NVIDIA submits MGX 1.2, CPU rack standards to OCP to streamline global AI deployment
NVIDIA contributes MGX 1.2 specs and CPU rack standards to the Open Compute Project, aiming to streamline the global deployment of high-density AI infrastructure.
Key points
- NVIDIA released MGX 1.2 specifications at the OCP Global Summit to standardize liquid-cooled AI racks.
- The new standard supports 250kW rack densities and blind-mate cooling for faster deployment.
- Partners like Meta, Cisco, and Google announced networking and software advances alongside the hardware specs.
NVIDIA contributed updated rack design specifications to the Open Compute Project (OCP) during its Global Summit in San Jose. The company released the third-generation MGX 1.2 reference design, which includes details for the Vera Rubin architecture intended for large-scale AI factory deployments.
The MGX 1.2 specification introduces support for rack deployments reaching 250 kilowatts. According to NVIDIA, this update adds blind-mate liquid cooling, interoperable interconnect infrastructure, and a liquid-cooled 5,000-ampere busbar. These components are designed to handle the power demands of next-generation accelerated computing systems while simplifying assembly and field service times.
Standards and Partnerships
NVIDIA shared these designs to give system builders a common foundation for developing AI infrastructure. The announcement included several partnerships aimed at creating a unified approach to AI factories. Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA, stated that open infrastructure is critical for scaling these facilities worldwide.
The MGX 1.2 update also expands modularity with configurable I/O bays and fanless liquid-cooled compute trays. It includes standardized power and cooling interfaces to help ecosystem partners build interoperable platforms. This approach aims to improve power efficiency and drive higher tokens-per-watt performance across the industry.
Broader Ecosystem Advances
Several partners announced complementary developments at the event. Meta and NVIDIA advanced open AI networking with Spectrum-X Ethernet SuperNIC and Spectrum-6 innovations. Visionbay introduced an 800-volt DC design for AI factories, while NVIDIA and Cisco announced the Ultra-Dense Small Form Factor Pluggable standard for 400Gb/s networking fabrics.
NVIDIA and Google announced Falcon Transport support for future ConnectX-10 SuperNICs and BlueField DPUs. Additionally, NVIDIA launched new open source software components for its DSX AI Factory. These efforts span the AI factory itself, the utility grid, and the rack-scale architecture.
What should you do?
Infrastructure planners should review the MGX 1.2 specifications for compatibility with their liquid-cooling strategies. Teams should evaluate how the 250kW rack density impacts their power grid resilience and facility design. Organizations should monitor partner announcements for networking standard updates that affect their current hardware roadmap.
What to do and how to stay safe: NVIDIA
- Review MGX 1.2 technical documents to assess compatibility with existing liquid-cooling infrastructure plans.
- Evaluate power distribution requirements for 250kW rack densities in upcoming data center expansions.
- Monitor updates from networking partners like Meta and Cisco for new interconnect standards.
- Check open-source DSX AI Factory components for integration opportunities in current workflows.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the maximum power density supported by the new MGX 1.2 spec?
The MGX 1.2 specification supports rack deployments up to 250 kilowatts.
Who announced the Ultra-Dense Small Form Factor Pluggable standard?
NVIDIA and Cisco announced the USFP standard for 400Gb/s-based networking fabrics.
Does the Vera CPU rack support the same architecture as the GPU racks?
Yes, the Vera CPU rack is built on the third-generation NVIDIA MGX rack architecture.
