[Videocardz/3DCenter] GK110 rumors/specs

[badb0y]

AMD is probably releasing new cards around December/Janurary time do its not crazy to expect GK100/110 around that time.

 

[-Slacker-]

Sorry for the off topic, but if it's confirmed that the GTX680's die is <300mm^2, then I can't see the HD7970, which has 70mm^2 on it, being slower on average, as the newest rumored benchmarks suggest. I mean sure, it's possible, but I wouldn't put much faith into it.


**To try and avoid a flame war, I will not reply to any response to this post**

 

[tviceman]

Sorry for the off topic, but if it's confirmed that the GTX680's die is <300mm^2, then I can't see the HD7970, which has 70mm^2 on it, being slower on average, as the newest rumored benchmarks suggest. I mean sure, it's possible, but I wouldn't put much faith into it.


**To try and avoid a flame war, I will not reply to any response to this post**

I think it's already a foregone conclusion, though. It's only a matter of how much faster / more efficient GK104 is. Good, reliable reviews are coming in less than 48 hours. We will all know then, but I'm sure the argument won't stop there. The one glimmer of bright light in all of the "go premium" banter is that the MSRP for gtx680 is now rumored to be $499. So it's faster and cheaper than hd7970, it'll spark price cuts across AMD's entire lineup and will finally push 28nm cards into a noticeably better price/performance category than 40nm cards.

 

[MrTeal]

I'm going to go out on a limb here, and predict that even at 550mm^2, GK110 won't be more than 30% faster than the 300mm^2 GK104 in gaming performance. Looking at the AT review, NVIDIA has done an amazing job of stripping Kepler down to the bones to get the best possible gaming experience out of it while sipping power. For compute though it is lacking, outside some scenarios where speed is solely dependent on shaders and clock. When GK110 is released I expect that all or at least the majority of the CUDA cores will be FP64 capable, and this is going to eat up die space without giving a huge boost to gaming performance.

My guess is that GK110 is going to be an absolute compute beast, but if you want gaming performance don't wait the 6 months, buy the 680 now and buy another in the fall.

 

[Subyman]

With the scientific community eating up gamer oriented cards for relatively cheap prices, I can see why Nvidia wants to offer a very high priced compute card. National Labs, universities, etc will pay 2K+ all day long for hardware where as the gamer is very price sensitive. I wouldn't be surprised if the GK110 is aimed at the compute crowd and Nvidia separates the lines.

 

[Dribble]

I'm going to go out on a limb here, and predict that even at 550mm^2, GK110 won't be more than 30% faster than the 300mm^2 GK104 in gaming performance. Looking at the AT review, NVIDIA has done an amazing job of stripping Kepler down to the bones to get the best possible gaming experience out of it while sipping power. For compute though it is lacking, outside some scenarios where speed is solely dependent on shaders and clock. When GK110 is released I expect that all or at least the majority of the CUDA cores will be FP64 capable, and this is going to eat up die space without giving a huge boost to gaming performance.

My guess is that GK110 is going to be an absolute compute beast, but if you want gaming performance don't wait the 6 months, buy the 680 now and buy another in the fall.

From what we have seen in the reviews so far the 680 is memory limited. I agree 50% more cores clocked lower to keep thermals reasonable doesn't make 50% faster, but a memory bus at least 50% wider might have a large impact.

 

[MrTeal]

From what we have seen in the reviews so far the 680 is memory limited. I agree 50% more cores clocked lower to keep thermals reasonable doesn't make 50% faster, but a memory bus at least 50% wider might have a large impact.

The memory bus is supposed to be twice as wide, though I would be shocked if they got a 512bit bus up to 6GHz. Where have you seen results showing that the 680 is being severely memory limited?

 

[MrK6]

Keeping my eye on big K. If it overclocks well, this could be my next card .

 

[Mars999]

The memory bus is supposed to be twice as wide, though I would be shocked if they got a 512bit bus up to 6GHz. Where have you seen results showing that the 680 is being severely memory limited?

I think he meant bus interface limited vs. the 7970's 384bit in some situations is showing a better performance for it.

What about 4GB in a 680GTX? I want the high res texture packs for Skyrim, Rage, ect... and I heard they can suck up to 2GB of data alone.... Anyone seen any benchmarks with the 680 vs. 7970 3GB to see if it matters?
Thanks

 

[blckgrffn]

I'm going to go out on a limb here, and predict that even at 550mm^2, GK110 won't be more than 30% faster than the 300mm^2 GK104 in gaming performance. Looking at the AT review, NVIDIA has done an amazing job of stripping Kepler down to the bones to get the best possible gaming experience out of it while sipping power. For compute though it is lacking, outside some scenarios where speed is solely dependent on shaders and clock. When GK110 is released I expect that all or at least the majority of the CUDA cores will be FP64 capable, and this is going to eat up die space without giving a huge boost to gaming performance.

My guess is that GK110 is going to be an absolute compute beast, but if you want gaming performance don't wait the 6 months, buy the 680 now and buy another in the fall.

+1. I agree with your speculation. That 30% more gaming performance could need ~2x the power as well.

Go SLI and forget about it for a couple years if you have the dough. Obviously if you have 7970 or three, its likely not enough to move you.

I am pleasantly surprised at nvidias execution this time around

 

[RampantAndroid]

Some people need SOMETHING from the nV side to cheer/argue about. I mean, they must be really disappointed that nV didn't release a 7970 killer for $300.

Speculation is fun sometimes but are we all really gonna now start arguing about a card that is SO FAR from release?

nVidia released a card that undercuts the competition and outperforms or holds even in all but synthetic compute benchmarks. Seems like they have something to be happy about. There's some good competition once again, and nVidia wasn't terribly late to the game this time either.

 

[n0x1ous]

As good as the 680 is, I know that Kepler can be so much more with a little more silicon.

I like the monstaz. 500mm2 or bigger for me!