NVIDIA DGX SuperPOD B300 Quantum-X800 InfiniBand Reference Architecture

Type: Reference Architecture Tags: NVIDIA, DGX SuperPOD, DGX B300, Quantum-X800, InfiniBand, XDR, AC Power, Blackwell Ultra, Mission Control, AI factory Related: NVIDIA-DGX-SuperPOD, NVIDIA-DGX-B300, NVIDIA-DGX-B200, NVIDIA-Blackwell-Architecture, NVIDIA-Quantum-X800-InfiniBand, NVIDIA-Quantum-InfiniBand, NVIDIA-Spectrum-X, NVIDIA-Mission-Control, NVIDIA-AI-Enterprise, NGC, NVIDIA-UFM, NVIDIA-Certified-Storage, NVIDIA-Enterprise-AI-Factory Sources: https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/index.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/abstract.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/dgx-superpod-components.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/network-fabrics.html, https://docs.nvidia.com/dgx-superpod/reference-architecture/scalable-infrastructure-b300-xdr/latest/components.html Last Updated: 2026-05-09

Summary

NVIDIA DGX SuperPOD B300 Quantum-X800 InfiniBand Reference Architecture is the DGX B300 SuperPOD design focused on NVIDIA Quantum-X800/XDR InfiniBand switching and AC power. It describes a large SuperPOD configuration built from NVIDIA-DGX-B300 systems, 800 Gb/s InfiniBand compute fabric, storage and Ethernet management fabrics, NVIDIA-Mission-Control, NVIDIA-AI-Enterprise, CUDA, Magnum IO, NGC, Slurm, and NVIDIA Enterprise Support.

Detail

Purpose

Use this page for the DGX B300 SuperPOD reference architecture document titled “NVIDIA SuperPOD with DGX B300 Systems, NVIDIA Quantum-X800 InfiniBand switching and AC Power Reference Architecture.” Use NVIDIA-DGX-SuperPOD for the generic SuperPOD concept and NVIDIA-Quantum-X800-InfiniBand for the network platform.

Architecture notes

  • Document number RA-11339-001 V01; publication date 2025-07-23; docs last updated 2025-11-19.
  • Based on DGX B300 systems powered by NVIDIA Blackwell Ultra GPUs.
  • Built around scalable units; each SU contains 72 DGX B300 systems.
  • The fully tested system scales to 8 SUs, or 576 DGX B300 nodes, with larger deployments possible based on customer requirements.
  • The compute fabric uses managed Quantum-X800 switches and a rail-optimized full-fat-tree topology.
  • Storage fabric options include InfiniBand and Ethernet storage fabrics; the design also includes in-band and out-of-band Ethernet management networks.
  • Full system support, including compute, storage, network, and Mission Control software, is provided by NVIDIA Enterprise Experience (NVEX).

Software stack

The software table includes NVIDIA-Mission-Control, NVIDIA-AI-Enterprise, Magnum IO, NGC, and Slurm. The docs note that Mission Control includes Base Command Manager and Run:ai functionality, and that SuperPOD supports multiteam environments through Base Command Manager while multitenancy is not currently supported with SuperPOD.

NVIDIA context

This page captures the InfiniBand/AC-power DGX B300 SuperPOD path. It complements NVIDIA-DGX-SuperPOD-B300-Spectrum-4-Ethernet-RA, which captures the Spectrum-4/DC-busbar SuperPOD document, and NVIDIA-NVL72-AI-Factory, which captures the Enterprise RA path for GB300 NVL72 rack-scale systems.

Connections

Source Excerpts

  • NVIDIA describes this RA as a DGX B300 SuperPOD design based on Blackwell Ultra GPUs.
  • The abstract says each scalable unit contains 72 DGX B300 systems.
  • The network fabrics page describes a full 576-node SuperPOD with rail-aligned groups of 72 nodes.

Resources