Error:torch.cuda.is_available() return true in cmd but return false in pycharm

1

In cmd, I printed “import torch” then “torch.cuda.is_isavailable()”, it returned true;
but at python console in pycharm I got false.
I have set the right python interpreter, and there is ‘pytorch’ under the package menu. Are there any possible causes? Thanks in advance.

2

You are most likely using different Python environments and the one used in PyTorch might have installed the CPU-only binary.
Also, according to this topic the user claimed os.environ["CUDA_VISIBLE_DEVICES"]='1' was set somehow in Pycharm without manually setting it, which was hiding the only available GPU.

3

I see the python envirment I set is …\anaconda3\python.exe, you mean this is python2 but not python3? I see the environment in cmd is python 3.10 indeed.But what is CPU-only binary?

4

No, I meant different virtual Python environments which allow you to isolate installed packages.
“CPU-only” binary refers to the pip wheel or conda binary shipping with CPU kernels only without any CUDA kernels or CUDA libraries.

5

Thanks a lot. I find where is wrong. My pytorch Build is py3.9_cpu_0. I will uninstall it and try again.