we're still working on this code, and it has a few problems - like fully maxing out your CPU *and* your GPUs (because Fermat testing is still more efficient to handle on the CPU, although we now have also an alternative Fermat path on the GPU).