OEM
HGX B200
Blackwell
Built for AI and High-Performance Computing
Artificial intelligence, complex simulations, and large-scale datasets require GPUs with multiple high-speed interconnects and a fully accelerated software stack. The NVIDIA HGX™ platform brings together the full capabilities of NVIDIA GPUs, NVIDIA NVLink™, and NVIDIA networking, along with a fully optimized AI and high-performance computing (HPC) software stack. This delivers the best application performance and accelerates time to insight for every data center.
Specifications
HGX B200 (Intel/AMD) | |
GPU | 8x NVIDIA Blackwell GPUs (SXM) |
GPU Memory | 8x 180GB |
Performance | 72 petaFLOPS FP8 training and 144 petaFLOPS FP4 inference |
NVIDIA® NVSwitch™ | Included |
NVIDIA NVLink Bandwidth | 14.4 TB/s aggregate bandwidth |
System Power Usage | ~14.3 kW |
CPU (Intel/AMD) | Dual Intel® Xeon® 6900 series processors with P-cores Intel® Xeon® Gen4/5 scalable Dual EPYC 9004/9005 |
System Memory | 24x DIMM slots, ECC RDIMM / MRDIMM DDR5 up to 8800MT/s (Up to 6TB) 32x DIMM slots, ECC RDIMM / MRDIMM DDR5 up to 5600MT/s (Up to 4TB) 32x DIMM slots, ECC RDIMM / MRDIMM DDR5 up to 4400MT/s (Up to 8TB) 24x DIMM slots, ECC RDIMM / MRDIMM DDR5 up to 4800MT/s (Up to 6TB) 24x DIMM slots, ECC RDIMM / MRDIMM DDR5 up to 6400MT/s (Up to 6TB) |
Interface | 2 RJ45 1 GbE 2 USB 3.0 ports 1 VGA port 1 TPM header |
Network | Max 8 NIC/IB |
Storage | 8/10 x 2.5" Gen5 NVMe/SATA |
Power | 6 5250W(3+3) |
Rack Units (RU) | 10 RU |
Brand | SMCI, Asus, Gigabyte, Dell |