Is there evidence to show that feature extraction is actually any better using CPU than GPU? The comparisons that I’ve seen have had less variability in accuracy between CPU-GPU than there was between CPU-CPU or GPU-GPU suggesting there is no real difference.