Super Micro Computer (SMCI, Financials) announced the full-scale production of its AI data center solutions, powered by NVIDIA's Blackwell platform.
The range of Building Block Solutions from the firm currently includes multiple air-cooled and liquid-cooled systems, enabling several CPU choices. Liquid-to-liquid and liquid-to-air configurations are among the advanced cooling methods designed to improve thermal efficiency. Supermicro also provides a complete suite of data center management tools along with rack-level integration services, including networking, cabling, and cluster-level validation.
Next-generation cooling methods are abundant in the Supermicro NVIDIA HGX B200 8-GPU systems. Within the same 4U form factor, newly designed cold plates and a 250kW coolant distribution unit more than double the cooling capacity of the previous model. Available in 42U, 48U, or 52U versions, the rack-scale design incorporates vertical coolant distribution manifolds that free up valuable rack space. With this architecture, a 52U rack can house up to 12 computers totaling 96 NVIDIA Blackwell GPUs.
Supermicro introduces a new 10U air-cooled NVIDIA HGX B200 system with a redesigned chassis to accommodate eight 1000W TDP Blackwell GPUs for conventional data centers. By fully integrating up to four of these 10U systems into a rack, one can maintain the same density as the previous generation while achieving up to 15 times the inference performance and three times the training performance.
Supermicro's SuperCluster designs feature NVIDIA Quantum-2 InfiniBand or NVIDIA Spectrum-X Ethernet networking in a centralized rack, enabling a non-blocking, 256-GPU scalable unit across five racks or an expanded 768-GPU scalable unit across nine racks. Designed specifically for NVIDIA HGX B200 systems, this architecture supports the NVIDIA AI Enterprise software platform for building and running production-grade AI pipelines.