Supermicro NVIDIA HGX Systems power the world’s largest liquid-cooled AI data centers. The new 4U NVIDIA HGX B200 8-GPU system features new cold plates and tubing design that further enhances efficiency and serviceability over its predecessor. The system features 8 NVIDIA Blackwell GPUs, each with 180 GB HBM3e memory. The GPUs are interconnected at 1.8 TB/s through the latest NVIDIA NVLink, with 1.4 TB of GPU memory capacity per system.
The SuperCluster creates a massive pool of GPU resources, acting as one AI supercomputer, featuring 1:1 networking to GPU with 8× 400 Gb/s NVIDIA ConnectX-7 adapters or BlueField-3 SuperNICs, as well as 2 NVIDIA BlueField-3 DPUs per system.
| Node Configuration | |
|---|---|
| Overview | 4U Liquid-cooled system with NVIDIA HGX B200 8-GPU |
| CPU | Dual Intel® Xeon® 6900 series (SYS-422GA-NBRT-LCC) Dual AMD EPYC™ 9005/9004 (AS-4126GS-NBRT-LCC) Dual 5th/4th Gen Intel® Xeon® Scalable (SYS-421GE-NBRT-LCC) |
| Memory | 24 DIMMs up to DDR5-6400 24 DIMMs up to DDR5-6000 32 DIMMs up to DDR5-5600 |
| GPU | 8× NVIDIA HGX B200 (180 GB HBM3e each) 1.8 TB/s NVIDIA NVLink with NVSwitch |
| Networking | 8× single-port NVIDIA ConnectX-7 or BlueField-3 SuperNICs 2× dual-port BlueField-3 DPUs |
| Storage | 8× hot-swap 2.5” NVMe bays 2× M.2 NVMe slots |
| Power Supply | 4× 6600 W redundant Titanium PSUs |
*Recommended configuration; other memory, networking, storage options available.
