This is exactly why I'm really hoping that OpenCL takes off. While a good read, to me the article shows why you shouldn't necessarily continue with the status q

This is exactly why I'm really hoping that OpenCL takes off. While a good read, to me the article shows why you shouldn't necessarily continue with the status quo just because "it works" (granted, the article was written back in '02 when there was no other real option). We have tons of insanely cheap hardware now that handles SIMD better than any CPU, it's about time we start making use of it.