I am getting nccl error while running my code on multiple gpu’s.so i want to install pytorch version compatible with cudaWhen i was running on singlen gpu everything was working fine but on multiple gpu’s it is throwing error so can anyone tell me how to rsolve it. I have pytorch 1.10.0
CUDA 9 is quite old by now and not supported anymore in current PyTorch builds. However, note that the PyTorch binaries ship with their own CUDA dependencies and will not use your locally installed CUDA toolkit unless you build PyTorch from source or custom CUDA extensions.
I thus guess you mean the NVIDIA driver?
Could you update PyTorch to the latest stable or nightly release and check if you would still see these errors?