Skip to content

Latest commit

 

History

History
102 lines (80 loc) · 3.88 KB

quickstart.md

File metadata and controls

102 lines (80 loc) · 3.88 KB

Quickstart

You need for your <tenant_id> number and your credentials, i.e. a kubeconfig file <tenant_id>.kubeconfig to access the remote GPUs infrastructure.

Currently Helm is used to install the agent. Make sure you have helm installed in your workstation and you have admin permissions on your cluster.

  1. Install the Chart:
$ helm repo add clastix https://clastix.github.io/charts
$ helm repo update

$ helm upgrade --install k8sgpu clastix/k8sgpu \
    --namespace kube-system \
    --set "k8sgpuOptions.tenantName=your_tenant_name" \
    --set-file "kubeConfigSecret.content=/path/to/your/gpu-remote-cluster/kubeconfig"
  1. You will end up with a virtual node in your kubernetes cluster:
$ kubectl get nodes
NAME    STATUS ROLES AGE VERSION
k8s.gpu Ready  agent 37s v1.2.2
...
  1. Check your assigned RuntimeClasses:
$ kubectl get runtimeclasses
NAME HANDLER AGE
seeweb-nvidia-2xa6000 nvidia 13m
  1. Usage

Now you can schedule your workloads on the virtual node:

$ cat << EOF | kubectl create -f -
apiVersion: v1
kind: Pod
metadata:
  name: nvidia-smi
spec:
  restartPolicy: OnFailure
  runtimeClassName: seeweb-nvidia-2xa6000
  containers:
  - name: nvidia
    image: nvidia/cuda:12.6.0-base-ubuntu20.04
    command: ["/bin/bash", "-c", "--"]
    args: ["sleep 3600"]
    imagePullPolicy: Always
EOF

The pod is now running on the virtual node and can access the assigned GPU specified by the RuntimeClass:

$ kubectl get pods
NAME        READY   STATUS    RESTARTS      AGE
nvidia-smi  1/1     Running   0             37s

$ kubectl exec nvidia-smi -- nvidia-smi

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15              Driver Version: 550.54.15      CUDA Version: 12.6     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA RTX A6000               On  |   00000000:06:00.0 Off |                  Off |
| 30%   34C    P8             31W /  300W |       1MiB /  49140MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  NVIDIA RTX A6000               On  |   00000000:07:00.0 Off |                  Off |
| 30%   39C    P8             34W /  300W |       1MiB /  49140MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
  1. Uninstall the Chart

Delete all pods running on the virtual node and remove the chart using Helm:

$ helm uninstall k8sgpu --namespace kube-system

The virtual node is then removed from your kubernetes cluster.