Do I need to install CUDA and CuDNN on a WSL2 instance?

It seems that WSL2 can directly use the Windows host instance of the GPU and therefore does not need me to install CUDA support within Ubuntu 22.04. Running nvidia-smi inside the WSL reports a Persitence-M graphics card, and Nvidia GeForce RTX 2060 so looks good. Any Tensorflow code I run does use my GPU. Great. However i have some code that needs CuDNN (specifically DLIB) and I am wondering do I install this in Windows 11 host or in the Ubuntu WSL2 instance?

1 Like

Hi @simon.bunn ,
I am checking on this with the engineering team.
Will share the update.
Thanks

Hi @simon.bunn ,
If you are using your application in WSL2, you would need to install cuDNN within WSL2.
Thanks

1 Like

should we also install cuda toolkit?

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.