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.
H200 hits 329 tok/s/GPU for $1.19 per million tokens at 41 tok/s/user on GLM 5/5.1. No GB300 NVL72 data at this operating point.
H200: 212 tok/s/GPU, $1.85 per million tokens at 61 tok/s/user on GLM 5/5.1. GB300 NVL72 is unmeasured here.
At 82 tok/s/user on GLM 5/5.1, H200 delivers 132 tok/s/GPU at $2.93 per million tokens; GB300 NVL72 hasn't been benchmarked at this target. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:—H200:329.5 | GB300 NVL72:—H200:212.5 | GB300 NVL72:—H200:132.4 |
| Cost ($/M tok) | GB300 NVL72:—H200:$1.190 | GB300 NVL72:—H200:$1.847 | GB300 NVL72:—H200:$2.933 |
| tok/s/MW | GB300 NVL72:—H200:190452 | GB300 NVL72:—H200:122811 | GB300 NVL72:—H200:76544 |
| Concurrency | GB300 NVL72:—H200:~34 | GB300 NVL72:—H200:~14 | GB300 NVL72:—H200:~7 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.