Liquid-Cooled Al SuperCluster With 256 NVIDIA HGX™ B200 GPUs, 32 4U Liquid-cooled Systems

The Supermicro Liquid-Cooled AI SuperCluster delivers unmatched AI training and inference performance with 256 NVIDIA HGX™ B200 GPUs and next-generation liquid cooling technology. Designed for rapid deployment, energy efficiency, and scalability, this solution is the ultimate infrastructure for high-performance AI workloads​.

Key Features:

• Unprecedented GPU Power: 256 NVIDIA HGX B200 GPUs per 32-node cluster, each with 180GB HBM3e memory and 1.8TB/s NVLink bandwidth​.
• High-Density Compute: 32 liquid-cooled 4U systems powered by dual Intel® Xeon® or AMD EPYC™ processors​.
• Advanced Networking: 8x NVIDIA Quantum-2 400G InfiniBand or Spectrum-4 400GbE switches for non-blocking, low-latency connectivity​.
• Massive Scalability: Expand to a 96-node cluster with 768 NVIDIA B200 GPUs across 9 racks​.
• Efficient Cooling: 250kW liquid cooling distribution units (CDU) with redundant hot-swap pumps for superior thermal management.