To avoid: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation. If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].

[yuedajiong]

in ./nvdiffrast/torch/ops.py

os.environ['TORCH_CUDA_ARCH_LIST'] = '.'.join(map(str,torch.cuda.get_device_capability())) #=''

To avoid: UserWarning: TORCH_CUDA_ARCH_LIST is not set, all archs for visible cards are included for compilation. If this is not desired, please set os.environ['TORCH_CUDA_ARCH_LIST'].

[Clement-Lelievre]

in my case, what worked was to comment out this line