Giters

mothguib / arch-cuda-parallel

Multiple Version of CUDA Libraries On The Same Machine

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Multiple Versions of CUDA Libraries On The Same Arch Linux Machine

This file contains some useful aliases and way to handle multiple CUDA versions in Arch-based distributions. After doing all steps from CUDA installation, reboot system. Initially, CUDA will not be set up. You need to set it using alias set-cuda-X-Y. If you want to change version, you must first use unset-cuda alias, and then set new version.

Edit content of ~/.bashrc

# some useful aliases, not related to CUDA
alias ll='ls -alF'
alias la='ls -A'
alias l='ls -CF'

# Alias definitions.
# You may want to put all your additions into a separate file like
# ~/.bash_aliases, instead of adding them here directly.
# See /usr/share/doc/bash-doc/examples in the bash-doc package.

if [ -f ~/.bash_aliases ]; then
    . ~/.bash_aliases
fi

# All things that you want to add to the path on login add before this comment!
RVG_PATH=$PATH

Create if does not exist, otherwise edit ~/.bash_aliases

alias clean-path='export PATH="${RVG_PATH}"'  
alias clear-path='export PATH="${RVG_PATH}"'  
  
alias unset-cuda='export PATH="${RVG_PATH}"'  
alias unset-cudas='export PATH="${RVG_PATH}"'  
alias set-cuda-9-2='export PATH="${RVG_PATH}:/opt/cuda-10.0/bin"'  
alias set-cuda-8-0='export PATH="${RVG_PATH}:/opt/cuda-11.0/bin"'
  
alias gpu-temp="nvidia-smi --query-gpu=temperature.gpu --format=csv,noheader"

CUDA installation

  1. Download CUDA & CUDNN from either https://archive.archlinux.org/packages/c or https://archive.org/download/archlinux_pkg_cuda if too old.
  2. Install CUDA & CUDNN: pacman -U cuda-X-Y cudnn.
  3. Rename /opt/cuda to /opt/cuda-X.Y: mv /opt/cuda /opt/cuda-X.Y or cp /opt/cuda /opt/cuda-X.Y
  4. Rename /etc/ld.so.conf.d/cuda.conf to /etc/ld.so.conf.d/cuda-X.Y.conf, and change the content of the file to point to correct location.
  5. Uninstall CUDA & CUDNN: pacman -R cudnn cuda (first remove cudnn).
  1. (Optional) Remove file /etc/profile.d/cuda.sh. if it still exists.

Repeat steps 1-5 to install all versions you need. At the end, if you want to have the latest version that will receive regular updates, you can install it with pacman -S cuda cudnn. pacman will update only that version. After system reboot, you should be able to activate X.Y version, simply typing in terminal set-cuda-X-Y.

About

Multiple Version of CUDA Libraries On The Same Machine