NVIDIA B200

NVIDIA B200 is their Blackwell-generation GPU. Each GPU has:

• 10 GPCs¹
• ? TPCs
  • ? SMs (? per TPC)
  • ? FP32 cores (128 per SM)
  • ? tensor cores (4 per SM)
  • ? GHz (low precision), ? GHz (high precision)
  • 2:4 structured sparsity
• 192 GB HBM3 (8 stacks)
  • 8 TB/s (max)
• 2x 900 GB/s NVLink v5 (D2D)²
• 2x 256 GB/s PCIe Gen6 (H2D)²
• 1000 W maximum

B100 GPUs are a lower-power variant of B200 (700W) that is meant to be a “drop-in replacement” for HGX H100 platforms.³ That is, you can take a server platform built for 8-way H100 baseboards, swap in B100 baseboards, and sell them without having to re-engineer power or thermals.

Performance

The following are theoretical maximum performance in TFLOPS:³

Data Type	VFMA	Matrix	Sparse
FP64	90	90
FP32	180
TF32		1250	2500
FP16		5000	10000
BF16		2500	5000
FP8		5000	10000
FP6		5000	10000
FP4		10000	20000
INT8		5000	10000

Footnotes

1. open-gpu-kernel-modules/kernel-open/nvidia-uvm/uvm_blackwell_fault_buffer.h at main · NVIDIA/open-gpu-kernel-modules (github.com) ↩

2. GB200 NVL2 | NVIDIA ↩ ↩²

3. https://resources.nvidia.com/en-us-blackwell-architecture ↩ ↩²