So the limit of GPU speed is wholly dependent on die shrinks? Look at the history of GPUs for the last 6 years. Even with die shrinks, the power consumption has continued to rise.
I can't possibly predict but something more radical needs to happen for Moore's GPU laws (is there such a thing?!?) to continue.
Well, it is partially our fault.. It would certainly be possible to keep creating slightly faster, much lower power cards. But the market for an upgrade for power reasons is nearly non existent. Though what seems to be common is the mainstream performacne cards of each generation are lower power/heat, higher feature set versions of the previous high end.
As far as Moore's law goes.. we are always increasing the density.. I don't think anyone ever expected that trend to mean lower power as well, at least not lower power of the total package (though power/performance has been getting better over the years).
What I think they will do is more than likely continue with each generations high end cards becoming more and more power hungry to a point. By the time pcie 4.0 comes out we might have drastically different architecture to play with, so who knows. We are at a point now where we need to migrate to pcie 3.0 more for power constraints on the high end than bandwidth.. which I find rather silly..
I can't think of a situation where more performance wouldn't be sought after, even if resolution starts to reach a plateau there are other things we can compute on a GPGPU. Unfortunately the more transistors we make the more power it takes to run them, and our improvements in manufacturing efficiency will likely never catch up to our increases in transistor count so far as to plateau energy consumption. That is to say until we migrate to something totally different.
Facebook
Twitter