Hi fellow developers, As mentioned in a couple of previous mails, e.g. "... tensorflow ...", and "... anaconda ...". The license of software stack on nvidia's hardware accelerator is always a problem. In the past, I helped Debian's nvidia-team update the nvidia-cuda-toolkit several times, and packaged some software around it, even if their proprietary licenses looked disgusting.
AMD finally started to do somthing to attempt ending nvidia's CUDA monopoly, and created ROCm, a free (freesoftware licensed) counterpart to CUDA. https://rocm.github.io/index.html Nowadays ROCm has gained support from some major applications such as tensorflow and pytorch, etc. And they also created a portability layer so that existing CUDA code can be compiled to use AMD GPUs: https://github.com/ROCm-Developer-Tools/HIP That means AMD is trying to end nvidia's proprietary CUDA monopoly. As a free distribution developer I personally and spiritually support AMD's effort on this direction. I created a Salsa team, which could be used to hold the ROCm software stack packaging: https://salsa.debian.org/rocm-team However I personally don't have any AMD hardware, so currently I'm still quite conservative in actual packaging work. Welcome to join the Salsa ROCm team, in (spiritual) support of AMD's free software effort.