Building the Modular Foundation for AI Factories with NVIDIA MGX

Table of Contents

1. Why modular architecture matters now
2. Inside the MGX rack system
   • Mechanical components
   • Electrical components
   • Plumbing (cooling) infrastructure
3. Get started

Why modular architecture matters now

• NVIDIA MGX offers a modular reference architecture for accelerated computing, allowing partners to design multiple systems with a building-block approach.
• It supports multiple product generations and hundreds of GPU, DPU, CPU, storage, and networking combinations for AI, HPC, and digital twins.
• Enables high-density, rack-scale deployments without compromising performance or reliability.
• Streamlines build process, cutting deployment timelines from 12 months to less than 90 days.
• Empowers organizations to optimize data center builds for cost, performance, and supply chain resilience.

Inside the MGX rack system

Mechanical components

• The modular MGX rack provides structural integrity and serviceability for high-density data center deployments.
• The 33 kW Power Shelf supplies substantial power, and the MGX Power Whip offers flexible connections for power distribution.
• MGX components in NVIDIA GB200 NVL72 and GB300 NVL72 systems manage power density and thermal loads for liquid-cooled, rack-scale platforms.
• Advanced liquid-cooled MGX architecture addresses energy demands for high-performance computing, like the Blackwell compute nodes.

Electrical components

• The system enables training models with up to 1.8 trillion parameters on coherent 72-GPU domains using NVLink switches.
• Deploy inference clusters with low latency variance across 72-node racks for AI workloads.

Plumbing (cooling) infrastructure

• Ensures efficient thermal management for high-powered systems like the GB200 NVL72 and GB300 NVL72.

Get started

• NVIDIA MGX is the foundation for the AI factory era, enabling evolution with silicon innovations and modular upgrade paths for data center investments.