Skip to content

Chapter 1: Operating Systems, Management Platforms, and First Power-On · Part 2 of 9

The software stack -- what runs where

Jump to section

1. Before you touch anything2. The software stack -- what runs where3. The management philosophy -- centralised vs distributed4. Is it all CLI? Or all orchestrated?5. The platforms you need to know -- and why NVIDIA dominates6. First power-on -- what actually happens7. First access -- what you expect to see8. How the connectivity looks -- the full picture9. What you now know

Every piece of hardware in an AI cluster runs its own software stack. These are not interchangeable. Understanding which OS lives on which device is the foundation of everything.

What runs where — click each device

DGX H100

Ubuntu 22.04 LTS (NVIDIA-customised)

Compute node

DGX OS is a customised Ubuntu image with NVIDIA drivers, CUDA, NCCL, container runtime, and management agents pre-installed. Underneath it is standard Ubuntu — you can SSH in, run apt, use systemctl. NVIDIA adds nvsm, dcgm, and the fabric manager daemon.

How to access

SSH to management IP, or iDRAC virtual console for boot-time access

First commands
nvidia-smi  →  verify all 8 GPUs are visible
ibstat      →  verify all 8 ConnectX-7 NICs are active
Note: Each DGX also has a separate iDRAC BMC accessible on a dedicated 1GbE port — completely independent of the host OS.

Select each device in the visualisation above to see its OS, role, and management surface. The four stacks -- DGX OS, ONYX, Cumulus Linux, and UFM -- are what every CLI command in this curriculum assumes you are running. (The DGX B200 platform additionally runs BlueField-3 firmware on its storage adapters, but that is not covered in this foundational chapter.) The next sections cover how they are managed and accessed.

22% complete

Previous

Before you touch anything

Next

The management philosophy -- centralised vs distributed