I don’t expect there to be much difference. We have already enabled CUDA for the main functions where a parallel approach would be faster than a serial. We do keep the CUDA versions up to date and the newer versions require newer device drivers as you found out, NVIDIA might have changed the performance of their code/drivers.