Skip to content

Provision and configure a Google Kubernetes Engine platform for NVIDIA Omniverse applications

This document describes how to provision and configure a Google Kubernetes Engine (GKE) cluster to deploy NVIDIA Omniverse applications on Google Cloud.

The provisioning and configuration process follows GKE security and operational best practices.

Provision and configure a GKE cluster

To provision and configure a GKE cluster that implements best practices, you deploy an instance of the Core GKE Accelerated Platform.

To deploy an instance of the Core GKE Accelerated Platform, you do the following:

  1. Clone the Accelerated Platforms repository and set the repository directory environment variable:

    git clone https://github.com/GoogleCloudPlatform/accelerated-platforms && \
cd accelerated-platforms && \
export ACP_REPO_DIR="$(pwd)"

  2. Configure the Core GKE Accelerated Platform by adding setting the following Terraform configuration variables in platforms/gke/base/_shared_config/initialize.auto.tfvars:

    # Disable initializing node pools without GPUs
initialize_container_node_pools_cpu             = false
# Disable initializing node pools with GPUs that don't offer NVIDIA RTX cores
initialize_container_node_pools_gpu_without_rtx = false
# Disable initializing node pools with Google TPUs because they are not needed for this workload
initialize_container_node_pools_tpu             = false

  3. Deploy an instance of the Core GKE Accelerated Platform:

    CORE_TERRASERVICES_APPLY="networking container_cluster container_node_pool workloads/cluster_credentials" "${ACP_REPO_DIR}/platforms/gke/base/core/deploy.sh"

Note: the instructions in this section are a specialization of the Core GKE Accelerated Platform deployment instructions that are tailored to this use case.

Destroy the GKE cluster

To destroy the instance of the Core GKE Accelerated Platform you created, you do the following:

  1. Destroy the instance of the Core GKE Accelerated Platform:

    CORE_TERRASERVICES_DESTROY="workloads/cluster_credentials container_node_pool container_cluster networking" "${ACP_REPO_DIR}/platforms/gke/base/core/teardown.sh"

What's next