Using colours as an analogy for processing on a graphics chip is particularly poor, don'tcha think?
However, tests have already proven that GROMACS runs many, many times faster on an R520 than current processors do, and that was bound by DX. The primary processing for F@H has always been a math bound problem, and inhernatly lends itself to the massess of vector math processing that GPU's have in relation to standard CPU's, which is why Stanford have always called the GPU research the "High Performance Client".
However, tests have already proven that GROMACS runs many, many times faster on an R520 than current processors do, and that was bound by DX. The primary processing for F@H has always been a math bound problem, and inhernatly lends itself to the massess of vector math processing that GPU's have in relation to standard CPU's, which is why Stanford have always called the GPU research the "High Performance Client".