Re: Well, I'm waiting for the refresh!
Really new parts from either Nvidia or ATI will only happen once the bulk 28nm fab process is available sometime next year,and let's put it into perspective here:
GF100 has about 3 billion transistors with a 530mm^ die size at 40nm,so at 28nm,and assuming whatever high end GPU Nvidia releases next at 28nm is also at that same 530mm^ size,and also has roughly the same ratio between logic and cache as the GF100,the 28nm version will have roughly 4.8 billion transistors.
Ati's next GPU,if it stays the same physical size as the current cypress right now at 344mm^,but built at 28nm and keeping within the same guidelines as for the GF100,it'll have about 3.4 billion transistors.
In either case,the transistor budget increase is well over 1 billion transistors,so it'll either allow to make already existing features much more powerfull,or add new features to the architecture(seems DX11.1 is coming),so for large performance and feature support increase,the 28nm fab process is what's needed for both Nvidia and ATI.
Whatever features are going to be added with 40nm refresh parts from Nvidia,given that it's current GF100 is already large at 40nm,might simply resume to having all shaders enabled(512),running at faster clocks and also using faster GDDR 5 memory,while still staying at or preferably below that of the current GTX480 cards in terms of power consumption.
ATI has a biit more flexibility since their current cypress is quite a bit smaller(344mm^),and uses less power too,so these upcoming southern islands parts are more of a wildcard,but rumors to suggest it might be using a 384 bit memory bus,with the attendant increase in memory bandwith and extra memory onboard the cards....we'll see in 3~4 months from now.