NVIDIA B100 Blackwell 

Specification NVIDIA B100
GPU Architecture NVIDIA Blackwell
GPU NVIDIA B100 Tensor Core GPU
GPU Memory 192GB HBM3e
Memory Interface 8192-bit
Memory Bandwidth Up to 8 TB/s
CUDA Cores Blackwell GPU architecture
Tensor Cores 5th Generation
RT Cores N/A for this data-center accelerator
FP4 Tensor Performance 14 PFLOPS
FP8 / FP6 Tensor Performance 7 PFLOPS
INT8 Tensor Performance 7 POPS
FP16 / BF16 Tensor Performance 3.5 PFLOPS
TF32 Tensor Performance 1.8 PFLOPS
FP32 Performance 60 TFLOPS
FP64 Tensor Performance 30 TFLOPS
FP64 Performance 30 TFLOPS
GPU Interconnect NVLink 5
NVLink Bandwidth 1.8 TB/s
PCIe PCIe Gen6, 256 GB/s
Form Factor SXM
TDP 700W
Cooling Passive / Server Cooling
Display Outputs None
ECC Yes
Memory Type HBM3e
Target Market Data Center / AI / HPC
Typical Platform NVIDIA HGX B100
HGX Configuration 8 × B100 GPUs