I think it's more likely to be like 400 < 5000 this time because we're at another node change, and step one (AMD launching first) has already happened. The big weakness of 500 mm2 + dies is that it takes longer to get them right. Kepler 1 will probably be a bit like fermi 1, and, just like the recent generation, the refresh part will benefit immensely from the matured manufacturing process.
Hmmh... well, we'll have to wait and see.
I'll be waiting though. We never know what'll really be best. We can presume or predict situations through trends in the past, but something may be different this time around. :|
AMD is averaging out GPU performance with increased computational power for situations such as GPGPU, rather than heading for fully fledged gaming performance. Once the new architecture is done, then I suppose they can focus on big boosts in all of those, across the board, in the future. ^_^
