Michael Suess over at Thinking Parallel has a thoughtful short piece on potential inflection points in the value curves between using the fruits of the gaming world (GPUs and Cell processors) and traditional CPUs, mostly motivated by all the press that Folding@Home is getting these days (for example, here).
He briefly considers why FLOPS don’t really matter as a measure of value and moves quickly into the very large real-world obstacles to adoption of a totally new processing paradigm: code, and everything to do with it. Current applications, compilers, and expertise are all heavily invested in traditional CPUs.
I agree with all of this. Of course, similar arguments could have been made (were!) during the move from vector to RISC architectures, and I would have agreed with those then, too. Ok, so my career as a futurist is limited. But I do know that people hate change and that groups of people hate change even more. When they are free to choose, people rarely leave an old technology for a new technology without a clear order-of-magnitude type benefit.
[Update: Michael sent me a very nice email pointing out that his name is Suess, not Seuss as I originally had it. I blamed my kids because that's the kind of guy I am. I also updated the title of this post to reflect the correct spelling.]