The latest landmarks show that supermicro systems with the Nvidia B200 have surpassed the previous generation of systems with 3x the generation of tokens per second
San Jose, California.,, April 3, 2025 / Prnewswire / – Super Micro Computer, Inc. (SMCI)A total supplier of IT solutions for AI / ML, HPC, Cloud, Storage and 5G / EDGE, announcement of the leading performance of the industry industry on several MLPERF inference V5.0 references, using the NVIDIA HGX ™ B200 8-GPU. 4u liquid cooling systems and 10u air cooled have achieved the best performance in certain benchmarks. SuperMicro has demonstrated more than 3 times the generation of tokens per second (token / s) for LLAMA2-70b and llama3.1-405B Benchmarks compared to H200 8-GPU systems.
“Supermicro remains a leader in the AI industry, as evidenced by the first new benchmarks published by MLCommons in 2025,” said Charles LiangPresident and chief executive officer of Supermicro. “Our architecture of construction blocks allows us to be of the first market with a diversified range of optimized systems for various workloads. We continue to collaborate closely with Nvidia to refine our systems and guarantee a leadership position in the workloads of the AI.”
Find out more about the new MLPERF V5.0 Inferences: https://mlcommons.org/benchmarks/inference-datacenter/
SuperMicro is the only publication system for MLPERF inference performance system suppliers (on selected references) for NVIDIA HGX ™ B200 B200 B200 systems with air and liquid cooled. Air -cooled and liquid -cooled systems were operational before the MLCommons reference start date. Supermicro engineers optimized systems and software to present impressive performance. In the operating margin, the B200 REGROIDIDE SUPERMICRO system had the same level of performance as the B200 system cooled by liquid. SuperMicro delivered these systems to customers while we carry out the benchmarks.
MlCommons underlines that all the results are reproducible, that the products are available and that the results can be verified by other members of MLCOMS. Supermicro engineers optimized systems and software, as allowed by MLCOMMONS rules.
The SYS-421GE-NBRT-LCC (8x NVIDIA B200-SXM-180GB) and SYS-A21GE-NBRT (8x NVIDIA B200-SXM-180GB) showed performance leadership performing 8X7B mixture inference, the benchmarks mixture with 129,000 tokens / second. The system based on Nvidia B200 cooled by Air Supermicro and cooled by liquid provided more than 1,000 tokens / second inference for the Grand Llama3.1-405B Model, while previous generations of GPU systems have much smaller results. For smaller inference tasks, using lama2-70b Benchmark, a supermicro system with the NVIDIA B200 SXM-180 GB installed shows the highest performance of a level 1 system supplier.
Specifically:
- Stable XL (server)
SYS-A21GE-NBRT (8x B200-SXM-180GB)
# 1 requests / s, 28.92 - LAMA2-70b-Interactive-99 (server)
SYS-A21GE-NBRT (8x B200-SXM-180GB)
# 1 tokens / s, 62 265.70 - LAMA3.1-405B (offline)
SYS-421GE-NBRT-LCC (8xB200-SXM-180GB)
# 1 tokens / s 1521.74 - LAMA3.1-405B (Server)
SYS-A21GE-NBRT (8x B200-SXNM-180GB)
# 1 tokens / s, 1080.31 (for an 8-GPU node) - Mixtral-8X7B (server)
SYS-421GE-NBRT-LCC (8x B200-SXM-180GB)
# 1 tokens / s, 129,047.00 - Mixtral-8X7B (offline)
SYS-421GE-NBRT-LCC (8x B200-SXM-180GB)
# 1 tokens / s, 128,795.00
“Mlcommons congratulate SuperMicro for their submission to the reference MLPERF inference V5.0. We are delighted to see their results presenting significant performance gains in relation to previous generations of systems,” said David KanterMLPERF chief in MlCommons. “Customers will be satisfied with the performance improvements obtained which are validated by the neutral, representative and reproducible MLPERF results.”
SuperMicro offers a complex IA portfolio with more than 100 GPU optimized systems, both air -cooled and cooled by liquid, with a choice of processors, ranging from optimized systems to one place to 8 -way multiproacious systems. Rack SuperMicro systems include computer science, storage and network components, which reduces the time necessary to install them once they are delivered to a customer site.
NVIDIA HGX B200 8-GPU systems of Supermicro use liquid cooling and new generation air cooling technology. The newly developed cold plates and the new 250 kW cooling liquid distribution unit (CDU) double than double the cooling capacity of the previous generation in the same 4u shape factor. Available in 42U, 48U or 52U configurations, rack level design with new vertical coolant distribution collectors (CDM) no longer occupies precious rack units. This allows eight systems, including 64 NVIDIA Blackwell GPU in a Rack 42U and up to 12 systems with 96 NVIDIA Blackwell GPU in a 52U rack.
The new NVIDIA HGX B200 system, air -cooled, has a redesigned chassis with extended thermal margin to accommodate eight GPU Blackwell TDP 1000W. Up to 4 of the new air -cooled 10U systems can be installed and fully integrated into a rack, the same density as the previous generation, while offering up to 15x inference and 3x training performances.
About Super Micro Computer, Inc.
Supermicro (NASDAQ: SMCI) is a world leader in total IT solutions optimized by the application. Founded and operating in San Jose, CaliforniaSuperMicro undertakes to provide a first -market innovation for the company, the Cloud, the AI and the IT infrastructure of telecommunications / EDGE 5G. We are a total supplier of computer solutions with server, AI, storage, IoT, switching systems, software and support services. The design of the design of the motherboard, energy and supermicro chassis also allows our development and production, allowing new generation of the cut -off generation for our global customers. Our products are designed and manufactured internally (in the United States, AsiaAnd The Netherlands), Take advantage of global operations for scale and efficiency and optimized to improve TCO and reduce the environmental impact (Green IT). The award -winning portfolio of Building Block Solutions® servers allows customers to optimize their accurate workload and application by selecting from a wide family of systems built from our flexible and reusable construction blocks that support a full set of formal factors, processors, memory, GPU, storage, networking, electricity and cooling (Cooling in the open air or cash cooling).
Supermicro, Server Building Block Solutions, and we keep it green are commercial brands and / or registered trademarks of Super Micro Computer, Inc.
All other brands, names and brands are the property of their respective owners.
Super micro Computer source, Inc.
