Supermicro Starts Shipping NVIDIA's Rubin AI Racks, Plumbed for Liquid Cooling
Supermicro said on Sept. 23 that it has begun shipping data center racks built on NVIDIA's Vera Rubin NVL72 design, the rack-sized systems that carry NVIDIA's new Rubin generation of AI processors.
The company says a buyer can now order a complete unit and take delivery of production-ready systems, with one Supermicro team running the site survey, design, integration, testing and on-site deployment. The racks arrive with their cooling plumbing attached rather than with fans: the release says NVIDIA's Vera Rubin platform was designed alongside direct liquid cooling because air cooling alone cannot reach the same computing throughput per watt, and that the heat has to be carried the whole way, from the cold plates inside the rack out to a cooling tower.
The company's published configuration puts 72 Rubin GPUs and 36 Vera CPUs in one liquid-cooled rack that operates as a single machine.
For larger sites, Supermicro says its standard building block, which it calls a scalable unit, comes to 1,152 Rubin GPUs across 16 compute racks. Its blueprints cover power envelopes from 5 MW up to gigawatt scale.
Charles Liang, Supermicro's president and chief executive, said in the release that "our development of DLC-2 technology is what sets Supermicro apart and allows us to deliver the most complete solution for the NVIDIA Vera Rubin platform."
