Python 3.8 and incompatible CUDA drivers

[KevinMusgrave]

Current Behavior

In a python 3.8 environment, the following type of error occurs when installing two entirely separate packages:

The following specifications were found to be incompatible with your CUDA driver:

  - feature:/linux-64::__cuda==10.2=0

Your installed CUDA driver is: 10.2

You can see specifically what I mean in this spyder-feedstock issue, and this pytorch-metric-learning issue. In both cases, users find that switching to python 3.7 fixes the problem. Also, here's a stackoverflow question in which the user has the same incompatible CUDA error when installing PyTorch.

This problem is difficult to debug because the error message says the CUDA driver is incompatible, even though the driver numbers match.

Steps to Reproduce

In a python 3.8 environment:

conda install spyder

or

conda install pytorch-metric-learning

Expected Behavior

The CUDA driver should be compatible, or the provided error message should be more informative.

Environment Information

Here is an environment one user posted in the pytorch-metric-learning repo:

channels:
  - anaconda
  - conda-forge
  # - nodefaults
dependencies:
  - anaconda::flake8
  - anaconda::nb_conda
  - anaconda::notebook
  - anaconda::pandas
  - anaconda::pip
  - anaconda::pylint
  - anaconda::pytest
  - anaconda::python=3.8
  - conda-forge::black
  - conda-forge::jupyter_contrib_nbextensions
  - conda-forge::nb_black
  - nodejs
  - pip

[KevinMusgrave]

There was an error in the way I wrote the meta.yaml file. For any conda-building noobs out there, you can take a look at how I fixed my problem here

[interwebjill]

I am having this issue with wand. While you solved your individual issue, there is still a more general problem:

https://stackoverflow.com/questions/63335480/install-wand-package-in-conda-on-windows

emcconville/wand#492

[KevinMusgrave]

@interwebjill I'm not familiar with that package, but you could try this as a workaround

conda install pip
pip install Wand

[interwebjill]

Yes, that is the solution, thank you. Turns out that there is not a conda package for wand. So the larger issue is that this conda error message isn't accurate.

[github-actions]

Hi there, thank you for your contribution to Conda!

This issue has been automatically locked since it has not had recent activity after it was closed.

Please open a new issue if needed.