Managing multiple CUDA versions using environment modules in Ubuntu

Steps_multiple_cuda_environments.md

Steps to manage multiple CUDA environments

Latest Update: May 19th, 2024

This gist contains all the steps required to:

• Install multiple CUDA versions (e.g., CUDA 11.8 and CUDA 12.1
• Manage multiple CUDA environments on Ubuntu using the utility called environment modules.
• Use this approach to avoid CUDA environment conflicts.

Environment Modules is a package that provides for the dynamic modification of a user's environment via modulefiles. You can find more on it at https://modules.readthedocs.io/en/latest/

1. Install the Compatible NVIDIA Drivers (if required)

• Add PPA GPU Drivers Repository to the System

  sudo add-apt-repository ppa:graphics-drivers/ppa

• Check GPU and available drives

  ubuntu-drivers devices
  # install it using: sudo ubuntu-drivers

• Install the compatible driver

  # best to allow Ubuntu to autodetect and install the compatible nvidia-driver
  sudo ubuntu-drivers install

  For example, I tried to install nvidia-driver-545 using sudo ubuntu-drivers install nvidia:545 command. However, I was unable to install it. There was always some or the other issue.

  Note: Please restart your system after installing the nvidia driver. Ideally, you should be able to get GPU state and stats using nvidia-smi

• Check the installed NVIDIA driver

  nvidia-detector 

• Additionally, you can also install NVIDIA drivers using the Software & Updates Ubuntu app. Just go to the Additional Drivers tab, choose a driver, and click Apply Changes.

2. Install CUDA 11.8 and CUDA 12.1

• Go to the https://developer.nvidia.com/cuda-toolkit-archive and select CUDA Toolkit 11.8 from the available options.

• Choose your OS, architecture, distribution, version, and installer type. For example, in my case:

  Option	value
  OS	Linux
  Architecture	x86_64
  Distribution	Linux
  Version	22.04
  Installer type	deb(local)

• Follow the provided installation instructions by copying and pasting the commands into your terminal. This will install CUDA 11.8. Use the following commands:

  wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-ubuntu2204.pin
  sudo mv cuda-ubuntu2204.pin /etc/apt/preferences.d/cuda-repository-pin-600  
  wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda-repo-ubuntu2204-11-8-local_11.8.0-520.61.05-1_amd64.deb
  sudo dpkg -i cuda-repo-ubuntu2204-11-8-local_11.8.0-520.61.05-1_amd64.deb
  sudo cp /var/cuda-repo-ubuntu2204-11-8-local/cuda-*-keyring.gpg /usr/share/keyrings/
  sudo apt-get update
  sudo apt-get -y install cuda

• Similarly, install CUDA 12.1 using the following commands:

  wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-ubuntu2204.pin
  sudo mv cuda-ubuntu2204.pin /etc/apt/preferences.d/cuda-repository-pin-600
  wget https://developer.download.nvidia.com/compute/cuda/12.1.0/local_installers/cuda-repo-ubuntu2204-12-1-local_12.1.0-530.30.02-1_amd64.deb
  sudo dpkg -i cuda-repo-ubuntu2204-12-1-local_12.1.0-530.30.02-1_amd64.deb
  sudo cp /var/cuda-repo-ubuntu2204-12-1-local/cuda-*-keyring.gpg /usr/share/keyrings/
  sudo apt-get update
  sudo apt-get -y install cuda

• Make sure to copy and execute the commands above in your terminal to install CUDA 11.8 and CUDA 12.1 on your system.

3. Install cuDNN library

• Go to https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/ and download the cuDNN tar for CUDA 11.x. Note that you might need to create a developer's account first.

