NVIDIA Physical AI Reference Architecture Deployment Guide

NVIDIA Enterprise Reference Architecture

• Overview
• Solution Overview
  • Application Layer
    • Application Layer Software
      • RUN.AI
      • OSMO
      • Prometheus
  • Infrastructure Layer
    • Storage
    • Infrastructure Layer Software
      • Host Operating System
      • Container Runtime
      • Kubernetes
      • NVIDIA Driver
      • CUDA Toolkit
      • NVIDIA GPU Operator
      • NVIDIA Network Operator
      • HGX B200 Configuration
      • Additional Components
• Install Operating System on Bare Metal Nodes
  • Accessing BMC Interface — Example of an RTX PRO Server
  • Accessing BMC Interface — HGX B200
  • Provisioning Ubuntu 24.04 via BMC
  • Expanding Root Filesystem (Recommended for GPU Nodes)
    • Expand LVM Root Filesystem (Ubuntu with LVM)
  • Prepare the Future Kubernetes Nodes
    • Update the OS Packages and Disable Sudo Prompt
  • Prepare the Installation Node
• Kubernetes Deployment
  • Deploy Kubernetes with Kubespray
    • Configure the Inventory
    • Execute Kubespray in the Container
  • Longhorn Storage Class Provider
  • GPU Operator Installation
    • Normal Installation Procedure
    • Installation Procedure with a Workaround for the Driver 580.82.07
    • HGX B200 Node
    • Firmware Version
    • Install Essential Networking Modules
  • Kubernetes Node Configuration
    • Node Labeling
    • Ingress Configuration
    • LoadBalancer Configuration
    • Prometheus Install
• Network
  • Driver Installation
  • Update the DPU OS and Firmware
  • Network Operator Installation
    • Define and Apply the NicClusterPolicy
    • Define IP Pools
    • Define SriovNetworks
    • Define the Network Node Policy
    • Spawn a Test DaemonSet to Test the Connectivity
• Run:ai Install and Configuration
  • Run:ai Control Plane Install
  • Run:ai Cluster Setup
  • Run:ai Cluster — Optional Additional Components
    • Distributed Training Install
    • Inference Install
    • Distributed Inference Setup
  • User Access
  • Run:ai Install Complete
• OSMO Helm Charts
  • OSMO Architecture
  • Registering Backend
  • Map Run:ai Project to the osmo-workflows Namespace
• Physical AI Workflow
  • Overall Workflow
  • Source of Truth
  • Setup
  • Execution
  • Pipeline Summary
  • Monitoring and Data Locations
  • References

Notices

• Notices
  • Notice
  • Trademarks
  • Copyright