Inference Volume$52.1B+5.1%
Model Rankings165updated
AI Traffic624.8K+38.2%
Global Latency11msoptimal
API Calls102.4M+12.4%
Live Agents582,391+42.1%
AI Market Cap$58.2B+5.2%
GPU Nodes6,247+10.2%
Inference Volume$52.1B+5.1%
Model Rankings165updated
AI Traffic624.8K+38.2%
Global Latency11msoptimal
API Calls102.4M+12.4%
Live Agents582,391+42.1%
AI Market Cap$58.2B+5.2%
GPU Nodes6,247+10.2%

NVIDIA B200 (Blackwell)

NVIDIA Data Center180GB HBM3e • 8 TB/s • 120 TFLOPS FP8
$40,000 - $60,000-1.03%Coming 2025
TDP
1000W
VRAM
180GB HBM3e
Bandwidth
8.0 TB/s
Compute
120 FP32 / 20,000 FP8
Released
Late 2025
Availability
Coming 2025

Description

The B200 is NVIDIA's next-generation Blackwell architecture GPU with 2nd-gen Transformer Engine, offering up to 4x performance over H100 for AI training and inference.

Benchmarks

MLPerf Training
4.0x vs H100
LLM Inference
30x vs H100
FP8 Throughput
20,000 TFLOPS
NVLink Bandwidth
1.8 TB/s