GLM 5/5.1 — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (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 21–67 tok/s/user interactivity band, at 33 tok/s/user on GLM 5/5.1: GB300 NVL72 runs 3417 tok/s/GPU at $0.28/M tokens, H200 runs 396 at $0.99/M. GB300 NVL72 is 257% cheaper per token; GB300 NVL72 delivers 763% more tok/s/GPU.
Setting 44 tok/s/user as the target on GLM 5/5.1, GB300 NVL72 produces 252 tok/s/GPU ($2.88 per million tokens) and H200 produces 306 ($1.28). H200 is 126% cheaper per token; H200 delivers 21% more tok/s/GPU.
At 56 tok/s/user interactivity on GLM 5/5.1, GB300 NVL72 delivers 83 tok/s/GPU at $8.90 per million tokens; H200 delivers 237 tok/s/GPU at $1.65. H200 is 441% cheaper per token; H200 delivers 186% more tok/s/GPU at this point. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:3416.6H200:395.7 | GB300 NVL72:252.1H200:306.0 | GB300 NVL72:83.0H200:237.4 |
| Cost ($/M tok) | GB300 NVL72:$0.277H200:$0.990 | GB300 NVL72:$2.880H200:$1.277 | GB300 NVL72:$8.903H200:$1.646 |
| tok/s/MW | GB300 NVL72:1626935H200:228711 | GB300 NVL72:120062H200:176875 | GB300 NVL72:39507H200:137204 |
| Concurrency | GB300 NVL72:~4993H200:~64 | GB300 NVL72:~229H200:~28 | GB300 NVL72:~61H200:~18 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.