Approach is actually the official installation guide - https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html Whenever kernel is updated (possbily via yum/dnf update/upgrade) the drivers need recompilation which means that one would reinstall them via package manager as given by the script.
Check if system has NVIDIA GPU on PCI bus
lspci | grep -i nvidia
Whatever is the base folder, recommended is to be at home. To go there, execute:
cd ~
Download the 'custom' made installation script:
curl -LO https://raw.githubusercontent.com/cgi-estonia-space/ALUs-platform/main/rhel/8/ootpa/install_nvidia_driver.sh
chmod +x install_nvidia_driver.sh
Install stuff (depends on the system's permission setup, might need sudo)
./install_nvidia_driver.sh
or
sudo ./install_nvidia_driver.sh
and finally when all OK
sudo reboot
After bootup run the following command:
nvidia-smi
It should display the GPU properties and other runtime info
Download the 'custom' made installation script:
curl -LO https://raw.githubusercontent.com/cgi-estonia-space/ALUs-platform/main/rhel/8/ootpa/install_cuda_base.sh
chmod +x install_cuda_base.sh
Install stuff (depends on the system's permission setup, might need sudo)
./install_cuda_base.sh
or
sudo ./install_cuda_base.sh
and finally when all OK
source ~/.bashrc
nvcc --version