I got RTX 4060 with driver 560.76. I installed CUDA 11.8 and copied cuDNN 8.9.7. files to the correct directories in the CUDA installation folder. I added the right paths to the System variables Environment. Finally, created virtual environment and installed Pytorch 2.2.1 with torchaudio 2.2.1 and torchvision 0.17.1.
When I run this to check if Pytorch detect CUDA:
print(torch.cuda.is_available())
I get FALSE.
Same when I run
nvidia-smi
No running processes found
In a different virtual environment, I use TensorFlow and it does defecting my GPU
What am I doing wrong?
Ok I was able to fix it by reinstalling the three libraries, specifying the CUDA version:
pip install torch==2.2.1+cu118 torchvision==0.17.1+cu118 torchaudio==2.2.1+cu118 -f https://download.pytorch.org/whl/torch_stable.html
Close everything and rerun it from the beginning.