Description
I realised that most of my problems are rooted from a version incompatibility among one of those cuda, cudnn, pytorch and python on windows 11.
I have 2 NVIDIA GeForce RTX 2080 Ti
Windows 11
First try:
I installed following version combination inside my 'stylegan' conda environment:
- python=3.7.16
- cudatoolkit=11.0 (manual installation, can be downloaded here )
- cudnn=8.05.39
- pip install torch===1.7.1+cu110 torchvision===0.8.2+cu110 torchaudio===0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
- pip install click requests tqdm pyspng ninja imageio-ffmpeg==0.4.3 psutil scipy
everything is great except for cudatoolkit=11.0(it only supports windows 10). So, some drivers support cuda 11.0 not windows 11 and vice versa. cudatoolkit>=11.5 supports windows 11.
Second try:
I have uninstalled all cuda related software from my PC. I have created a new 'stylegan2' conda environment to keep second installation separately. This time I have tried to install cuda's compatible version with windows 11.
- python=3.11.9
- cudatoolkit=11.8(manually installed using .exe file, can be downloaded here )
- cudnn=8.2.0 (for all cuda 11.x)
- conda install pytorch==2.2.2 torchvision==0.17.2 torchaudio==2.2.2 pytorch-cuda=11.8 -c pytorch -c nvidia
- pip install click requests tqdm pyspng ninja imageio-ffmpeg==0.4.3 psutil scipy
In that case, there are following errors
'Failed to build CUDA kernels for upfirdn2d' because 'conv2d_gradfix not supported on PyTorch 2.2.2'.
This time, everything is great except for pytorch=2.2.2.
Can I never ever find cuda, cudnn, pytorch and python version combination that meets all requirements on my windows 11?
If anyone run the project on Windows 11 without errors. Please let me know cuda, cudnn, pytorch, python versions that you used.