The first one seems to take a long time on CUDA. Thinking maybe I should stick to PG on CUDA while the getting is good.