Over the past couple of months, I have used my minimal VMware Cloud Foundation (VCF) 9.0 lab to explore nearly every major VCF 9.0 capability including NVMe Tiering, vSAN ESA, NSX VPC, vSphere Supervisor Services like vSphere Kubernetes Service (VKS), VM Service & ArgoCD, VCF Operations (VCFO), VCF Automation (VCFA), VMware AVI Load Balancer, Data Services Manager (DSM) and the list goes on and on! š
One area that I have been interested in diving deeper into is with our Private AI offerings and with the announcement of VMware Private AI Services (PAIS) now included as part of VMware Cloud Foundation (VCF), it gave me that additional motivation to see if I could also run PAIS in my lab!
As part of my exploration, I will be documenting my experience in a mini-blog series on installing and configuring PAIS to deploy an AI Agent using an NVIDIA GPU with DirectPath I/O. Since NVIDIA vGPU requires an NVIDIA AI for Enterprise (NVAIE) license, this approach can especially be useful for lab and learning purposes.
Requirements:
- VCF 9.0 environment deployed
- NSX VPC configured with Centralized Transit Gateway
- vSphere Supervisor configured with NSX VPC
- VCF Automation configured to support vSphere Supervisor with (All Apps) Tenant Portal
Step 1 - Login to the Broadcom Support Portal (BSP) to download PAIS and the registry download token.
Navigate to My Downloads and search for "Private AI Services" (direct download link) and select the latest version to download the deployment manifest (pais.yml) file.
Next to the download link, you should see a green icon, click on that to retrieve the registry information along with your unique download token.
Note: If you need to deploy PAIS in an air-gapped environment, you can follow this blog post which outlines the instructions for transporting the required container images into your own private container registry.
Step 2 - Before we can install the PAIS Supervisor Service from Broadcom's container registry, we need to add the registry information and download token into our vSphere Supervisor Cluster. Navigate to your Supervisor Cluster->Configure->Container Registries and click Add to begin the workflow.
Step 3 - Now, we can register the PAIS Supervisor Service by navigating to your Supervisor Cluster->ServicesĀ and click Add to begin the workflow. Upload the pais.yml file that you had downloaded from Step 1 to register the new Supervisor Service.
Step 4 - To install the PAIS Supervisor Service, navigate to your Supervisor Cluster->Services->Private AI ServicesĀ and under actions, select Manage Service and leave the defaults and click next to begin the installation.
The installation will create a new vSphere Namespace along with a vSphere Pod that represents the PAIS Supervisor Service and should only take a few minutes. You can verify that everything was setup correctly by navigating to your Supervisor Cluster->Configure->Supervisor Services->Overview->InstalledĀ and ensure the status is green as shown in the screenshot below.
Pretty easy to install PAIS, right?
In the next blog post, we will cover the installation and configuration of Data Services Manager (DSM), which is a requirement for providing a Vector Database using Postgres, so stay tuned!
Thanks for the comment!