NVIDIA HGX B100 8-GPU
Specification HGX B100
Product Name NVIDIA HGX B100
GPU 8 × NVIDIA B100 Tensor Core GPU
Architecture NVIDIA Blackwell
GPU Form Factor SXM
Memory per GPU 192GB HBM3e
Total GPU Memory 1.5TB HBM3e
Memory Bandwidth / GPU Up to 8TB/s
Aggregate Memory Bandwidth Up to 64TB/s
NVLink 5th Generation
GPU-to-GPU NVLink Bandwidth 1.8TB/s
Total NVLink Bandwidth 14.4TB/s
PCIe PCIe Gen6
GPU TDP Up to 700W
FP4 Tensor Performance 112 PFLOPS
FP8/FP6 Tensor Performance 56 PFLOPS
FP16/BF16 Tensor Performance 28 PFLOPS
INT8 Tensor Performance 28 POPS
FP32 480 TFLOPS
FP64 240 TFLOPS
Platform Type OEM / ODM Server Platform
Target Applications AI Training, AI Inference, LLM, HPC