0

I installed CUDA & cuDNN using the conda method described in Nvidia guide inside a conda environment "tf_gpu_01". I'm guessing this is not a system wide install as compared to using something like sudo apt install cuda or whatever steps like that.

Now if I want to create another conda environment "tf_gpu_02" with the same version of CUDA & cuDNN do I have to install them again in the new env or clone the old env.
Or could I refrence the CUDA path from the older env to the newer env,
My main concern is I create a lot of envs for different projects and I don't want to end up with multiple huge files eating disk-space all for the same versions of CUDA & cuDNN.

Amr Yasser
  • 17
  • 10
  • 1
    Also see https://stackoverflow.com/a/55602940/570918. That is, if you use exactly the same version, then Conda will use hardlinking to minimize redundant files. – merv Sep 27 '22 at 23:29
  • @merv That is exactly what I'm asking about. Thanks – Amr Yasser Oct 02 '22 at 10:55

0 Answers0