Supermicro delivers direct-liquid-cooled infrastructure solutions optimized for NVIDIA Blackwell GPUs.

Supermicro, Inc. (NASDAQ: SMCI), a Total IT Solution Provider for AI, Cloud, Storage, and 5G/Edge, announced at the Supercomputing Conference in San Jose (California) and Atlanta (Georgia) on November 19 the highest-performing SuperCluster, an end-to-end AI data center solution featuring the NVIDIA Blackwell platform for the era of trillion-parameter-scale generative AI. The new SuperCluster significantly increases the number of NVIDIA HGX B200 8-GPU systems in a liquid-cooled rack, resulting in a large increase in GPU compute density compared to Supermicro's current industry-leading liquid-cooled NVIDIA HGX H100 and H200-based SuperClusters. Supermicro is also enhancing its portfolio of NVIDIA Hopper systems to address the rapid adoption of accelerated computing for HPC applications and mainstream enterprise AI.

"Supermicro has the expertise, delivery speed, and capacity to deploy the largest liquid-cooled AI data center projects in the world, containing 100,000 GPUs, which Supermicro and NVIDIA contributed to and recently deployed," said Charles Liang, president and CEO of Supermicro. He emphasized that these SuperClusters reduce power needs due to DLC efficiencies and highlighted how Supermicro's Building Block approach enables quick design of servers with NVIDIA HGX B200 8-GPU in both liquid-cooled and air-cooled configurations. "Our SuperClusters provide unprecedented density, performance, and efficiency, and pave the way toward even more dense AI computing solutions in the future. The Supermicro clusters use direct liquid cooling, resulting in higher performance, lower power consumption for the entire data center, and reduced operational expenses."

The upgraded SuperCluster features a rack-scale design with innovative vertical coolant distribution manifolds (CDMs) allowing for increased compute nodes per rack. The new Supermicro NVIDIA HGX B200 8-GPU systems include improvements to thermals and power delivery with support for dual 500W Intel Xeon 6 processors with DDR5 MRDIMMs at 8800 MT/s, or AMD EPYC 9005 Series processors. A new air-cooled 10U form-factor system features a redesigned chassis accommodating eight 1000W TDP Blackwell GPUs. These systems employ a 1:1 GPU-to-NIC ratio supporting NVIDIA BlueField-3 SuperNICs or NVIDIA ConnectX-7 NICs, with two NVIDIA BlueField-3 data processing units (DPUs) per system streamlining data handling to attached high-performance AI storage.

Supermicro's lineup of NVIDIA MGX designs supports the NVIDIA GB200 Grace Blackwell NVL4 Superchip, which delivers four NVIDIA NVLink-connected Blackwell GPUs unified with two NVIDIA Grace CPUs over NVLink-C2C, providing up to 2x performance for scientific computing, graph neural network (GNN) training, and inference applications. The NVIDIA GB200 NVL72 SuperCluster with Supermicro's end-to-end liquid-cooling solution delivers an exascale supercomputer in a single rack with SuperCloud Composer (SCC) software for comprehensive monitoring and management. This configuration includes 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs connected via fifth-generation NVIDIA NVLink and NVLink Switch, facilitating 130TB/s of total GPU communication bandwidth.

Supermicro's 5U PCIe accelerated computing systems are now available with NVIDIA H200 NVL for lower-power, air-cooled enterprise rack designs. These systems deliver up to 1.7x faster LLM inference performance over the previous generation and enable fine-tuning of LLMs in a few hours, with a 1.5x memory capacity increase and 1.2x bandwidth increase using HBM3e. The X14 and H14 5U PCIe systems support up to two 4-way NVIDIA H200 NVL systems through NVLink technology with a total of eight GPUs per system, providing 900GB/s GPU-to-GPU interconnection and a combined pool of 564GB of HBM3e memory per 4-GPU NVLink domain. These systems support up to 10 PCIe GPUs and feature the latest Intel Xeon 6 or AMD EPYC 9005 Series processors, with each H200 NVL including a five-year subscription to NVIDIA AI Enterprise.