NVIDIAEstimated

B200

Blackwell · Blackwell · SXM · 2024

Blackwell generation with second-gen Transformer Engine and FP4. A large step in FP8 throughput and memory over Hopper.

LLM trainingLLM inferenceMultimodalHPC
Precision fingerprint
6432t3216BF168i8
Memory

192 GB

HBM3e

Bandwidth

8 TB/s

peak

TDP

1000 W

liquid

Max model

~80B

FP16, planning est.

Compute throughput

FP6440 TFLOPS
FP3240 TFLOPS
TF321.10 PFLOPS
FP162.25 PFLOPS
BF162.25 PFLOPS
FP84.50 PFLOPS
INT84.50 PFLOPS

Platform & software

InterconnectNVLink 5 — 1.8 TB/s
PCIePCIe 6.0 x16
Coolingliquid
MIGSupported
PartitioningMIG
VirtualizationMIG, vGPU
FrameworksCUDA, TensorRT-LLM, Triton, NeMo, vLLM
AvailabilityAzure, GCP, OCI, bare-metal
Known limitations
  • ·1000 W typically requires direct-liquid cooling
  • ·Early-life availability constrained
Related GPUs
NVIDIA

H100 SXM5

80 GB · 989 TFLOPS FP16

NVIDIA

H100 PCIe

80 GB · 756 TFLOPS FP16

NVIDIA

H100 NVL

94 GB · 989 TFLOPS FP16