Add GPU

In the first step to add a GPU to a VM in the Deka GPU Portal Service you need to export the downloaded kubeconfig file via the Deka GPU Portal Service. If you haven't downloaded kubeconfig, you can press this link for further explanation. To export the kubeconfig file on the Linux operating system, you can run the following syntax.

set KUBECONFIG=C:\path\to\your\kube-config.yaml

Meanwhile, if you use the Windows operating system, you can run the following syntax.

export KUBECONFIG="/ path/to/your/kube-config.yaml"

After exporting the KUBECONFIG Variable file, verify the connection to the Kubernetes cluster using the “cluster-info” command.

kubectl cluster-info

Turn off the VM that is being configured via the Deka GPU Service Portal by click the More button and select Turn Off.

The "Turn Off Virtual Machine" window appears. Click the Yes button to continue turning off the VM. After successfully shutting down the VM, run the following syntax to ensure the VM has stopped running.

kubectl get vm

Change the manifest that the VM uses by running the following command, and add the GPU to the devices list.

kubectl edit vm vm-l40s-passthrough

In the specs section, above the disk, add this line.

gpus: 
  - deviceName: nvidia.com/AD102GL_L40S
    name: gpu1

So the vm-l40s-passthrough manifest looks like below.

Restart the VM in use via the Virtual Machine page in the Deka GPU Portal Service by pressing the More button and selecting Turn On.

Run the following command to see a list of NVIDIA devices.

lspci | grep -i nvidia