GK110 = Kepler-based Tesla K20 GPU, with up to 2,880 SPs, 7.1 Billion transistors!

[RussianSensation]

And just as I suspected, there will be a larger performance gap between GK104 and GK110 than there was between GF114 and GK110. But then again, with the amount of time that it will have taken to get GK110 out, all of this info is both appropriate and expected. I am guessing GK110, when it makes it's way into a geforce product, will be 60-80% faster than GK104 across the board in graphics performance. I think we will be "lucky" to see it's initial msrp at or below $650.

I removed 3072 CUDA cores from the title until it's confirmed by a secondary source. I can't seem to find official info for now. For that reason I put 50% faster for conservative reasons.

I read the GK110 white paper and it says SMX unit has 192 CUDA cores not 128 as mentioned by the source I used. So I am going to hold off on 3072 CUDA cores for now and put that under Rumor.

 

[coffey]

Hmmm, nobody has dared the "will it run..." joke yet.

 

[kevinsbane]

Where did they mention the CUDA count on the K20?

 

[AdamK47]

Want

 

[RussianSensation]

Where did they mention the CUDA count on the K20?

From original sources I found but I can't find official data to confirm it. For now I put that 3072 # under rumors.

The official GK110 white paper has 192 CUDA / SMX. I'll update as soon as I find more info.

I never insinuated that, you continue to misrepresent my posts, but regardless.

Vindicated!

You said until GK110 is launched GK104 is NV's high-end chip. GK110 isn't going to launch until Q4 at the earliest which means we are both right - right now GK104 is NV's highest end gaming chip, but at the same time it's not a high-end Kepler chip.

That is being way too optimistic. 50% more shaders is not going to mean 50% more performance with the overhead of a compute GPU.

3072 cores is actually 100% more than 1536 in GTX680. I agree with you that we won't see linear performance increase with double the CUDA cores though.

 

[coffey]

Want

I hope you have a huge fat wallet

 

[railven]

You said until GK110 is launched GK104 is NV's high-end chip. GK110 isn't going to launch until Q4 at the earliest which means we are both right - right now GK104 is NV's highest end gaming chip, but at the same time it's not a high-end Kepler chip.

That is not what I said, my scenarios is exactly this just that played out:

http://forums.anandtech.com/showpost.php?p=33115350&postcount=56

This ties in nicely to what I said in another thread. With nVidia's experience in the GPGPU sector, they could easily create a two prong attack.

With PC gaming (and gaming in general) in a slump of innovation, why cater to the lower profits margin when you can milk your top end GPU at the $5,000 price point. EDIT: ANd face it, anything that will be based on the new Unreal Engine won't manifest in our hands until at least Q4 2012 or Q1 2013. By then - Refresh city!

I should dig up my post. I'm definitely buying a lottery ticket this weekend. If I win the lotto, I promise to buy each of you a last generation card - since of course, they are good enough

I can't find our debate through search but found this, which covers what I said.

I should have bought that lottery ticket

EDIT:

And I found this post too which gave me a light chuckle:

http://forums.anandtech.com/showpost.php?p=33115890&postcount=82

Good times...good times...

 

[RussianSensation]

That is not what I said, my scenarios is exactly this just that played out:

http://forums.anandtech.com/showpost.php?p=33115350&postcount=56

Yup, in terms of strategy, you called it 100%. :thumbsup:

P.S. Those emoticons are fun. We could use them on our forum.

 

[BallaTheFeared]

Is there any word at all for a GK110 gamer version, at the very least in the upcoming refresh?

This gen is just ridiculous, $550 mid-range cards. When do the new consoles come out, I can probably buy one for that much.

 

[RussianSensation]

Is there any word at all for a GK110 gamer version, at the very least in the upcoming refresh?

This gen is just ridiculous, $550 mid-range cards. When do the new consoles come out, I can probably buy one for that much.

Don't hold me to it. This is what I read:

1) Wii U will launch in the US on Sunday, 18th November, 2012.
2) Xbox Next Q4 2013/Q1 2014
3) PS4 Q4 2013

Rumored GPU specs for next gen consoles are underwhelming, especially since Sony is losing $ and can't afford to sell PS4 at a loss. PS4 is unlikely to be cutting edge like PS3 was. Also, it's not like Hellspawn GTX470s in SLI are slowing you down

I have no idea when GK110 will launch in the consumer space. If NV can milk GK104 for $400-500, they have no urgent reasons to launch a 7 billion transistor gaming chip for $650 when they can sell it for $3K in professional markets. Until HD8000 launches, NV really has no reason at all to launch a larger die Kepler.

