GPU Operator

1. Review the system requirements of the add-on to make sure you have enough resources to install it.
2. (Optional) Perform manual scaling of worker node groups or set up automatic scaling.

The GPU Operator addon works on worker nodes with GPU, so only installation on dedicated nodes is available for it. To be able to add worker nodes with GPU to the cluster, connect the Cloud GPU service.

1. Prepare a dedicated group of worker nodes for installing the add-on, if not done so already:

   Management console

   1. Go to your VK Cloud management console.

   2. Select the project where the required cluster is located.

   3. Go to Kubernetes Clusters → Kubernetes Clusters.

   4. Click the name of the cluster.

   5. Make sure that the cluster has a dedicated group of worker nodes with GPUs on which addons will be hosted.

      If there is no such group, add it.

   6. (Optional) If nodes with GPUs should only run processes that require GPU resources, set a taint for this node group:

      • effect NoSchedule
      • key nvidia.com
      • value gpu

2. Install the add-on:

   Management console

   1. Go to your VK Cloud management console.

   2. Select the project where the required cluster is located.

   3. Go to Kubernetes Clusters → Kubernetes Clusters.

   4. Click the name of the cluster.

   5. Go to the Addons tab.

   6. Click the Install button on the gpu-operator add-on card, then click Install addon.

   7. (Optional) Edit the following:

      • the selected version
      • the application name
      • the namespace where the add-on will be installed
      • the add-on settings code

   8. Click the Install addon button.

      The installation of the add-on in the cluster will begin. This process can take a long time.

3. (Optional) Read the official NVIDIA documentation about working with the add-on.

You can find the full add-on settings code along with the description of the fields on GitHub.

Once done with editing the add-on code, continue with the installation process.