Five versions of CUDA on a Redhat server

I went to Redhat support with this issue, and they said it was an nVidia issue. The OS was installed with their media and gets updated from their repos. ???

I have been fighting with this for weeks and hope someone here can help.

It appears that I have five different versions of whatever CUDA is installed on this system:

5.4G /usr/local/cuda-11.8
4.8G /usr/local/cuda-11.8/targets/x86_64-linux/lib
4.8G /usr/local/cuda-11.8/targets/x86_64-linux
4.8G /usr/local/cuda-11.8/targets
4.6G /usr/local/cuda-11.4
4.2G /usr/local/cuda-11.6
4.1G /usr/local/cuda-11.7
4.1G /usr/local/cuda-11.5
4.0G /usr/local/cuda-11.4/targets/x86_64-linux/lib
4.0G /usr/local/cuda-11.4/targets/x86_64-linux
4.0G /usr/local/cuda-11.4/targets
3.7G /usr/local/cuda-11.6/targets/x86_64-linux
3.7G /usr/local/cuda-11.6/targets
3.6G /usr/local/cuda-11.7/targets/x86_64-linux
3.6G /usr/local/cuda-11.7/targets
3.6G /usr/local/cuda-11.6/targets/x86_64-linux/lib
3.5G /usr/local/cuda-11.7/targets/x86_64-linux/lib
3.5G /usr/local/cuda-11.5/targets/x86_64-linux
3.5G /usr/local/cuda-11.5/targets
3.4G /usr/local/cuda-11.5/targets/x86_64-linux/lib

I am running low on disk space on root and want to remove all that I can. Can anyone guide me how to do this?

Redhat Enterprise Linux Server 8.6
Dell PowerEdge R740xd
2x Tesla V100-PCIE-16GB cards

Thanks in advance.

  • Dave

Hi @dclausse

Welcome to the NVIDIA Developer forums. The regional category does not offer support, so I am moving your topic to the CUDA forums for help.

The reason you have anything in /usr/local/cuda-XXXX/ is because someone with access to the machine installed it, not due to anything that RedHat did or that NVIDIA did. This is not an NVIDIA issue. It is completely legitimate for development work to have multiple CUDA versions installed. It’s a matter of user preference.

The CUDA linux install guide gives removal instructions. It’s impossible to tell you what you “can” remove in the sense of what your machine users may want/need. If you have no need of CUDA at all, then simply doing

sudo rm -Rf /usr/local/cuda*  

will free up the disk space used by those CUDA installations, but no one will be able to use that machine for CUDA development work after that (unless another CUDA install is done at that point).