Friday, Nov 13 | --:--
Back to home

Nvidia Opens Vera Rubin Rack Designs at OCP; Meta Adopts Spectrum-6 Switches

At the OCP Global Summit on Oct. 12, Nvidia contributed Vera Rubin rack and compute-tray designs through its MGX architecture, including an MGX 1.2 spec for racks up to 250 kW and a Vera CPU rack for agent workloads. Meta said it will adopt Spectrum-6 Ethernet switches, Nvidia and Cisco unveiled USFP optics for 400 Gb/s-per-lane fabrics, and Nvidia co-published open data-center and grid-resilience specs.

Times of AI Desk 3 min read San Jose, CA View as Markdown
Engraved cutaway of a tall liquid-cooled server rack with blueprints unrolling from it onto a long public table where engineers from several companies lean in, coolant pipes in green, emerald and gunmetal intaglio.

At the Open Compute Project Global Summit in San Jose on Monday, October 12, Nvidia contributed Vera Rubin rack and compute-tray designs to OCP through its open MGX architecture, the company said in a blog post. Vera Rubin itself was unveiled at GTC in March; this is the follow-on step of opening its hardware designs.

What Nvidia announced#

  • MGX 1.2: adds rack-scale blind-mate liquid cooling and a liquid-cooled 5,000A busbar, and supports rack deployments up to 250 kW, Nvidia says.
  • A CPU rack for agents: a Vera CPU rack specification designed for "tool calling, orchestration and sandboxed execution" alongside Vera Rubin NVL72 systems. Nvidia's Joe DeLaere told a press briefing that dedicated Vera racks are "shipping now," with over 20,000 cores per rack, SDxCentral reports.
  • Meta networking: Meta is adopting Nvidia Spectrum-6 102.4 Tb/s Ethernet switches to build larger Vera Rubin deployments on open networking.
  • Optics: Nvidia and Cisco announced plans for USFP, a pluggable optics form factor providing a path to 400 Gb/s per lane.
  • Facility specs: Nvidia joined AMD, Google and Meta in publishing an Open Data Center Specification v0.7, and Google, Meta, Microsoft and Nvidia published an OCP Grid Resilience Specification. Nvidia also released open-source components for its DSX AI Factory software.

CPUs for agents#

On the same day, Intel CEO Lip-Bu Tan said frontier model companies were asking for more CPUs for inference orchestration and reinforcement learning, Computer Weekly reported in its coverage of the DBS conference in Singapore.

Limits#

  • These are vendor announcements; Nvidia gave no pricing, and apart from the Vera rack, no shipment dates.
  • The Vera performance claims reported by SDxCentral are Nvidia's own and are not repeated here.

Sources#

Prior Coverage

Earlier Times of AI reporting on this thread.

Scroll to continue reading