What do you think about using a Cell CPU that has 30x FPU performance to that of a Intel PC as the CPU for a computing platform?
Cell has around 200 GFLOPS of single-precision floating point power. If you code for them explicitly (you can't just recompile standard FP code). Impressive for its time, but irrelevant today when you can get GPUs 20x faster.
A modern Intel CPU's HD4000 graphics can achieve up to 294 GFLOPS, if you are happy coding in OpenCL. The general purpose AVX isn't a slouch either.
