GLM 5/5.1 — GB200 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB200 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.
GB200 NVL72 / H200 on GLM 5/5.1 at 37 tok/s/user: 2074 / 705 tok/s/GPU, $0.31 / $0.56 per million tokens. GB200 NVL72 is 79% cheaper per token; GB200 NVL72 delivers 194% more tok/s/GPU.
Around the middle of the 29–62 tok/s/user interactivity band, at 45 tok/s/user on GLM 5/5.1: GB200 NVL72 runs 635 tok/s/GPU at $0.96/M tokens, H200 runs 620 at $0.63/M. H200 is 51% cheaper per token; GB200 NVL72 delivers 2% more tok/s/GPU.
Setting 54 tok/s/user as the target on GLM 5/5.1, GB200 NVL72 produces 158 tok/s/GPU ($4.50 per million tokens) and H200 produces 537 ($0.73). H200 is 515% cheaper per token; H200 delivers 240% more tok/s/GPU. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:2073.9H200:705.2 | GB200 NVL72:635.4H200:620.3 | GB200 NVL72:157.8H200:536.9 |
| Cost ($/M tok) | GB200 NVL72:$0.311H200:$0.556 | GB200 NVL72:$0.957H200:$0.632 | GB200 NVL72:$4.499H200:$0.731 |
| tok/s/MW | GB200 NVL72:987580H200:407621 | GB200 NVL72:302553H200:358552 | GB200 NVL72:75151H200:310353 |
| Concurrency | GB200 NVL72:~931H200:~18 | GB200 NVL72:~150H200:~13 | GB200 NVL72:~27H200:~10 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.