 

[Joseph F]

Hmmm, nobody has dared the "will it run..." joke yet.

Will it run BF3 Ultra @ 7680*1600 with 16x SSAA?

 

[sontin]

I expect that BigK will come in November - like GF110. There is no reason to hold back the chip for the consumer market.

 

[coffey]

Will it run BF3 Ultra @ 7680*1600 with 16x SSAA?

Depends on the fps you can live with

 

[sontin]

nVidia confirmed 15 SMX - 2880 Cores and maybe 240 FP16 TMUs :'(

German:
http://www.heise.de/newsticker/meld...hip-hat-bis-zu-2880-Shader-Kerne-1576464.html

http://www.pcgameshardware.de/aid,8...Us-auf-GTC-2012-vorgestellt/Grafikkarte/News/

 

[Cookie Monster]

Just been reading about whats been presented at GTC, the GK110 or BigK seems like a complete GPGPU monster. I mean just looking at the die shot, you will notice the big fat command processor in the middle.

Here are some observations:
-Die size seems to be roughly ~520mm^2 or around that vicinity.
-It seems to only have 15 SMXs, not 16.
-The SMXs found on the BigK is bigger, am thinking its due to the increased/beefed FP64 units. Everything else seems the same to the Gk104 SMX.
-If GK104 has 4 GPCs, the BigK has 6 which results in 48 ROPs and a 384bit memory interface.
-6 setup engines (um 6 traingles/Clock?? you kidding me? will eat through tessellation benchmarks)

So in short, if they do make a Geforce version of these maybe in early 2013, or fall 2012 (less likely) we are going to see a 2880CC, 48ROPs, 240 TMUs, 384bit card.

Id say a 50% performance increase over the GK104 and probably near double that of the GTX580 isn't too far fetched.

 

[soccerballtux]

hm, I wonder if you could just use this for video games. You wouldn't need to upgrade for a long time.

 

[JAG87]

Just in case anyone didn't notice, GK110 looks to be 225W-300W. 8+6 pin on the board.

 

[Don Karnage]

They are professional cards, it doesn't matter what the PCB looks like when they spend their entire life in a rack, obscured from sight.
Actually, I prefer plain green PCBs to the ridiculous, gaudy colors sometimes used that are associated with gamers, nowadays.

Though, I'm sure you're just kidding.

I'd buy this card even with a hot pink PCB. It would match my hello kitty fans

 

[Cookie Monster]

Just in case anyone didn't notice, GK110 looks to be 225W-300W. 8+6 pin on the board.

Which doesn't necessarily tell us anything e.g GTX480

 

[Cookie Monster]

I'd buy this card even with a hot pink PCB. It would match my hello kitty fans

I wouldn't.. because it doesn't even have any output video ports while costing +$5000 or so.

 

[Tempered81]

jeezuz what a beast.

 

[blackened23]

I'd buy this card even with a hot pink PCB. It would match my hello kitty fans

It has to come with hello kitty case stickers as well.

Stickers or NO DEAL.

 

[tviceman]

That is being way too optimistic. 50% more shaders is not going to mean 50% more performance with the overhead of a compute GPU. GK104 has nothing in the way of nvidia's usual compute arsenal and is aimed squarely at gaming. This card most likely will give the true performance leap over the GTX580 that we are used to from nvidia with a node shift, 60-70%, and make up the slack that the GTX680 left with it's small 30% gain on the GTX580.

3072 cores is actually 100% more than 1536 in GTX680. I agree with you that we won't see linear performance increase with double the CUDA cores though.

I think it's probably more valid to compare the gtx580 and gtx560ti to each other, since both were on the same node and based on the same fundamental architecture. The gtx580 had 33% more cores, 50% more bandwidth, and it's cores operated at a 6.4% slower frequency than the gtx560ti. So with those 33% more cores, it performed 40% faster than the gtx560ti.

I don't think there will be =>100% scaling with cores and performance moving from GK104 to GK110, but I do think it will scale pretty good given that it won't be at all bandwidth constrained.

 

[sandorski]

These are not Video Cards.

 

[tviceman]

These are not Video Cards.

GK110 will be in a video card at some point within the next year.