gpt-oss 120B — GB200 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and H200 (NVIDIA Hopper) on gpt-oss 120B. 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.
Setting 101 tok/s/user as the target on gpt-oss 120B, GB200 NVL72 produces 30885 tok/s/GPU ($0.02 per million tokens) and H200 produces 3322 ($0.12). GB200 NVL72 is 488% cheaper per token; GB200 NVL72 delivers 830% more tok/s/GPU.
At 157 tok/s/user interactivity on gpt-oss 120B, GB200 NVL72 delivers 17516 tok/s/GPU at $0.04 per million tokens; H200 delivers 1753 tok/s/GPU at $0.22. GB200 NVL72 is 537% cheaper per token; GB200 NVL72 delivers 899% more tok/s/GPU at this point.
GB200 NVL72 posts 8710 tok/s/GPU for $0.07 per million tokens at 214 tok/s/user on gpt-oss 120B; H200 posts 834 tok/s/GPU for $0.47. GB200 NVL72 is 578% cheaper per token; GB200 NVL72 delivers 944% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:30884.9H200:3322.0 | GB200 NVL72:17516.2H200:1753.2 | GB200 NVL72:8709.5H200:834.3 |
| Cost ($/M tok) | GB200 NVL72:$0.020H200:$0.118 | GB200 NVL72:$0.035H200:$0.223 | GB200 NVL72:$0.069H200:$0.469 |
| tok/s/MW | GB200 NVL72:14707096H200:1920239 | GB200 NVL72:8341061H200:1013432 | GB200 NVL72:4147398H200:482244 |
| Concurrency | GB200 NVL72:~512H200:~64 | GB200 NVL72:~3055H200:~53 | GB200 NVL72:~105H200:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.