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 98 tok/s/user as the target on gpt-oss 120B, GB200 NVL72 produces 32977 tok/s/GPU ($0.02 per million tokens) and H200 produces 3421 ($0.11). GB200 NVL72 is 517% cheaper per token; GB200 NVL72 delivers 864% more tok/s/GPU.
At 156 tok/s/user interactivity on gpt-oss 120B, GB200 NVL72 delivers 17595 tok/s/GPU at $0.03 per million tokens; H200 delivers 1692 tok/s/GPU at $0.24. GB200 NVL72 is 577% cheaper per token; GB200 NVL72 delivers 940% more tok/s/GPU at this point.
GB200 NVL72 posts 8620 tok/s/GPU for $0.07 per million tokens at 215 tok/s/user on gpt-oss 120B; H200 posts 803 tok/s/GPU for $0.49. GB200 NVL72 is 600% cheaper per token; GB200 NVL72 delivers 973% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:32977.4H200:3421.3 | GB200 NVL72:17595.4H200:1692.0 | GB200 NVL72:8620.4H200:803.2 |
| Cost ($/M tok) | GB200 NVL72:$0.019H200:$0.114 | GB200 NVL72:$0.035H200:$0.236 | GB200 NVL72:$0.070H200:$0.488 |
| tok/s/MW | GB200 NVL72:15703542H200:1977650 | GB200 NVL72:8378763H200:978028 | GB200 NVL72:4104941H200:464280 |
| Concurrency | GB200 NVL72:~656H200:~64 | GB200 NVL72:~3067H200:~42 | GB200 NVL72:~103H200:~15 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.