GPU-enabled clusters
Overview
With CAPZ you can create GPU-enabled Kubernetes clusters on Microsoft Azure.
Before you begin, be aware that:
- Scheduling GPUs is a Kubernetes beta feature
- NVIDIA GPUs are supported on Azure NC-series, NV-series, and NVv3-series VMs
- NVIDIA GPU Operator allows administrators of Kubernetes clusters to manage GPU nodes just like CPU nodes in the cluster.
To deploy a cluster with support for GPU nodes, use the nvidia-gpu flavor.
An example GPU cluster
Let’s create a CAPZ cluster with an N-series node and run a GPU-powered vector calculation.
Generate an nvidia-gpu cluster template
Use the clusterctl generate cluster command to generate a manifest that defines your GPU-enabled
workload cluster.
Remember to use the nvidia-gpu flavor with N-series nodes.
AZURE_CONTROL_PLANE_MACHINE_TYPE=Standard_B2s \
AZURE_NODE_MACHINE_TYPE=Standard_NC6s_v3 \
AZURE_LOCATION=southcentralus \
clusterctl generate cluster azure-gpu \
--kubernetes-version=v1.22.1 \
--worker-machine-count=1 \
--flavor=nvidia-gpu > azure-gpu-cluster.yaml
Create the cluster
Apply the manifest from the previous step to your management cluster to have CAPZ create a workload cluster:
$ kubectl apply -f azure-gpu-cluster.yaml
cluster.cluster.x-k8s.io/azure-gpu serverside-applied
azurecluster.infrastructure.cluster.x-k8s.io/azure-gpu serverside-applied
kubeadmcontrolplane.controlplane.cluster.x-k8s.io/azure-gpu-control-plane serverside-applied
azuremachinetemplate.infrastructure.cluster.x-k8s.io/azure-gpu-control-plane serverside-applied
machinedeployment.cluster.x-k8s.io/azure-gpu-md-0 serverside-applied
azuremachinetemplate.infrastructure.cluster.x-k8s.io/azure-gpu-md-0 serverside-applied
kubeadmconfigtemplate.bootstrap.cluster.x-k8s.io/azure-gpu-md-0 serverside-applied
Wait until the cluster and nodes are finished provisioning...
$ kubectl get cluster azure-gpu
NAME PHASE
azure-gpu Provisioned
$ kubectl get machines
NAME PROVIDERID PHASE VERSION
azure-gpu-control-plane-t94nm azure:////subscriptions/<subscription_id>/resourceGroups/azure-gpu/providers/Microsoft.Compute/virtualMachines/azure-gpu-control-plane-nnb57 Running v1.22.1
azure-gpu-md-0-f6b88dd78-vmkph azure:////subscriptions/<subscription_id>/resourceGroups/azure-gpu/providers/Microsoft.Compute/virtualMachines/azure-gpu-md-0-gcc8v Running v1.22.1
... and then you can install a CNI of your choice.
Once all nodes are Ready, install the official NVIDIA gpu-operator via Helm.
Install nvidia gpu-operator Helm chart
If you don’t have helm, installation instructions for your environment can be found here.
First, grab the kubeconfig from your newly created cluster and save it to a file:
$ clusterctl get kubeconfig azure-gpu > ./azure-gpu-cluster.conf
Now we can use Helm to install the official chart:
$ helm install --kubeconfig ./azure-gpu-cluster.conf --repo https://helm.ngc.nvidia.com/nvidia gpu-operator --generate-name
The installation of GPU drivers via gpu-operator will take several minutes. Coffee or tea may be appropriate at this time.
