Has Anyone Successfully Built and Loaded the nvidia-peermem Module on Clear Linux?

Q&A
1

Hello everyone,

I’m currently using Clear Linux on our lab’s cluster, but I’m having trouble getting the nvidia-peermem module to load properly via modprobe. I suspect this might be related to the InfiniBand drivers, as the default rdma-core package may not support this module. It seems like installing MLNX OFED or DOCA OFED might be necessary, but I’m not entirely sure.

Has anyone successfully built and loaded nvidia-peermem on Clear Linux before? If so, could you please share your setup and steps to achieve this? Any guidance would be greatly appreciated.

Thanks in advance!

2

Hi, a few months ago I tried installing the Nvidia drivers (GT730) on ClearLinux following the guide, but unfortunately I got stuck at a certain point. I’m afraid I wouldn’t be able to help you

3

The easiest way to install NVIDIA drivers on Clear Linux is to use @marioroy 's github project.

After downloading the latest drivers and CUDA from NVIDIA’s archive today, I installed them. Copy and paste from my terminal history:

git clone https://github.com/marioroy/nvidia-driver-on-clear-linux
cd nvidia-driver-on-clear-linux
bash ./pre-install-driver update
bash ./pre-install-driver
bash ./install-driver ~/projects/NVIDIA-Linux-x86_64-570.124.04.run
bash ./install-cuda ~/projects/cuda_12.8.1_570.124.06_linux.run

Then edited the ~/.bashrc to look like

me=${BASH_SOURCE[0]//\//_}; me=${me//./_}; if [[ ${SOURCED[${me}]} == "yes" ]]; then return; else declare -A SOURCED; SOURCED[${me}]=yes>

# Use global profile when available
if [ -f /usr/share/defaults/etc/profile ]; then
    . /usr/share/defaults/etc/profile
fi
export CUDA_HOME=/opt/cuda
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

If you are paranoid about the scripts, check it out :laughing:

It appears nvidia-peermem requires a specific older CUDA version and I have not tried to build it. I installed the latest CUDA.