Have you tried removing all that stuff?
I’m talking about:
- driver: nvidia
- /dev/nvidia-caps:/dev/nvidia-caps
- /dev/nvidia0:/dev/nvidia0
- /dev/nvidiactl:/dev/nvidiactl
- /dev/nvidia-modeset:/dev/nvidia-modeset
- /dev/nvidia-uvm:/dev/nvidia-uvm
- /dev/nvidia-uvm-tools:/dev/nvidia-uvm-tools
count: all
And just leave it with:
deploy:
resources:
reservations:
devices:
count: all
capabilities: [gpu]
Finally, make sure you have the cuda docker toolkit installed.
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html
Make sure to then test with:
sudo docker run --rm --runtime=nvidia --gpus all ubuntu nvidia-smi
As per:
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/sample-workload.html
EDIT: NVM see you fixed it! Could you post your final docker-compose.yml for reference? I would like to dive deep and figure out why some people need to add more to their docker-compose.yml and some don’t.
I’m a zillenial (26 yo) here and I’d like to know too.
I might also be up for making such communities from scratch, what sort of stuff would you be interested in?
I should say though I don’t feel like an outcast. But then again I’ve literally never used Facebook, Twitter/X/Bluesky/Mastodon or TikTok, only used insta a bit. I grew up on 4chan before the chuds fully took over so my background is a bit different.