Originally Posted by Newbie2009
Would they not do that before the die shrink?
It depends. If they already have engineering samples of 28nm and realize its not enough yield, or whatnot, then tweaks will begin. We know its not anything dramatic but minor revisions are allowed. This scenario is far more than likely, but it would be still be my assumption till release.
The fact that they screwed something up is their (SA) conclusion which I think its a leap / stab at nVidia. Concerning their architecture plans and based of the performance per watt roadmap of kepler/maxwell etc, something tells me that performance will not be a problem.
The 5xx generation was mightily impressive and could have easily been the 4xx generation had there not be a massive push for market share and releasing GPUs. Call it a mistake or what not, I personally see that even the performance of the 4xx generation was far greater than the 2xx generation. I could anticipate either the same mistake of greater performance at a cost of heat or power usage, or possibly straight to the equivalent 5xx release, and bypassing a similar 4xx release. Either way, transitioning to 28nm has been hard for both sides from everything I have picked up. I would suspect that its not cut and dry die shrinking as anticipated and its more complicated then we would like to believe.