Last week I watched a video on Stanford on iTunes by Dave Patterson about the end of Moore's law for uniprocessor speed increases and the sea change in chip designs that's going to happen as we move towards massively parallel systems for the mainstream market. One possibility seems to be that we'll have computers with, say a hundred simpler smaller CPUs and 1 fast CPU to run some tasks that can't be parallelised (a lot of OS responsibilities I guess). It's definitely worth checking out - it's pretty dense, he covers a lot of ground but he's a great speaker. His book was one of my favorites as an undergrad. By coincidence there was a seminar in the computer lab here on the same subject that i went to on Wednesday.
I'm curious as to when the current SMP approach will move to a new, very parallel design. How many x86 CPUs will be packed into a single machine before the chip manufacturers cave and seriously push a totally new design?
