Nvidia shares Vera Rubin rack designs through Open Compute Project
Nvidia contributed Vera Rubin infrastructure designs to OCP and introduced shared standards and software for building larger AI data centers.
NVIDIA
Nvidia announced contributions of Vera Rubin rack and computing tray designs to the Open Compute Project on October 12. Unveiled at the group's summit in San Jose, the designs extend Nvidia's MGX modular infrastructure architecture.
The MGX 1.2 specification supports racks drawing as much as 250 kilowatts. It includes liquid cooling, fanless computing trays, support for 19-inch equipment and a cooled power conductor rated at 5,000 amps. Nvidia also released software for network configuration, cluster testing, rack management and server provisioning.
The effort addresses the power, cooling and integration demands of increasingly dense AI systems. Nvidia said common designs should help manufacturers build compatible equipment and simplify assembly and maintenance. Separate specifications developed with partners cover facility design and responses to electrical grid disruptions.
Additional networking work remains under development. Nvidia and Google announced planned Falcon Transport support for future ConnectX-10 adapters and BlueField processors, while Nvidia and Cisco outlined a new optical module format for faster connections.
The day's biggest AI news in one short email, every morning. Plus an instant heads-up when a reset or a big release lands.
More on NVIDIA
- Oct 10Nvidia reportedly weighs acquisition or further investment in Reflection AIStrong rumor
- Oct 10Nvidia reportedly stops supplying RTX 5090 chips as 24GB RTX 5080 rumor emergesRumor
- Oct 7Nvidia and Microsoft open RTX Spark PC preorders for local AI workloadsOfficial
- Oct 7Google opens SynthID Detector to the public worldwideOfficial