Example configurations for using containerized Pixi environments with HTCondor and GPUs

These examples are inspired by the Center for High Throughput Computing examples, the pixi-docker project, and Pavel Zwerschke's pixi-docker-examples.

CUDA enabled Docker images

These examples assume that you want to use GPU resources to take advantage of hardware acceleration and so focus on using the Pixi Docker base images built on the NVIDIA CUDA enabled images for runtime use with the the NVIDIA Container Toolkit.

Local installation

• Make sure that you have the nvidia-container-toolkit installed on the host machine
• Check the list of available tags on Github Container Registry to find the tag you want

Tip

You can use crane to do this from the command line as well

$ pixi global install crane
$ crane ls ghcr.io/prefix-dev/pixi

• Use docker pull to pull down the image corresponding to the tag

Example:

docker pull ghcr.io/prefix-dev/pixi:0.45.0-jammy-cuda-12.6.3

Local use

To check that NVIDIA GPUS are being properly detected run

docker run --rm --gpus all ghcr.io/prefix-dev/pixi:0.45.0-jammy-cuda-12.6.3 'nvidia-smi'

and check if the nvidia-smi output appears correctly.

To run (interactively) using GPUs on the host machine:

docker run --rm -ti --gpus all ghcr.io/prefix-dev/pixi:0.45.0-jammy-cuda-12.6.3