GLM 5/5.1 — B300 vs H200
Head-to-head AI inference benchmark comparison of B300 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on GLM 5/5.1. Latency, throughput, and cost across LLM workloads. Use the chart controls below to switch sequences, precisions, and metrics — same interactions as the main inference chart.
Near the low end of the 21–101 tok/s/user interactivity band, at 41 tok/s/user on GLM 5/5.1: B300 runs 825 tok/s/GPU at $0.79/M tokens, H200 runs 329 at $1.19/M. B300 is 50% cheaper per token; B300 delivers 150% more tok/s/GPU.
Setting 61 tok/s/user as the target on GLM 5/5.1, B300 produces 576 tok/s/GPU ($1.12 per million tokens) and H200 produces 212 ($1.85). B300 is 65% cheaper per token; B300 delivers 171% more tok/s/GPU.
At 82 tok/s/user interactivity on GLM 5/5.1, B300 delivers 410 tok/s/GPU at $1.57 per million tokens; H200 delivers 132 tok/s/GPU at $2.93. B300 is 87% cheaper per token; B300 delivers 209% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | B300:825.1H200:329.5 | B300:575.9H200:212.5 | B300:409.7H200:132.4 |
| Cost ($/M tok) | B300:$0.791H200:$1.190 | B300:$1.119H200:$1.847 | B300:$1.569H200:$2.933 |
| tok/s/MW | B300:380246H200:190452 | B300:265399H200:122811 | B300:188790H200:76544 |
| Concurrency | B300:~82H200:~34 | B300:~39H200:~14 | B300:~21H200:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.