Why support starts from CUDA 10 and not from CUDA 8 ?
Why support starts from CUDA 10 and not from CUDA 8 ?