[citation][nom]dragonsqrrl[/nom]I think it's really funny how performance per W and cost were the most important arguing points for AMD fanboy's... up to this launch. For generations performance per W and pricing were the only things that mattered, more than absolute performance, and certainly more than compute performance. Compute? What's that? It's irrelevant for consumer cards. But now that Nvidia has implemented performance efficiency better than AMD ever has, while in addition coming in at a lower price, the emphasis in their arguments shift to compute performance. Well, I guess that's what makes them fanboys. Ignore weaknesses, emphasize strengths, and throw consistency and self respect out the door... amazing.[/citation]
If compute performance didn't matter, Fermi wouldn't exist as we know it. Civ V makes use of compute shaders which is why Cayman didn't do so well at it and, despite the lack of driver command lists, Tahiti does far better. So, yes, compute can and does matter in some games.
How can you be sure that NVIDIA provides a more efficient performance solution "than AMD ever has"? I remember when the 6xxx series came out and a lot of people thought it was less efficient than the 5xxx series, plus also remember that every new architecture doesn't perform at its best with the initial drivers. I may sound like an AMD apologist here, but the performance of the GCN architecture will improve over time as will that of Kepler. You MIGHT be 100% correct about Kepler, but until the drivers are mature and detailed tests are conducted, you can't very well say that for now. Let's put it this way - compared to Fermi, it could hardly go backwards, though admittedly, I personally never expected such an improvement.
Your last statement certainly made me chuckle... "Ignore weaknesses, emphasize strengths, and throw consistency and self respect out the door... amazing." Funny, because that's what NVIDIA fans have been doing as regards Fermi, though I won't be so harsh as to factor consistency and self respect into the equation.
The 680 is the new king, and it should bring down the price of the 7970, but this is AMD's own first generation Fermi, if you like, and NVIDIA seem to have gone for a more Caymen-style architecture in comparison. It'll be interesting to see where they both end up.
[Small edit - it was the 5xxx series that was considered less powerful per shader than the 4xxx series]