What they did is massively improve the back end of the design ,all on 40nm.Now all they will do is use 28nm to scale up SP count,IMO big time. I can see 2xVLIW4 we have in Cayman,with 28nm version of NI. They are already awfully close to 580GTX with Cayman XT,especially in high resolution new DX11/tessellation heavy games. If AMD and NV effectively double the compute power with next gens,we may have a deja vu next year:AMD having 5-10% less performance with 36-40% smaller die- a clear design efficiency advantage at the same node.