What I know is that both TSMC and GF said they will "drop" the 32nm node and move directly to the 28nm node. Judging from the problems TSMC had with the 40nm process that affected both ATI (now AMD) and nVidia at the time, I think 28nm should be at least one year away, so the end of 2011 2H. I think it will be more like the end of 2012 1H, but that's just me.
As for the roadmap above, please note the "768 core" designation... 768 CUDA cores on 40nm... that's not ONE chip, that's a CODENAME for a dual GPU card. It has to be. 768 CUDA cores on 40nm will be ONE FREAKIN' HUGE chip. About 50% larger then GF100. I can't even begin to comprehend how can such a beast exist. Imagine the cooler on these cards. Impossible. Not to mention the overly complicated 512bit memory bus. 512 is codename for 2x256... see how everything drops into place