GPU enabled image about devpod HOT 6 OPEN

yup we do, so it should assign the gpu to the pod!
Testing out now with the devcontainer setup you provided :)

we were initially using a setup.sh and init.sh file in .devcontainer so i've switched it to follow the setup provided and added in docker features as we're going to use docker compose/build to run our repo

from devpod.

Hi @raouln97

If you already have a machine with GPU support, then you can use a devcontainer setup like this one: https://github.com/psaboia/devcontainer-nvidia-base in the repo, and then use devpod for that repo

from devpod.

we're actually trying to use a pod_manifest_template to run a gpu in the pod

something like this:

apiVersion: v1
kind: Pod
metadata:
name: raouls-gpu-test
spec:
priorityClassName: gpu-apps
containers:
- name: raouls-gpu-container
image: ghcr.io/loft-sh/dockerless:0.1.4
resources:
requests:
cpu: "500m"
memory: "5Gi"
nvidia.com/gpu: "1" # Requesting 1 NVIDIA GPU
limits:
cpu: "1000m"
memory: "8Gi"
nvidia.com/gpu: "1" # Limiting to 1 NVIDIA GPU
tolerations:
- key: "sku"
operator: "Equal"
value: "gpu"
effect: "NoSchedule"
volumeMounts:
- mountPath: /workspaces/.home
name: devpod
subPath: devpod/home

from devpod.

if you already have the gpu-operator already in place, this should work, at most I'd say to add a node affinity to ensure it gets scheduled on the node with the gpus

from devpod.

Hey @89luca89 ,

i tried applying our pod manifest template separately and tested the gpu access with "nvidia-smi" and i could see the availability of the gpu
However, using the nvidia configurations you provided in the .devcontainer folder seems to rebuild the pod without the gpu requirements and hence, gpu can't be accessed inside the devpod container

any ideas how to get around this?

Thanks! :)

from devpod.

I think for the k8s provider to pick up the resourse correctly you need to name the container devpod instead of raouls-gpu-container

from devpod.