Is there any way to get Pytorch to use an old GPU (NVS 300)

I have an old computer at work with a NVS 300 card which only supports up to CUDA 6.5. It uses Windows 7 (never upgraded to 10 due to compatibility issues with some of the software.) I’ve tried compiling Pytorch on this system to no avail (although I’ve successfully compiled Pytorch from source on multiple Windows machines with GPUs that support CUDA 8+.)

Has anyone ever gotten Pytorch GPU to compile on such an old machine? Or should I just give up and use CPU on that one?


You can surely compile without CUDA support.
But CUDA 6.5 is too old and most of the cuda code use recent features. So you won’t be able to compile with CUDA support.

Oh well, no point in compiling if there are packages available for the CPU version of Pytorch