Originally Posted by Crap Daddy
Don't expect a "performance" chip to beat AMD's top dog. That will come later with the GK110. I think GK104 will be (at stock clocks) between the 7950 and the 7970.
Well I fully expect that to happen actually. GF114 (aka fully enabled GF104) did match AMD's top dog, Cypress. 40 nm didn't play very well for Nvidia, BUT if everything had gone well, their performance chip GF104 would have matched AMD's fastest, and GF100 would have been 40% faster (like GF110 is).
Most people don't take these things into account, but they are very important, because that's what Nvidia is going to aim for this time around too. Why? Because of the die size, we know they are aiming at the same kind of improvement they tried with GF104 over the previous gen. And we know how that would had played if 40 nm had not become a nightmare for Nvidia. AMD released Barts and Cayman, replacements that were significantly bigger than the chips they were replacing at the time that Nvidia only "fixed" GF100/104, and that's why GF110/GF114 vs Cypress/Juniper is a comparison no one does, but in order to better understand what are the posibilities with Kepler it's very very important to note them, because once again the die sizes are the same. Conditions are pretty much the same, except that 28nm is simply much better.
This time around things are supposed better for Nvidia, and GCN is clearly less efficient (for gaming) than VLIW, so 10% here + 10% here and GK104 could easily end up being significantly faster than HD7970.