, it's bad enough that they are 3 months behind ATI/AMD on the release but worse even that they don't put out specs so it leaves people open to fake stuff and conjecture.

Should drop the price of the 580 though.

Rumors are rumors..

look at the Bulldozer's Rumored Performance -> Actual Performance....

Seem like the "big" Kepler chip has "just taped out". So, one might expect the top card in late Summer to Fall 2012. Just about when HD8xxx might hit the market (or might not ).

GK110 Tapes Out

BTW, the rumor indicates Nvidia's first try at 28nm failed -- that's why they are late. Heck, AMD had a good, working 28nm design over a year ago and were just waiting for the TSMC 28nm manufacturing process to catch up. Some of those good features were "added" to the 40nm HD69xx cards while they waited (cards take 3 years to design/develop/produce -- so next year's cards have already been in the works for 2 years).

if only AMD can catch up to Intel....

Q3?! That's an awful ways away... AMD will be lining its pockets in the meantime and likely working to have a counter ready for NV's release date.

That article also mentions 300W (!) power consumption, I know, I know, just rumours- but if true that sounds like another hot and noisy NV high-end card. We shall see.

Isn't AMD going the one new GPU, one respin route? I'd expect there to be a 7975 and a 7955 in July, and the 8xxx cards in or around December.

8000 will be "improved" GCN

http://www.tomshardware.com/news/Nvidia-Kepler-GPU-GeForce-600-Series,14642.html#xtor=RSS-998


even the 690 will only have 3.5 GB of Ram. (1.75GBx2).


i bet they took their readings with PhysX set to ON.

Your guess seems very plausible, to me. Later seems more likely than sooner, to me.

Three years back (or so), AMD indicated they wanted to come out with a substantially new card every 12 months -- very importantly, without slips. In between, it was supposed to be minor tweaks, if anything. Now, there is new management. So, they may not be on that update path any longer. Nonetheless, sales wise, Fall/Winter would be a good time for an HD8xxx.

Post-recession I don't know that the market really would take that same breakneck pace. I always thought 18 months for a major update would be fine. The 6 months between major and minor update cycle was just too fast. Besides, as a result of the console mentality meaning games aren't really pushing like they used to, the previous generation winds up being totally capable for a long time.

Unless you're pushing multiple monitors (which I find impractical for space reasons) at very high resolutions, there's nothing a card over a year old at 1920x1080 can't manage today.



The Jedi Master

New rumoured specs:

nVidia GK104

Name (high end): GeForce GTX 680 (not certain yet)
28nm manufacturing process at TSMC, die-size ca. 340mm² (inofficial estimate)
4 Graphics Processing Clusters (GPC)
4 Streaming Multiprocessors (SM) aka Shader-Cluster pro GPC, so in all 16 Shader-Cluster for the GK104-Chip
96 Stream Processors (SP) aka Shader units per Shader-Cluster, so in all 1536 Shader units for the GK104-Chip
8 Texture units (TMU) aka Texture units per Shader-Cluster, so in all 128 Texture units for the GK104-Chip
32 Raster Operation Units (ROPs)
256 Bit DDR memory interface (up to GDDR5)
Chip frequency (high end): 950 MHz
No more Hotclocks, no extra frequency of the shader unit (anymore)
SinglePrecision performance: 2,9 TFlops, DoublePrecision with 1:6 = DP-processing power 486 GFlops, Texture performance 121 GTex/sec
Memory frequency (high end): 2500 MHz, memory bandwidth about 160 GB/sec
2048 MB GDDR5 memory placement



Google translated from German

Original article @ 3DCenter.org

So this rumor has a far faster GPU clock and a slower memory clock. Interesting.

However, I'm not interested in any card over $350, so I'll see what's in that price range come the end of this year/beginning of next when my card is 18-24 months old.



The Jedi Master

those numbers dont add up, something fishy...

Still the rumors have not improved on the 2GB VRAM and I'm guessing that still means it won't have 3 monitor support otherwise it would be higher VRAM amount, lame... I just hope there is some unmentioned higher end (685?) card with 3+GB VRAM and 3 ports (including DP).

For now I got tired of waiting for Kepler and picked up a 560Ti/448 for <$250, these are close to 570 performance but very overclockable (some of the same model I bought to 570/580 speeds ) and I needed the upgrade from my 5870's for FSX and COD and rF2 and am tired of waiting.

