Supermicro said on Sept. 29 it has started shipping Nvidia's Vera Rubin NVL72 rack that integrates its data center building block solution, DCBBS, and its direct liquid cooling technology stack, DLC-2. The company said it will take orders and supply the rack in scalable units. Each scalable unit supports 1,152 GPUs and 331TB of HBM4 memory.

The Nvidia Vera Rubin platform is designed on the premise of direct liquid cooling. It is intended to deliver AI processing performance per watt that is difficult to achieve with air cooling alone. The structure requires the entire coolant circulation system, from cold plates and manifolds to the cooling tower, to transfer heat.

Supermicro said it tests and validates all racks with the liquid cooling system integrated as one. It can reduce the time from on-site deployment to the start of operations. The company handles mechanical design of system nodes, mass production of liquid-cooled racks, and on-site deployment of clusters based on Nvidia reference architecture. A dedicated team also manages site surveys, project design, integration, L11 and L12 testing, on-site deployment and post-deployment support.

The company said it has built its own portfolio of cooling components based on thermal management technologies developed through DCBBS. The lineup includes cold plates, manifolds, hose kits and rack power shelves. It also includes in-row and in-rack coolant distribution units, liquid-to-air sidecar CDUs, rear-door heat exchangers and facility cooling towers. DCBBS blueprints by power scale are available from 5 MW to gigawatt-class.

Charles Liang (찰스 리앙), president and CEO of Supermicro, said, "Supermicro has built its liquid cooling technology stack, manufacturing capabilities and specialist deployment team for years for this moment." He added, "Customers can now order scalable units and receive a system that can be applied immediately to a commercial environment with end-to-end integration completed." He also said, "Supermicro's DLC-2 technology, developed by Supermicro, is at the core of these differentiated capabilities, enabling it to provide a highly complete integrated solution for the Nvidia Vera Rubin platform."