CODEWITHSUNDEEP
pythonpytorchconda
Rong
Rong

Reputation: 21

torch.cuda.is_available() True in (base), but False in other conda env

I run this code in Anaconda prompt, and it returns True.

(base) C:\User
torch.cuda.is_available()
True

But when I run other conda environment, it just doesn't work.

(pytorch_project) C:\User
torch.cuda.is_available()
False

The problem seems to be different results of torch.version.cuda.

(base) torch.version.cuda = 10.1

(pytorch_project) torch.version.cuda = 10.2

But I don't know why they are different...

How do I make 10.2 down to 10.1 and make is_available() == True ?

Here is my info.

 Windows 10 / nvidia-smi=425.31 / CUDA ver=10.1 / pytorch=1.4.0 / torchvision=0.5.0

conda list

Upvotes: 1

Views: 3801

Answers (2)

Rong
Rong

Reputation: 21

Thank you for your answers and comments. <3

I've solved the problem.

I use Visual Studio Code as the developer environment tool, but as shown in the picture I uploaded, conda list pointed at exact the same directory, which means I didn't actually activate my environment.

It should be like (pytorch) C:\User , but it's (Power Shell) PS C:\User instead.

To solve this, I went to Settings → Terminal > Integrated > Shell Args: Windows and edited settings.json with

"terminal.integrated.shellArgs.windows": ["-ExecutionPolicy", "ByPass", "-NoExit","-Command","& 'C:\\Users\\user\\miniconda3\\shell\\condabin\\conda-hook.ps1'" ]

Reference: https://blog.lcarbon.idv.tw/vscode-設定-anaconda-路徑至-visual-studio-code-終端機中windows/

By starting a new terminal I got (pytorch) PS C:\User correctly, and saw my torch=1.5.0 using conda list.

Then I run conda install pytorch torchvision cudatoolkit=10.1 -c pytorch to change the pytorch version.

And Voilà!!

torch.cuda.is_available()
True

Upvotes: 1

Sergii Dymchenko
Sergii Dymchenko

Reputation: 7209

Try to uninstall PyTorch and torchvision from pytorch_project environment and install again in this way:

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch

Or, if you prefer pip:

pip install torch==1.6.0+cu101 torchvision==0.7.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

See https://pytorch.org/get-started/locally/ for details and more options.

Upvotes: 0

Related Questions