Installing the Nvidia datacenter drivers on AL2023

[openmarmot]

[edit] - see the reply to this with the official (and much better) solution.

In case anyone is struggling with the Nvidia driver installs for AL2023, I recently did the work to get the drivers working for the P3 series (Tesla V100)

          # install pre-reqs
          dnf install -y kernel-devel-$(uname -r) kernel-headers-$(uname -r) kernel-modules-extra
          
          # silent install of nvidia datacenter drivers for the p3 series
          BASE_URL=https://us.download.nvidia.com/tesla
          DRIVER_VERSION=550.54.14
          curl -fSsl -O $BASE_URL/$DRIVER_VERSION/NVIDIA-Linux-x86_64-$DRIVER_VERSION.run
          chmod +x NVIDIA-Linux-x86_64-$DRIVER_VERSION.run
          ./NVIDIA-Linux-x86_64-$DRIVER_VERSION.run --tmpdir . --silent

If you want a full cloudformation template - I have created some examples here :
https://github.com/openmarmot/aws-ec2-nvidia-drivers

[limmike]

Thanks for sharing.
NVIDIA has added repository support for AL2023. This makes it convenient to update NVIDIA GPU drivers and CUDA toolkit.

Step-by-step instructions available at re:Post

[openmarmot]

Awesome thanks!
This is superior to what I was doing manually. I have tested and confirmed that it works on a p3.2xlarge

[elsaco]

There's a typo in the instructions at https://docs.nvidia.com/cuda/cuda-installation-guide-linux/#network-repo-installation-for-amazon-linux. Distro name should be amzn2023 and not amz2023

Network install instructions from the online doc:

sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/amz2023/x86_64/cuda-amz2023.repo
will result in a 404 error:

[ec2-user@amazon1 ~]$ sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/amz2023/x86_64/cuda-amz2023.repo
Adding repo from: https://developer.download.nvidia.com/compute/cuda/repos/amz2023/x86_64/cuda-amz2023.repo
Status code: 404 for https://developer.download.nvidia.com/compute/cuda/repos/amz2023/x86_64/cuda-amz2023.repo (IP: 152.195.19.142)
Error: Configuration of repo failed

[limmike]

The article at Repost.aws uses the correct URI.

As typo is on their documentation site , you can submit feedback to NVIDIA. The Send feedback link points to mailto:CUDAIssues@nvidia.com?subject=CUDA Toolkit Documentation Feedback.

I have emailed NVIDIA about the typo.

[limmike]

NVIDIA has updated AL2023 documentation for CUDA 12.5 Update 1