Summary:
- Super Micro Computer, Inc. expands its NVIDIA Blackwell architecture portfolio with new liquid-cooled systems for hyperscale data centers and AI factories.
- The new systems offer high GPU density and energy efficiency, setting new standards in the fast-paced AI infrastructure landscape.
- These additions include 4U and 2-OU (OCP) systems, aligning with OCP Open Rack V3 specifications for maximum GPU density.
Article:
Super Micro Computer, Inc. (SMCI), a leading provider of comprehensive IT solutions, has recently unveiled a significant expansion in its NVIDIA Blackwell architecture portfolio. This expansion includes the introduction and shipment of cutting-edge 4U and 2-OU (OCP) liquid-cooled NVIDIA HGX B300 systems, which are designed to cater to the needs of hyperscale data centers and AI factories.
President and CEO Charles Liang emphasized the importance of these new systems in meeting the demands of today’s rapidly evolving AI infrastructure landscape. By offering the market’s most compact NVIDIA HGX B300 solutions, Supermicro is able to achieve an impressive 144 GPUs in a single rack through its direct liquid-cooling technology. This not only reduces power consumption but also cuts down on cooling costs significantly.
The 2-OU (OCP) system is in line with the 21-inch OCP Open Rack V3 (ORV3) specifications, enabling up to 144 GPUs per rack. This high GPU density is crucial for hyperscale and cloud providers looking to maximize space efficiency without compromising serviceability. The innovative design features efficient cooling solutions, blind-mate manifold links, and a modular GPU/CPU tray setup, all aimed at enhancing AI workloads by leveraging eight NVIDIA Blackwell Ultra GPUs.
In addition to the 2-OU system, Super Micro Computer, Inc. also offers a 4U system variant that maintains its computational power in a traditional 19-inch EIA rack. Thanks to Supermicro’s DLC-2 technology, this system captures up to 98% of the heat generated, improving power efficiency, reducing noise, and enhancing serviceability for dense AI clusters.
Key performance enhancements in these systems include 2.1TB of HBM3e GPU memory for handling larger models and an improved compute fabric throughput up to 800Gb/s using integrated NVIDIA ConnectX-8 SuperNICs when paired with NVIDIA networking solutions. These enhancements accelerate AI workloads such as agent-driven applications, foundational model training, and large-scale inference.
Supermicro’s commitment to total cost of ownership, efficiency, and serviceability is evident in their use of DLC-2 technology, which allows data centers to optimize power utilization by up to 40%, reduce water usage through 45°C warm water operations, and eliminate the need for chilled water and compressors. These pre-validated systems streamline deployment, catering to hyperscale, corporate, and governmental affiliates.
The introduction of these new systems further expands Super Micro Computer, Inc.’s NVIDIA Blackwell portfolio, which includes NVIDIA GB300 NVL72, NVIDIA HGX B200, and other certified platforms for optimal AI application performance. These systems offer secure scalability from single nodes to comprehensive AI infrastructures, ensuring that Supermicro remains at the forefront of innovation in the IT solutions industry.