After a time, you may run the following command against the workload cluster to check if all the gpu-operator resources are installed:
$ kubectl --kubeconfig ./azure-gpu-cluster.conf get pods -o wide | grep 'gpu\|nvidia'
NAMESPACE NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
default gpu-feature-discovery-r6zgh 1/1 Running 0 7m21s 192.168.132.75 azure-gpu-md-0-gcc8v <none> <none>
default gpu-operator-1674686292-node-feature-discovery-master-79d8pbcg6 1/1 Running 0 8m15s 192.168.96.7 azure-gpu-control-plane-nnb57 <none> <none>
default gpu-operator-1674686292-node-feature-discovery-worker-g9dj2 1/1 Running 0 8m15s 192.168.132.66 gpu-md-0-gcc8v <none> <none>
default gpu-operator-95b545d6f-rmlf2 1/1 Running 0 8m15s 192.168.132.67 gpu-md-0-gcc8v <none> <none>
default nvidia-container-toolkit-daemonset-hstgw 1/1 Running 0 7m21s 192.168.132.70 gpu-md-0-gcc8v <none> <none>
default nvidia-cuda-validator-pdmkl 0/1 Completed 0 3m47s 192.168.132.74 azure-gpu-md-0-gcc8v <none> <none>
default nvidia-dcgm-exporter-wjm7p 1/1 Running 0 7m21s 192.168.132.71 azure-gpu-md-0-gcc8v <none> <none>
default nvidia-device-plugin-daemonset-csv6k 1/1 Running 0 7m21s 192.168.132.73 azure-gpu-md-0-gcc8v <none> <none>
default nvidia-device-plugin-validator-gxzt2 0/1 Completed 0 2m49s 192.168.132.76 azure-gpu-md-0-gcc8v <none> <none>
default nvidia-driver-daemonset-zww52 1/1 Running 0 7m46s 192.168.132.68 azure-gpu-md-0-gcc8v <none> <none>
default nvidia-operator-validator-kjr6m 1/1 Running 0 7m21s 192.168.132.72 azure-gpu-md-0-gcc8v <none> <none>
You should see all pods in either a state of Running or Completed. If that is the case, then you know the driver installation and GPU node configuration is successful.
Then run the following commands against the workload cluster to verify that the
NVIDIA device plugin
has initialized and the nvidia.com/gpu resource is available:
$ kubectl --kubeconfig ./azure-gpu-cluster.conf get nodes
NAME STATUS ROLES AGE VERSION
azure-gpu-control-plane-nnb57 Ready master 42m v1.22.1
azure-gpu-md-0-gcc8v Ready <none> 38m v1.22.1
$ kubectl --kubeconfig ./azure-gpu-cluster.conf get node azure-gpu-md-0-gcc8v -o jsonpath={.status.allocatable} | jq
{
"attachable-volumes-azure-disk": "12",
"cpu": "6",
"ephemeral-storage": "119716326407",
"hugepages-1Gi": "0",
"hugepages-2Mi": "0",
"memory": "115312060Ki",
"nvidia.com/gpu": "1",
"pods": "110"
}
Run a test app
Let’s create a pod manifest for the cuda-vector-add example from the Kubernetes documentation and
deploy it:
$ cat > cuda-vector-add.yaml << EOF
apiVersion: v1
kind: Pod
metadata:
name: cuda-vector-add
spec:
restartPolicy: OnFailure
containers:
- name: cuda-vector-add
# https://github.com/kubernetes/kubernetes/blob/v1.7.11/test/images/nvidia-cuda/Dockerfile
image: "registry.k8s.io/cuda-vector-add:v0.1"
resources:
limits:
nvidia.com/gpu: 1 # requesting 1 GPU
EOF
$ kubectl --kubeconfig ./azure-gpu-cluster.conf apply -f cuda-vector-add.yaml
The container will download, run, and perform a CUDA calculation with the GPU.
$ kubectl get po cuda-vector-add
cuda-vector-add 0/1 Completed 0 91s
$ kubectl logs cuda-vector-add
[Vector addition of 50000 elements]
Copy input data from the host memory to the CUDA device
CUDA kernel launch with 196 blocks of 256 threads
Copy output data from the CUDA device to the host memory
Test PASSED
Done
If you see output like the above, your GPU cluster is working!