• Untar the downloaded file using the following command:

  tar -xvf cudnn-linux-x86_64-9.1.0.70_cuda11-archive.tar.xz # CUDA 11.x
  tar -xvf cudnn-linux-x86_64-9.1.0.70_cuda12-archive.tar.xz # CUDA 12.x
  

• Copy the cuDNN files to the CUDA toolkit files:

  # for CUDA 11.8
  cd cudnn-linux-x86_64-9.1.0.70_cuda11-archive/
  sudo cp include/cudnn*.h /usr/local/cuda-11.8/include
  sudo cp lib64/libcudnn* /usr/local/cuda-11.8/lib64
  
  # for CUDA 12.1
  cd cudnn-linux-x86_64-9.1.0.70_cuda12-archive/
  sudo cp include/cudnn*.h /usr/local/cuda-12.1/include
  sudo cp lib64/libcudnn* /usr/local/cuda-12.1/lib64

• Make the files executable:

  sudo chmod a+r /usr/local/cuda-11.8/include/cudnn*.h /usr/local/cuda-11.8/lib64/libcudnn*
  sudo chmod a+r /usr/local/cuda-12.1/include/cudnn*.h /usr/local/cuda-12.1/lib64/libcudnn*

Note: Strictly speaking, you are done with the CUDA setup. You can use it by adding the CUDA bin and library path to the PATH and LD_LIBRARY_PATH environment variables. For example, you can set up CUDA 11.8 by adding the following lines in the ~/.bashrc:

PATH=/usr/local/cuda-11.8/bin:$PATH
LD_LIBRARY_PATH=/usr/local/cuda-11.8/extras/CUPTI/lib64:$LD_LIBRARY_PATH
LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:$LD_LIBRARY_PATH

Similarly, you can set up CUDA 12.1. However, manually changing the paths every time can be cumbersome!

Note: In case, you only want to install either of the one, CUDNN 11.x or CUDNN 12.x. The simpler way is to go to https://developer.nvidia.com/cudnn-downloads and install the CUDNN 11.x or CUDNN 12.x similar to CUDA installation.

4. Manage multiple CUDA versions using environment modules

a) Install the environment modules utility:

• Run the following commands:

      sudo apt-get update
      sudo apt-get install environment-modules

• Check the installation:

  # Check the installation by running
  module list

You should see a list of default installed modules like git and maybe their versions displayed when you run the command module list. This confirms that the environment modules utility has been successfully installed on your system.

b) Create modulefiles for CUDA distributions

Note: You might need root permissions to create directories and files. Use sudo in that case.

• Create a directory /usr/share/modules/modulefiles/cuda to hold modulefiles for cuda distributions

  sudo mkdir -p /usr/share/modules/modulefiles/cuda

• Create a modulefile /usr/share/modules/modulefiles/cuda/11.8 for CUDA 11.8 and add the following lines:

  #%Module1.0
  ##
  ## cuda 11.8 modulefile
  ##
  
  proc ModulesHelp { } {
      global version
      
      puts stderr "\tSets up environment for CUDA $version\n"
  }
  
  module-whatis "sets up environment for CUDA 11.8"
  
  if { [ is-loaded cuda/12.1 ] } {
  module unload cuda/12.1
  }
  
  set version 11.8
  set root /usr/local/cuda-11.8
  setenv CUDA_HOME	$root
  
  prepend-path PATH $root/bin
  prepend-path LD_LIBRARY_PATH $root/extras/CUPTI/lib64
  prepend-path LD_LIBRARY_PATH $root/lib64
  conflict cuda

• Similarly, create a modulefile /usr/share/modules/modulefiles/cuda/12.1 for CUDA 12.1 and add the following lines:

  #%Module1.0
  ##
  ## cuda 12.1 modulefile
  ##
  
  proc ModulesHelp { } {
      global version
      
      puts stderr "\tSets up environment for CUDA $version\n"
  }
  
  module-whatis "sets up environment for CUDA 12.1"
  
  if { [ is-loaded cuda/11.8 ] } {
  module unload cuda/11.8
  }
  
  set version 12.1
  set root /usr/local/cuda-12.1
  setenv CUDA_HOME	$root
  
  prepend-path PATH $root/bin
  prepend-path LD_LIBRARY_PATH $root/extras/CUPTI/lib64
  prepend-path LD_LIBRARY_PATH $root/lib64
  conflict cuda

c) Make CUDA 11.8 the default cuda version

• Create a file /usr/share/modules/modulefiles/cuda.version to make CUDA 11.8 the default cuda module:

  #%Module
  set ModulesVersion 11.8

Note: make sure to reload your terminal.

