Overview

Base Command Manager (BCM) supports deploying operating systems beyond its default Ubuntu image through a custom image import workflow. A compatible image root can be built for any BCM-supported OS, packaged as a tarball, and imported through BCM’s cm-create image tooling. BCM handles the rest: ramdisk creation, Preboot Execution Environment (PXE) boot configuration, disk setup, and image distribution.

This guide walks through the full process using Red Hat Enterprise Linux (RHEL) 9.6 as the example OS. The six-part structure and the BCM-specific steps (Parts 2–4) are the same regardless of distribution. The parts that vary by OS — how the image root is bootstrapped, which package manager is used, and which subscription or repository setup is required — are in Part 1, and are clearly marked as RHEL-specific where applicable.

This guide is organized into six parts:

1. Building the OS image — installing the target OS bare-metal on a DGX node and constructing a BCM-compatible image root

2. Importing the image into BCM — transferring and registering the image on the head node

3. Configuring BCM — assigning the image to a node category and preparing for PXE boot

4. Installing NVIDIA software — installing drivers, and DOCA-OFED inside the image

5. Reconfiguring Slurm — extending the existing Slurm installation to support the RHEL image

6. Validating Slurm — verifying GPU access and container runtime on the provisioned nodes

References

The following reference documents can be helpful during the deployment:

• DGX BasePOD B200 Deployment Guide

• DGX Software for Red Hat Enterprise Linux 9

• NVIDIA DGX B200 System User Guide

• Red Hat Customer Portal

• Installing Red Hat Enterprise Linux on NVIDIA DGX