Super Micro Computer Inc. Leads AI Performance with NVIDIA HGX™ B200 Systems

Supermicro’s 4U liquid-cooled and 10U air-cooled systems have outperformed their competitors, achieving record results in several critical MLPerf Inference v5.0 benchmarks. Notably, these systems generated over three times the tokens per second (Token/s) for the Llama2-70B and Llama3.1-405B benchmarks when compared to previous H200 8-GPU systems. This remarkable achievement underscores Supermicro's engineering prowess and its strategic collaboration with NVIDIA.

Charles Liang, President and CEO of Supermicro, commented on the milestone: “Supermicro remains a leader in the AI industry, as evidenced by the first new benchmarks released by MLCommons in 2025. Our building block architecture enables us to be first-to-market with a diverse range of systems optimized for various workloads.”

Supermicro’s systems have set new records across multiple benchmarks, illustrating their superior capabilities:

These benchmarks reflect substantial performance gains over earlier generations, as noted by David Kanter, Head of MLPerf at MLCommons, who praised Supermicro for its impressive results.

Supermicro’s NVIDIA HGX B200 8-GPU systems feature next-generation cooling technologies that enhance both air and liquid cooling capacities. The newly developed cold plates and a 250kW coolant distribution unit (CDU) have doubled the cooling capacity of previous generations within the same 4U form factor. This advanced design allows for higher density in rack configurations, accommodating multiple systems without sacrificing performance.

The air-cooled 10U NVIDIA HGX B200 system boasts a redesigned chassis that can house eight 1000W TDP Blackwell GPUs. Significantly, up to four of these systems can be integrated into a rack, maintaining the same density as prior models while delivering up to 15 times the inference performance and three times the training performance.

Supermicro continues to expand its offerings within the AI space, presenting a comprehensive portfolio of over 100 GPU-optimized systems. These systems cater to various customer needs, from single-socket optimized setups to 8-way multiprocessor configurations. The company emphasizes its commitment to reducing installation time and enhancing operational efficiency through its integrated rack-scale systems that encompass computing, storage, and network components.

Super Micro Computer, Inc. stands at the forefront of the AI and ML industry, with its cutting-edge NVIDIA HGX B200 systems setting new performance benchmarks. The company’s strategic collaborations, innovative cooling technologies, and comprehensive product offerings position it well to meet the increasing demands of the AI landscape. As Supermicro continues to deliver first-to-market solutions, stakeholders can anticipate further advancements that will shape the future of technology and performance in AI workloads.