NVIDIA Contributes Vera Rubin Platform to OCP

NVIDIA introduces a cable-free, liquid-cooled modular compute tray to the Open Compute Project to enhance AI factory efficiency.

NVIDIA has contributed its Vera Rubin platform to the Open Compute Project, introducing a liquid-cooled, modular compute tray that eliminates cables, hoses, and fans. Built on 3rd Gen MGX architecture, it features a liquid-cooled busbar supporting 5,000 amps.

The platform debuts MGX Quick Disconnect connectors and the USFP standard for 400Gb/s networking. Performance gains are significant: Meta reported a 2x boost for Hyperscale Cloud, while Red Panda saw 3.5x better streaming data performance on Apache Kafka.

To support AI factories, NVIDIA launched open-source DSX software and partnered with Meta on 102.4 Tb/s Ethernet via Spectrum ASICs. These systems feature sub-3ms failure rerouting and new 800V DC designs to ensure grid resilience and high-density efficiency.


Related news

NVIDIA and Cisco Launch USFP Networking Module

AMD Raises GDDR6 Memory Prices for AIB Partners

MONTECH Launches CyberFlow Liquid Cooling Series

NVIDIA Announces Faster DLSS 5 Model for October

Custom Quad-GPU V100 Workstation Beats M5 Ultra