Kepler Nvidia GeForce GTX 780

[Aten-Ra]

Oh and I count 8x8 = 64 ROPs. :P

I believe i need classes

yeap they are 8x8.

[AffenJack]

The only way is barely divide the cuda cores (SP ) size by 300%, or i dont see how they can do it in 340mm2.

I don't see such a problem there, take GF114 and you can roughly double it in 28nm at the same diesize. Then you'll have 768 Hotclock Shader, 128TMUs, 64 Rops. But it' s only 32 Rops so you save space and Hotclock-Shader are bigger by an unknown margin then normal ones. Fill this free diespace with shaders instead and you get the speculated Gk104.

[UrbanSmooth]

"9800 JTX (Jesus X-Treme)," anyone?

[cold2010]

phk Say next month!!!

http://bbs.expreview.com/thread-49260-1-1.html

[trinibwoy]

Well these guys surely seem the most confident out of the rumors I've read, but I still don't like how these numbers look eerily similar to AMD's design. I guess the simple answer would be that Nvidia decided their superscalar design was too expensive and resulted in poor yields compared to AMD, but even if they did triple the core count, if they removed the hotclock wouldn't that result in just a minimal increase?

Not necessarily. If the 950Mhz for GK104 is true then it has 3*950/1544=80% more flops than the 580. Unless you consider 80% minimal of course

[AliG]

Not necessarily. If the 950Mhz for GK104 is true then it has 3*950/1544=80% more flops than the 580. Unless you consider 80% minimal of course

That's assuming 100% scaling, but shoot if we see even a 50% increase that would spank the 7970 (I believe it's on average only 20% faster than the 580).

[trinibwoy]

That's assuming 100% scaling, but shoot if we see even a 50% increase that would spank the 7970 (I believe it's on average only 20% faster than the 580).

I was just referring to the raw flop comparison after dropping hot clocks. Doesn't tell us anything about actual performance vs the 580 or 7970 though.

[nksharp]

phk Say next month!!!

http://bbs.expreview.com/thread-49260-1-1.html

If thats true I might be able to pick one (or two, or three) up for my build in time for the lan I plan on going to in the spring :o

Hopefully some waterblocks come out soon after they are released.

[AliG]

I was just referring to the raw flop comparison after dropping hot clocks. Doesn't tell us anything about actual performance vs the 580 or 7970 though.

That's true, but while total FLOPS doesn't tell us too much, pixel fillrate tends to be an accurate benchmark. if Kepler architecture is truly more efficient than Fermi, and Fermi did scale quite linearly (almost a 2x increase in pixel fillrate 280 -> 580) then I would imagine we would see fairly linear gains here as well.

Edit: I should mention this is assuming the texel rate is sufficient to not hold back the card. But as far as I can tell, TMUs shouldn't be an issue here

Edit2: I think I'm just starting to confuse myself lol. All these rumors are driving me nuts

[GoldenTiger]

If thats true I might be able to pick one (or two, or three) up for my build in time for the lan I plan on going to in the spring :o

Hopefully some waterblocks come out soon after they are released.

Yeah, grabbing 2 assuming they are good, right away upon release . Would love a March launch.

[DarthShader]

Yeah, rumours from Chinese forums are reliable. March launch is DOUBLE CONFIRMED!

[LordEC911]

That's true, but while total FLOPS doesn't tell us too much, pixel fillrate tends to be an accurate benchmark. if Kepler architecture is truly more efficient than Fermi, and Fermi did scale quite linearly (almost a 2x increase in pixel fillrate 280 -> 580) then I would imagine we would see fairly linear gains here as well.

Edit: I should mention this is assuming the texel rate is sufficient to not hold back the card. But as far as I can tell, TMUs shouldn't be an issue here

Edit2: I think I'm just starting to confuse myself lol. All these rumors are driving me nuts

Bandwidth...

[BHD]

Hi ALL

Lock At Image: Kepler






GK100 63% Faster 3DMark11 > Hd 7970

GTX 680 8% Faster 3DMark > HD 7970

[kaktus1907]

Charts are similar though not from the same guy.. first one fits latest rumors but it doesnt give any credibility to benchmark chart.. that said id take the charts with a healthy dose of salt..

[Olivon]

[IMG]http://wccftech.com/wp-content/uploads/2012/02/220329m6tnt4axtfl8tnif-635x386.png[IMG]

Original link :

http://we.pcinlife.com/thread-1831876-1-1.html

[Lanek]

Charts are similar though not from the same guy.. first one fits latest rumors but it doesnt give any credibility to benchmark chart.. that said id take the charts with a healthy dose of salt..

Looking at the fact this is the first post of this guy in PCinlife.. yes i take a good dose of salt too. some numbers against the 580 look really a bit too high.. 30-40%-48% faster of a GTX580 in games as BF3, CIV5, Dirt3... huum...

Taking in account we are normally speaking about a card with beta drivers, and the "middle range " chip.. hard to compare but their number are higher of a GTX590 in BF3, Dirt3... etc http://www.computerbase.de/artikel/g..._battlefield_3 ... really hard to believe Nvidia had aim the GK104 vs the GTX590

[AliG]

I'm thinking that chart either has a typo or it's fake. Look under the texel rate for GK106, it is more than GK104 despite having half the SP clusters. Either it should have been 37 and not 73 (or something along those lines) or it's fake. If that's a typo, I would say the chart checks out. From what we've heard those numbers do seem realistic

[-=DVS=-]

I'm thinking that chart either has a typo or it's fake. Look under the texel rate for GK106, it is more than GK104 despite having half the SP clusters. Either it should have been 37 and not 73 (or something along those lines) or it's fake. If that's a typo, I would say the chart checks out. From what we've heard those numbers do seem realistic

You get Texture FillRate = TMUs X Core Clock , 1150x64=73600 , both GK106 and GK104 have same number 64 TMUs , but GK104 is clocked higher.

[xBanzai89]

GPUReview updated their site with the GTX 600 series specs. I find this odd given that we do not have any official word from Nvidia on actual specs, since I didn't know that site followed rumors for official specs.

[GoldenTiger]

GPUReview updated their site with the GTX 600 series specs. I find this odd given that we do not have any official word from Nvidia on actual specs, since I didn't know that site followed rumors for official specs.

I saw them once post gtx 4xx specs if I recall correctly, and they turned out to be the official ones at launch. I have no idea if they know but, they aren't typically a rumors site as far as I've ever seen. Very interesting.

Edit.... the site owner made a comment clarifying they are educated guesses.

[bill_d]

I saw them once post gtx 4xx specs if I recall correctly, and they turned out to be the official ones at launch. I have no idea if they know but, they aren't typically a rumors site as far as I've ever seen. Very interesting.

Edit.... the site owner made a comment clarifying they are educated guesses.

not very
gtx 690 Max Power Draw:700 W

[GoldenTiger]

Just appeared on PDXLAN site:

News

*** SPECIAL PDXLAN 19 ALERT!!! ***
Monday, 13 February 2012
NVIDIA and Gearbox are bringing something really special to PDXLAN 19.

Attendees are in for an exclusive treat that will blow your mind!!

Anyone?

[Cold Fussion]

A lathe, and they will turn a GTX 680.

[Vardant]

I'm thinking that chart either has a typo or it's fake.

That chart is really old now and some of the spec are completely wrong.

[Tim]

680 preview with upcoming game from Gearbox which shows off Kepler's new Physics hardware with fluid dynamics and other physics stuff.