NVIDIA RTX PRO AI Factory

Type: Reference Architecture Tags: NVIDIA, RTX PRO AI Factory, Enterprise Reference Architecture, RTX PRO 6000 Blackwell, Spectrum-X, BlueField-3, AI Enterprise Related: NVIDIA-Enterprise-Reference-Architectures, NVIDIA-Enterprise-AI-Factory, Red-Hat-AI-Factory-with-NVIDIA, NVIDIA-RTX-PRO-Server, NVIDIA-Blackwell-Architecture, NVIDIA-AI-Enterprise, NVIDIA-AI-Enterprise-Software-Reference-Architecture, NVIDIA-Enterprise-RA-Observability-Guide, NVIDIA-Spectrum-X, NVIDIA-BlueField-DPU, NVIDIA-Certified-Systems, NVIDIA-Certified-Storage, NVIDIA-Run-ai, NVIDIA-vGPU, NVIDIA-Omniverse, NVIDIA-Omniverse-Reference-Architectures, NVIDIA-NetQ Sources: https://docs.nvidia.com/enterprise-reference-architectures/rtx-pro-ai-factory/latest/index.html, https://docs.nvidia.com/enterprise-reference-architectures/rtx-pro-ai-factory/latest/overview.html, https://docs.nvidia.com/enterprise-reference-architectures/rtx-pro-ai-factory/latest/components.html, https://docs.nvidia.com/enterprise-reference-architectures/rtx-pro-ai-factory/latest/software.html, https://docs.nvidia.com/enterprise-reference-architectures/rtx-pro-ai-factory/latest/nvidia-certified-storage.html, https://docs.nvidia.com/ai-enterprise/deployment/red-hat-ai-factory/latest/prerequisites.html Last Updated: 2026-04-29

Summary

NVIDIA RTX PRO AI Factory is a current NVIDIA Enterprise Reference Architecture for building modular AI factories with NVIDIA RTX PRO Servers. It is based on a 2-8-5-200 infrastructure configuration with two CPUs, eight GPUs, and five 200 Gb/s-class network interfaces, using RTX PRO 6000 Blackwell Server Edition GPUs, BlueField-3 networking, Spectrum-X Ethernet, and NVIDIA AI Enterprise software.

Detail

Purpose

The RTX PRO AI Factory pattern targets enterprises that need an air-cooled, power-efficient, NVIDIA-certified platform for agentic AI inference, industrial and physical AI, visual computing, data analytics, scientific simulation, rendering, and video workflows.

Architecture notes

  • Each NVIDIA-Certified RTX PRO Server in the reference design carries eight RTX PRO 6000 Blackwell Server Edition GPUs.
  • Scalable Units are four servers, with documented 16-node/128-GPU and 32-node/256-GPU design points.
  • The architecture separates compute east-west traffic from north-south customer and storage traffic.
  • BlueField-3 SuperNICs and DPUs are used for compute fabric and converged infrastructure roles.
  • NVIDIA-Certified Storage guidance allocates storage bandwidth linearly, with about 12.5 Gb/s per GPU as a guideline.

Software stack

The RA points to NVIDIA Enterprise AI Factory Validated Design software, including NVIDIA-AI-Enterprise, NVIDIA-Base-Command-Manager, NVIDIA-Omniverse, NVIDIA-Run-ai, NVIDIA-vGPU, and NVIDIA-NetQ. Use NVIDIA-AI-Enterprise-Software-Reference-Architecture for the shared software stack across RA hardware choices.

NVIDIA context

This page is the durable RTX PRO Enterprise RA page. Use NVIDIA-RTX-PRO-Server for the broader platform/product identity and NVIDIA-Enterprise-AI-Factory for the strategy-level AI factory concept. The Red-Hat-AI-Factory-with-NVIDIA prerequisites point readers to this RA as an example certified hardware pattern.

Connections

Source Excerpts

  • The current docs describe the RTX PRO AI Factory as a practical Enterprise RA based on the 2-8-5-200 pattern with RTX PRO Servers, BlueField-3, and Spectrum-X networking.
  • NVIDIA lists agentic AI inference, industrial and physical AI, visual computing, and data analytics/scientific simulation as primary use cases.

Resources