can't help you there, but i have to say i'm very curious about CUDA. the "billions of little cores" parallel architecture seems good i think it's because it g

can't help you there, but i have to say i'm very curious about CUDA. the "billions of little cores" parallel architecture seems good i think it's because it gets straight to the heart of parallelism. when you only have two or three cores/threads you might be worried about how to apportion some task to each core/thread, and worried about things like thread safety, whereas having to deal with hundreds of cores cuts that out completely and encourages purer parallelism/collective/emergent/"natural" algorithms my thinking is that maybe instead of starting with RISC processors and trying to control some number of them, we can instead start with a full parallel architecture (perhaps like what CUDA uses) with millions of cores, and "emulate" RISC/x86/etc on top of that, wherefrom it would be natural to add parallel opcodes and whatbeit these are just notions though