Hello @Marvin I am starting a worker as follows:

prefect worker start -t docker --pool <pool-name>

I want this worker to utilize the underlying GPU so that when the docker image is ran in this worker, it has access to the cuda files that comes with the gpu

Starting a worker with GPU support in Prefect is not clearly detailed in the documentation excerpts I found. There are general guides on starting a Docker worker, and on cloud-based solutions that might leverage GPUs, but nothing specific to your case. For running Docker containers with GPU support, you typically need to install NVIDIA Docker support and use the

--gpus

flag when starting your container. However, this functionality would need to be integrated into the Prefect worker command, and it's not clear from the documentation if and how Prefect supports this feature directly. Here are the relevant parts of the documentation, but they don't address GPU access specifically: - Starting a worker - Serverless Workers, which mentions Google Vertex AI for GPU support - Managed execution I would advise checking for more specific instructions in the Prefect documentation regarding GPU support or reaching out to the Prefect community or support for assistance. Additionally, examining the command line options for

prefect worker start

or checking for a configuration setting related to GPU access in the Prefect settings could be another avenue to explore.