I have/had no idea what he was "just saying...." so I ignored it.
The only point I was making was if you had 32nm gpu's or not, it has zero bearing on the 28nm process, how many transistors can fit into a certain size or what configurations of the architecture work or don't. THe 7970 was around 350mm^2, it will be as big as they were going to go with a 384bit bus, and if they went with a 256bit bus they would have gone smaller.
They weren't planning, or going to do a 450mm^2 core on launch because they knew you couldn't do it with suitable yields, Nvidia couldn't do it either. You can't just make a 400mm^2 die because you want to, it has to be balanced Nvidia couldn't just make something inbetween the gk104/gk110, the later had poor yields even pushing for a later than usual launch. Gk104 was as big as they would go on 256bit bus.
A 32nm cayman would have made absolutely no difference, not even one transistor difference, to 28nm.
They don't make a chip on a new process a specific size based on the previous processes, it's an architecture that has balanced points. 7970/680gtx was what could be made at launch, you can't magically up performance by 30% because a previous gen card was faster.