NVIDIA opens Vera Rubin rack designs at OCP, plus AI factory standards
At the Open Compute Project Global Summit in San Jose, NVIDIA said it contributed its Vera Rubin rack and compute tray designs to OCP through the open MGX architecture; the updated MGX 1.2 specification adds rack-level blind-mate liquid cooling, interoperable rack-scale interconnect and a liquid-cooled 5,000A busbar, supports rack deployments up to 250 kW and is meant to simplify assembly, test and service. NVIDIA also published Open Data Center Specification v0.7 with AMD, Google and Meta, and an OCP Grid Resilience Specification with Google, Meta and Microsoft that calls for large AI infrastructure to stay connected during grid disturbances and restore to near-full load in under two seconds. On networking, Meta is adopting NVIDIA Spectrum-6 102.4Tb/s Ethernet switches with Spectrum-X Multiplane technology to support AI factories of more than 100,000 GPUs, while NVIDIA released four open source DSX AI Factory components: Config Manager, Cluster Readiness Engine, Rack Management Service and Cluster API Provider.
Common open specs for racks, power and networking could lower integration barriers and speed up deployment of high-density AI factories.