Nvidia BigK GK110 Kepler Speculation Thread

It sounds basically as if they committed everyone they could make to to the highly lucrative HPC market for this year and maybe more.
 
Thanks. Guess I'll just give up and get one of those ASUS TOP 680's as it seems there's nothing interesting coming down the pipe. Whenever they become available that is.

Well at least you hang in there longer than I did and that is a good choice for a 680, performance is very good out of the box and the cooler is excellent.
 
added quote for clarity:
What are the odds of a 28nm GK204 refresh later this year with higher clocks around 360mm^2?


They've only just started filling out the top two tiers of GK104, though.
If a GK204 came out in that time frame, it would result in the odd situation of the high-end big die being lapped in terms of design generation.
 
Last edited by a moderator:
If a GK204 came out in that time frame, it would result in the odd situation of the high-end big die being lapped in terms of design generation.

I don't think so. I think that GK110 will have all the small touches here and there which GK114 will be characterised with.

GK204 is a very interesting name, where did you get it from?
 
added quote for clarity:



They've only just started filling out the top two tiers of GK104, though.
If a GK204 came out in that time frame, it would result in the odd situation of the high-end big die being lapped in terms of design generation.

Maybe but thats less of a concern if nVidia is hell bent on market segmentation. Sounds like GK110 might never see the inside of a PC.
 
That's 22% larger than GK104. Are you implying that the GK204 could have a wider bus, more SMXes, and/or something else?

Nothing that specific. It's just that they have a lot of TDP to play with and could do some damage with a tweaked GK104 core with more units.
 
Can we stop to fight the nividia architecture vs AMD just for a while ?


each g104 and GCN are extrremly close in games peformance clock to clock, with the difference is GCN a lot aimed at computing ( just need watch SP numbers ) ( i will not even discuss DP numbers, this is a no comment )


How much ALU will need GCN for counter a GK110 Knowing the SP beteween the K20 and the K10 will not change so much ( +15 %- + 25% max on SP max or Nvidia will be happy to say his next generation will oblierate the K10 in SP too ) ...

(GK104 is basically in term of pure raw power, a lot inferrior of the GCN 1.0 .. or HD 7970 .. ) ( I dont speak for games, i just said it for pure capability power )
 
Last edited by a moderator:
Nothing that specific. It's just that they have a lot of TDP to play with and could do some damage with a tweaked GK104 core with more units.

Looking at the 670 performance compared to the 680, it's more than obvious that the latter is screaming for more bandwidth. In other words I'm afraid that any speculative scenario with higher frequencies than now or more units (or both) won't result into any worthwhile performance increase without an as generous bandwidth increase.
 
Looking at the 670 performance compared to the 680, it's more than obvious that the latter is screaming for more bandwidth. In other words I'm afraid that any speculative scenario with higher frequencies than now or more units (or both) won't result into any worthwhile performance increase without an as generous bandwidth increase.

I don't know if it's ever been done before, but is there any good reason not to make a GPU with a native 288- or 320-bit bus?
 
Well, it would give quite odd memory pool sizes of course. While this isn't really a technical drawback it could be "confusing to consumers", as the term goes.
 
Back
Top