Quote Originally Posted by prava View Post
Lets be honest, guys. Does anybody believe that a GK104 can be much faster (say, more than a 5%) than a 7970 while having LESS power consumption? I don't believe in miracles whatsoever, and this is something that has never happened before, not accounting for failures on either side. Either way, performance/power-consumption is more or less even within the same node for both brands (provided there are no mistakes, such as GF100 or R600).
Quoting this for emphasis since I want to bring up some points here:

Tahiti is the first AMD core that puts a ton of focus on compute and DX11 rendering rather than being a unified architecture with some DX11 bits tacked on.

Fermi was built from the ground up for compute and DX11.

AMD is currently grappling with the heat, power consumption and die space (albeit at on a more efficient process technology) restrictions that NVIDIA went through with Fermi.

As mentioned by NVIDIA in several public presentations, Kepler is meant to be a second generation "ground up" architecture which essentially means they have learned from past experiences and have focused upon perf per watt. At least that's what this slide from 2010 says.

AMD is in many ways still on their 1st generation dedicated compute / DX11 architecture due to Tahiti / GCN skipping 32nm and instead waiting for 28nm to come around.

Take that as you will. Those are just the facts.