I need CUDA on Workstation for an offline open source AI training tool. Running the Nvidia bin blob is trash with how often fedora updates the kernel. The automated kernel module build is awesome, but now I need CUDA and the documentation is all over the place and outdated. I know it is because it is shitvidia. AMD doesn’t make competitive laptop hardware with 16GB of GPU memory or more and nothing with full CUDA compatibility with HIPS yet. I’m disabled where a laptop is the only option for ergonomics. Does anyone here know a proven path for sustainable CUDA on Fedora?
I’m not sure an easy solution to your current problem, but you can always use hipify-clang to convert it to rocm which is supported on every gpu, not just amd ones.