5. Changing and Viewing the CUDA Module

• To change and view the loaded CUDA module, you can use the following commands:

  # Check the currently loaded module
  module list
  # Check the available modules
  module avail
  
  # Load a specific cuda version
  module load cuda/12.1
  # Unload the currently loaded CUDA module
  module unload cuda
  # Load CUDA 11.8
  module load cuda/11.8
  
  # verify the paths of the loaded CUDA
  nvcc --version # should give the loaded CUDA version
  echo $CUDA_HOME
  echo $PATH
  echo $LD_LIBRARY_PATH

Note: You can add additional CUDA versions or other packages by creating corresponding modulefiles and following the steps outlined in this gist.

6. Some Useful Tips

a) nvidia-smi does not works

• Sometime, after Ubuntu update or some other weird issue. The system might not be able to detect drivers. For example, you get erros such as nvidia-smi has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running. The best solution is to remove the current drivers and reinstall the compatible nvidia-driver.

  # removes all the nvidia drivers
  sudo apt-get --purge remove "*nvidia*" "libxnvctrl*"
  # reinstall the compatible driver and restart
  sudo ubuntu-drivers install

b) Purge CUDA from your computer

DO IT AT YOUR OWN RISK

```bash
# removes all the nvidia drivers
sudo apt-get --purge remove "*nvidia*" "libxnvctrl*"
# remove all cuda versions
sudo apt-get --purge remove "*cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*"  "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
# remove all cuda folders
sudo rm -rf /usr/loca/cuda*
```

Resources and helpful links

• https://ubuntu.com/server/docs/nvidia-drivers-installation
• https://developer.nvidia.com/cuda-toolkit-archive
• https://developer.nvidia.com/cudnn-downloads

Thank you for providing these instructions. I am having the same issue as @mizeller (I have executed all the steps prior to that). It defaults to the latest version which is 12.5 now. This is probably because I have a later version (12.1) already installed on my system and I am now trying to install 11.8

I have tried the method @mizeller mentioned (specify version explicitly such as cuda-11-8) and it worked.

Unfortunately I hit another roadblock with environment modules. After following the steps (install environment modules and adding cuda modulefiles), module avail does not show the added cuda files. It only shows the default ones like dot, module-git, etc. Running module load cuda/11.8 will result in ERROR: Unable to locate a modulefile for 'cuda/11.8'.

Note that this is a problem with environment modules and not the general solution mentioned here.

Thanks for grate guide! What is difference from managing multiple CUDA version using sudo update-alternatives --config cuda?

@cieske I don't use update-alternatives because it requires setting symbolic links, which I try to avoid. Initially, it can be a bit of too much work. Moreover, I am more used to managing multiple packages using module (a bit old school :-)).

If you are familiar with conda, you can use it to manage multiple CUDA environments. It is a lot simpler than these approaches. There is a great video by Jeremy explaining how you can install and manage different CUDA versions using conda: Getting Started With CUDA for Python Programmers.

@omsrisagar Defaulting to latest current version might happen, if you don't download the correct CUDA version package and set current keyring in Step 2: Install CUDA 11.8 and CUDA 12.1. I too agree, better to mention the package explicitly to be on safer side.

@omsrisagar ERROR: Unable to locate a modulefile for 'cuda/11.8' is most likely due to incorrect cuda version naming in module files. Best to take chatGPT help ;).