[Semiaccurate] GK104/Kepler/GTX680 Next Week?

[BallaTheFeared]

From wccf...

The specifications were acquired from a very reliable source which suggests that the GK104 would hold a total of 1536 Stream Processors, Core clock would be set at 705Mhz and Processor clock at 1411Mhz. Memory would consist of 2GB GDDR5 (256-bit) buffer clocked at 3004MHz (6.0Ghz Effective Dual Data Rate).

http://wccftech.com/nvidia-kepler-gk104-gtx-680-final-specifications-leaked/


No hot clocks, but 1536 stream processors at 1411MHz... Dat crazy if they scale the same as Fermi per core/per clock.

No word on ROPs, TMUs, or Poly Engines however... You know the good stuff.

 

[boxleitnerb]

^ 1411=705x2 -> hot clocks
But this is not true I guess. All the rumors have been pointing at no hot clocks and hot clocks are bad for improving transistor density which Nvidia was anxious to improve so they can make their chips smaller.

 

[BallaTheFeared]

1411 is one off though, lol it threw me... I dunno ;(

No way they're getting 1500 hot clocked fermi type shaders in that small of a die.

 

[boxleitnerb]

Rounding error probably.
And I agree.

 

[BenSkywalker]

I think the thing that people forget is that as GPU's become more and more advanced, it's harder to improve them. For example, everyone remembers how awesome the 8800 GTX was at release, and I still see it referenced now, but what people seem to forget is that it was the first unified shader architecture. This removed a tremendous bottleneck from previous architectures and drastically increased GPU efficiency. Now with that bottleneck removed, it's also one less way to make next year's model that much better. Each generation the returns will get smaller and smaller until something radical changes the currents state of things.

Something with a far larger impact happened this generation then with unified shaders. We skipped a half node. Double the transistor density/mm. Should be the largest single generation leap we have seen since the start of the GPU era. We'll have to see if anyone is capable of doing those facts justice.

 

[Crap Daddy]

Hot clocks again? I doubt it.

 

[antihelten]

Something with a far larger impact happened this generation then with unified shaders. We skipped a half node. Double the transistor density/mm. Should be the largest single generation leap we have seen since the start of the GPU era. We'll have to see if anyone is capable of doing those facts justice.

4890 -> 5870 and 285 -> 480 were full node jumps as well (55nm to 40nm)

 

[BallaTheFeared]

Which saw a doubling in performance for Nvidia.

 

[antihelten]

Which saw a doubling in performance for Nvidia.

Not really. At launch techpowerup.com found an average increase of about 40% and computerbase.de of about 65%, when comparing the 480 to the 285.

Although subsequent driver improvements and newer games could certainly have brought closer to 100% since then.

 

[Arzachel]

Something with a far larger impact happened this generation then with unified shaders. We skipped a half node. Double the transistor density/mm. Should be the largest single generation leap we have seen since the start of the GPU era. We'll have to see if anyone is capable of doing those facts justice.

AMD and Nvidia both have hit the TDP wall, if you look at the perf/w, the 7XXX series are a generational leap. It just didn't come with a 25%+ TDP increase like we're used to.

 

[railven]

Which saw a doubling in performance for Nvidia.

Are you saying that now, after driver refinements, because I don't recall that being true at launch. Why it got panned a lot when considering it increased every other measurable metric considerably.

 

[BallaTheFeared]

http://www.tomshardware.com/reviews/geforce-drivers-geforce-gtx-480,2875.html

I guess


Should I do a new one, 197.41 vs 295.73 in SLI stock vs OCed?

 

[Keysplayr]

4890 -> 5870 and 285 -> 480 were full node jumps as well (55nm to 40nm)

I thought that 65nm to 40nm would have been a full node jump. 55nm was a half node IIRC.

And 40nm was just 45nm renamed to 40nm. Idontcare, could you chime in here? My memory is a little foggy on this point.

Full nodes: 110nm, 90nm, 65nm, 45nm(40nm), 28nm
Half nodes: 80nm, 55nm, 22nm??

Unsure about 22nm.

 

[notty22]

http://www.tomshardware.com/reviews/geforce-drivers-geforce-gtx-480,2875.html

I guess


Should I do a new one, 197.41 vs 295.73 in SLI stock vs OCed?

Please, no

 

[BallaTheFeared]

Heh

 

[f1sherman]

That dude from OBR is a huge nvidia fanboy and is completely biased in 99% of his articles. He has a hard on for nvidia.

Charlie seems to be the AMD fanboy.

Charlie vs OBR, war of the fanboys?

Charlie is not fanboi. He just hates Jensen's guts and no one knows why

 

[boxleitnerb]

Jensen slept with his wife

 

[Keysplayr]

Charlie is not fanboi. He just hates Jensen's guts and no one knows why

Of course we know why.

 

[f1sherman]

Please tell


oh... wife comment?

 

[Lepton87]

I thought that 65nm to 40nm would have been a full node jump. 55nm was a half node IIRC.

And 40nm was just 45nm renamed to 40nm. Idontcare, could you chime in here? My memory is a little foggy on this point.

Full nodes: 110nm, 90nm, 65nm, 45nm(40nm), 28nm
Half nodes: 80nm, 55nm, 22nm??

Unsure about 22nm.

22nm is a full node jump from 32nm also 28nm is a half-node.

update: 110nm is also half-node, full node is 130nm

 

[96Firebird]

http://en.wikipedia.org/wiki/Die_shrink#Half-shrink

So 55nm --> 40nm was a full node shrink between two stopgap half nodes.

And 40nm --> 28nm is a full node shrink between two stopgap half nodes.

 

[tviceman]

A bit off topic, but what do you think about: http://www.zdnet.com/blog/computers...ionize-the-pc-gaming-industry-and-beyond/7627 Enough to resuscitate PC gaming? Roadkill?

I would support this, but it will still be inherently shackled by Microsoft's OS licensing fee's. The day Valve buys On-Live (or creates it's own similar service) and supports real-time streaming across any pc-like device (desktop, tablet, powerful smart phones, etc.) along side local play for those with good enough PC's and can just make steam it's own OS will be the day that "PC Gaming" will obsolete console gaming and will start becoming the overriding popular way to game.

 

[tviceman]

From wccf...



http://wccftech.com/nvidia-kepler-gk104-gtx-680-final-specifications-leaked/


No hot clocks, but 1536 stream processors at 1411MHz... Dat crazy if they scale the same as Fermi per core/per clock.

No word on ROPs, TMUs, or Poly Engines however... You know the good stuff.

So much conflicting information going around. That is a hotclock though.

 

[Riek]

http://www.tomshardware.com/reviews/geforce-drivers-geforce-gtx-480,2875.html

Should I do a new one, 197.41 vs 295.73 in SLI stock vs OCed?

at the cost of an additional 100W.

 

[antihelten]

Which saw a doubling in performance for Nvidia.

Actually it was quite a bit less than that. the launch reviews from techpowerup* and computerbase.de* showed an average increase over the 285 of 40% and 65% respectively.

although with newer games/drivers the current difference might be fairly close to 100%

*I looked at these 2 sites since they are the ones I know of, who show average results of all the games they tested.