The 560Ti/448 came in last night and from initial benchmarking it is a big upgrade to my 5870 for FSX and I have not even tried overclocking it yet.

Should tide me over until at least late this year once Kepler and ATI's mid generation answer is out, though I gave up 3 monitor support for now unless I am so happy with it that I choose to pick up another for SLI.

Still hoping Kepler will have some evolutionary card with triple monitor support and ATI releases a serious overclock response later this year, if so I can sell the 560ti/448 and pick one up at that time, for now I am *very* happy with this bang/buck upgrade of it's little brother that came out in Nov 2011.

doubt it, nVidia seems to think they are better off forcing SLi for 3 screens/surround sound gaming.

CEO even said in an interview "that way they can almost force a gamer to buy not 1 card but two"

Fortunately, there are sometimes manufacturers that don't follow the rules, like Galaxy with their MultiDisplayTechnology (MDT) cards. Those do the single card "surround gaming" trick.

I actually seriously considered their models for 570/580 but in the reviews I read the main thing that kept me from doing it is you are stuck with a single 3 monitor "surround" "eyefinity" resolution, they trick the OS into seeing it as one monitor, but since the OS doesn't see the separate monitors you can't switch to separate "extended desktop" mode when not gaming or in situations where the game performance is not good with 5760x1080p.

So I think it is pretty ingenious that they have been able to do that with NVidia's design, but it is still a bit of a workaround tricking the OS and you lose some of the flexibility of normal Extended monitor mode if you run into games that the video card can't handle at the large 3 monitor resolution.

i think Mini HDMIx4 is better.

The AMD Eyefinity is actually "in the chipset" -- not just the connectors or the driver (so they say). It was designed in "in secret" originally -- to give them a competitive edge with "extreme gamers" and business people who use multi-monitor extensively. Actually did catch folks by surprise -- rarely done these days.

Nvidia 3-monitor is/was solely in the drivers. Basically, it was/is a quick fix -- not a genuine solution. Surprising if the new models don't have a chipset designed for 3-monitors. On the other hand, Nvidia pushes 3D stereo -- which is an afterthought by AMD.

AMD renounced GPU physics (after being the FIRST to experiment with it). They found it was always better to do the physics on the CPU and graphics on the GPU -- for best overall performance. GPU physics is a "marketing feature". Only PhysX does it that way (and they have to cripple the CPU drivers to make it seem better than it really is according to rumor). HAVOK physics doesn't bother -- its CPU only.

Though the first GTX6xx cards may still get here in April/May, they could easily be in short supply.

According to Nvidia's most recent Conference Call, there are issues with their 28nm process products (among other things) -- perhaps yield issues. These issues are rumored to be related to weaknesses in the Nvidia designs -- rather than weaknesses in the TSMC 28nm process (which had problems in the past that were fixed -- it is being used successfully by AMD 28nm HD7xxx products).

Linked is a long speculative article that, based on the conference call, guesses what types of issues might be affecting Nvidia product developments (of several types). Not a great read for a games enthusiast -- its more for investors. I link it simply because its usually good practice to link.

Nvidia Conference Call

Wow, that reads like an nvidia horror story. Not only the video card, but the mobile/tablet side as well.

Seems to me if the next gen launches on time it will have very limited supply and perhaps a notable failure rate as well with problems for those needing RMAs.



The Jedi Master

so the limited supplies everyone does get is gonna have outrageous mark ups,.

even if it does beat AMDs cards, you can win if you cant put supply on the market. and you can put supply on the market if out of 100 Chips only 30 make it through to the bin process, and out of those 30, only 15 of them are prolly binned high enough for use.

Sounds identical to the situation 2.5 years ago:
1. ATI comes out with the 5870 which is a great card and surprises
2. Nv struggles to get the 470/480 out, it's late, limited initial supplies, it runs hot and uses lots of power
3. Eventually Nv comes out with the 570/580 which are the good cards they meant to come out with in the first place

Seems like the potential for a similar pattern, #1 is true, and the conference call sounds like the situation before #2 happened.

Glad I can wait this out for a while now that I found a great/cheap temporary card...

scrap the 600 series and move on to the 700 series..

I thought the 700 series IS the 600 series-that-should-have-been? Or did they reverse that decision to try and bring the next one in sooner, only to fall on their face?



The Jedi Master