white water boat

Nvidia Container Toolkit and Podman on Ubuntu 20.04

I did not immediately find a good tutorial on how to use the Nvidia Container Toolkit with Podman to be able to run GPU-accelerated containers. The solution turned out to be quite straightforward.

First, you can install Podman as usual:

sudo apt-get install -y curl wget gnupg2
source /etc/os-release
sudo sh -c "echo 'deb http://download.opensuse.org/repositories/devel:/kubic:/libcontainers:/stable/xUbuntu_${VERSION_ID}/ /' > /etc/apt/sources.list.d/devel:kubic:libcontainers:stable.list"
wget -nv https://download.opensuse.org/repositories/devel:kubic:libcontainers:stable/xUbuntu_${VERSION_ID}/Release.key -O- | sudo apt-key add -
sudo apt-get update && sudo apt-get install -y podman
podman -v

Afterwards, you can install the Nvidia Container Toolkit as usual:

distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
      && curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
      && curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
            sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
            sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

sudo apt-get update
sudo apt-get install -y nvidia-container-toolkit

However, before you can start using both together you need to manually define the OCI hook:

sudo mkdir -p /usr/share/containers/oci/hooks.d/

cat << EOF | sudo tee /usr/share/containers/oci/hooks.d/oci-nvidia-hook.json
{
    "version": "1.0.0",
    "hook": {
        "path": "/usr/bin/nvidia-container-toolkit",
        "args": ["nvidia-container-toolkit", "prestart"],
        "env": [
            "PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
        ]
    },
    "when": {
        "always": true,
        "commands": [".*"]
    },
    "stages": ["prestart"]
}
EOF

sudo systemctl restart podman

Once this is done you should be able to run GPU-accelerated images via Podman the way you are familiar with from nvidia-docker:

sudo podman run --rm --gpus all nvidia/cuda:11.6.0-base-ubuntu20.04 nvidia-smi

Finally, please note that if you should hit an error like

Error: OCI runtime error: error executing hook `/usr/bin/nvidia-container-toolkit` (exit code: 1)

that most likely indicates that your driver is not new enough. You can take a look at your card’s CUDA capability and try a driver upgrade. For instance, on my current machine I have two GTX 2080 SUPER with driver 510.73.08 and that means I can run CUDA toolkit 11.6, but not 11.7. If you were on a 470 driver, this number should be lower and you might be able to run 11.4, but not 11.5.

0 replies

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply