Nvidia opens Vera Rubin AI rack designs to industry

Nvidia has given its Vera Rubin rack and compute tray designs to the Open Compute Project (OCP). The company announced the move Monday, Oct. 12, at the OCP Global Summit in San Jose, along with networking deals and new data center standards. Those standards were written with AMD, Google, Meta and Microsoft. Nvidia says shared designs will let equipment makers start manufacturing sooner, so AI data centers, which it calls "AI factories," can be built faster.blogs.nvidia+1

The designs are published through version 1.2 of Nvidia's MGX modular architecture. The spec includes liquid-cooled compute trays with no fans or cables, rack-scale interconnects, a liquid-cooled 5,000-amp busbar and support for racks drawing up to 250 kilowatts. A separate spec covers Nvidia's new Vera CPU rack. Nvidia said that rack has started shipping, SDxCentral reported. Nvidia also opened its Blackwell designs in the past.sdxcentral+1

"When the specification is open, the equipment makers and component developers can start fabricating right away and build a supply chain instead of waiting for a singular customer with one proprietary spec," Joe DeLaere, Nvidia's product lead for AI infrastructure solutions, said in a press briefing. "The ecosystem grows, and we grow with them."sdxcentral

Industry standards for buildings and the power grid

Nvidia, AMD, Google and Meta published the Open Data Center Specification v0.7, a shared plan for AI data center buildings that covers power, cooling and physical layout. Google, Meta, Microsoft and Nvidia also released the OCP Grid Resilience Specification. It sets out how large AI sites should stay connected during power-grid disturbances and get back to near-full load in under two seconds.blogs.nvidia

Yahoo Finance noted that faster construction would also mean more Nvidia chip and networking sales. It also pointed to growing local opposition to data centers, including a one-year statewide moratorium on construction in New York and a pause on new projects in Texas.finance.yahoo

Networking partnerships

Nvidia will support Google's Falcon Transport, a protocol for sending data reliably over Ethernet, on its future ConnectX-10 SuperNICs and BlueField DPUs. Meta is adopting Nvidia's Spectrum-6 102.4-terabit-per-second Ethernet switches for its Vera Rubin AI factories. The switches will run Meta's open-source FBOSS network software. Nvidia and Cisco introduced USFP, a new pluggable optical module for 400-gigabit-per-lane networks. The companies say it offers four times the bandwidth density of current 200-gigabit optics.siliconangle+1

Ian Buck, Nvidia's vice president of hyperscale and high-performance computing, is set to give a keynote at the summit at 5:45 p.m. on Monday. The summit runs through Oct. 15.nvidia