If we want to use tensorflow that requires cuda 11 but we have cuda 12 system wide

Description

What do we need to do to install a version of cuda 11 so that we can use tensorflow from the precompiled pip packages.

My OS uses cuda 12 so I don’t want to install cuda 11 system wide and subvert my package manager.

Environment

TensorRT Version:
3053
530.41.03
8.1
Fedora 37
3.11
2.12

As it turns out. I found a version I downloaded a version of cuda 11.8 by using the search for “cuda 11.8.0 download”. The link from the version 12 download page to archived versions did not provide an actual working link but always ended back at 12. Then I used the run file with the --installpath option to install it to a local directory. I did not have to use sudo to install this, which I think is a change from older cuda runfiles.

Hi @matthew.smith3 ,
It seems like you solved it, but in any case, take a look here, with the cuda containers you can avoid having to install locally and modify your current enviroment in order to run other CUDA versions.

Regards,
Andres
Embedded SW Engineer at RidgeRun
Contact us: support@ridgerun.com
Developers wiki: https://developer.ridgerun.c om/
Website: www.ridgerun.com