How to install multiple CUDA versions on a single Linux server

After you installed the first CUDA toolkit (preferably the latest stable version), and you want to install others beside it, due to some compatibililty issues for example, just do the following:

Download the standalone file installer, e.g. https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda_11.7.0_515.43.04_linux.run

chmod +x file_name
sudo ./file_name

When prompted for the components, you MUST deselect the driver, better just keep the CUDA toolkit and deselect all others

When prompted to update the /usr/local/cuda symlink, DON’T accept

After installation is done, just run the following and you are good to go:

export PATH=/usr/local/cuda-11.7/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH

Tested and working properly on Ubuntu 22.04 with A100 GPUs

Enjoy!

Source: https://medium.com/@peterjussi/multicuda-multiple-versions-of-cuda-on-one-machine-4b6ccda6faae

Unknown's avatar

About Ahmed Tawfik

Cybersecurity Professional, Systems Engineer, OSS & Linux Geek
This entry was posted in Linux and tagged , , , . Bookmark the permalink.

Leave a comment