limitations

• only works for eg compute local commands. We'll add it for remote workloads in the future based on requests.
• only available on linux hosts. darwin is planned but we dont have the hardware to test. if you're interested file a github issue and we'll give guidance.
• amd support only currently. if you need nvidia support file a github issue and we'll walk you through getting the information we'd need to make it seamless.

Linux host setup

enabling gpu access requires populating /etc/cdi. and

why this needs manual setup is beyond our perview since all the information in the file can be read from userspace.

note: nvidia isnt hardware we have on hand. process will be similar however.

# on arch linux:
sudo pacman -S amd-container-toolkit
sudo amd-ctk cdi generate --output=/etc/cdi/amd.json

sudo apt update
sudo apt install amd-container-toolkit
sudo amd-ctk cdi generate --output=/etc/cdi/amd.json

# run a workload that requires the gpu
eg compute local --gpu test/workload