windows 10
I have install Pytorch cuda by this …
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
Now …
torch.cuda.is_available()
>>>False
But
torch.backends.cudnn.enabled
>>>True
windows 10
I have install Pytorch cuda by this …
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c nvidia
Now …
torch.cuda.is_available()
>>>False
But
torch.backends.cudnn.enabled
>>>True
Did the install logs show that the cudatoolkit
was indeed installed or did you see a fallback to a CPU version?
If you’ve verified that the CUDA version was indeed picked, the error might be raised by a setup issue, i.e. your GPU cannot be detected.
Is nvidia-smi
working properly and do you have a sufficiently new NVIDIA driver to run the CUDA11.1 runtime?
This could mean that a CPU-only package is already installed. Create a new virtual environment and reinstall PyTorch there (and check the install logs for cudatoolkit
or the corresponding CUDA tag in the wheel name) or alternatively uninstall all PyTorch packages in your current environment and reinstall it again.