Supermicro Expands Open AI Data Center Infrastructure Portfolio

en.Wedoany.com Reported - On October 13, U.S. server manufacturer Supermicro announced the expansion of its artificial intelligence data center infrastructure portfolio based on Open Compute Project (OCP) standards, adding OCP-certified server systems, open racks, and rear door heat exchangers, and incorporating 48V DC power delivery, liquid cooling equipment, network switches, and management software into its Data Center Building Block Solutions (DCBBS). The related products are designed for rack-level and cluster-level deployment in both new and existing data centers and are now available to customers.

This expansion adopts OCP Open Rack Version 3 (ORv3) and Open Rack Wide (ORW) rack architectures, replacing traditional distributed server power modules and some power cables with a centralized 48V DC busbar. The ORv3 series offers 44OU and 48OU configurations in 21-inch racks, with a certified static load capacity of 2,268 kilograms and provisions for liquid cooling system integration. The new solution is also compatible with traditional EIA racks, allowing customers to choose the deployment form based on existing room space and equipment configurations.

The liquid cooling system uses DLC-2 direct-to-chip cooling technology, which directly removes heat generated by processors and accelerators through cold plates, while rear door heat exchangers handle the remaining heat in server exhaust air. This series of rear door heat exchangers provides cooling capacities from 10 to 120 kilowatts, supports AC and DC power, and features N+1 redundant fans and intelligent anti-condensation control. The newly introduced 4U in-rack coolant distribution unit (CDU) offers a cooling capacity of up to 250 kilowatts and can operate with 45 degrees Celsius warm water.

Computing equipment includes a 2OU eight-GPU server node based on NVIDIA HGX B300, using 21-inch ORv3 racks and liquid cooling configurations, with up to 18 nodes per rack, totaling 144 GPUs. Another 2OU four-node rack-level solution is designed for round-the-clock agentic AI workloads, supporting up to 45,696 Arm AGI processor cores per rack. The portfolio also includes FlexTwin high-performance computing systems based on Intel Xeon 6 or AMD EPYC processors, as well as rack architectures compatible with the NVIDIA Vera Rubin NVL72 platform.

Networking equipment covers 800G and 1.6T Ethernet switches, as well as a new switching platform using co-packaged optics (CPO) technology. Among them, the 2OU CPO switch for ORv3 racks uses a 102.4Tbps switching architecture, is equipped with 128 800G ports, and supports 40 to 60V DC busbar power and liquid cooling connections. The software system, through SuperCloud Composer and other management tools, provides unified monitoring of computing, networking, cooling, and power facilities, and supports Redfish interfaces and OpenBMC architecture.