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 97 tok/s/user as the target on gpt-oss 120B, GB200 NVL72 produces 33471 tok/s/GPU ($0.02 per million tokens) and H200 produces 3459 ($0.11). GB200 NVL72 is 519% cheaper per token; GB200 NVL72 delivers 868% more tok/s/GPU.
At 155 tok/s/user interactivity on gpt-oss 120B, GB200 NVL72 delivers 17665 tok/s/GPU at $0.03 per million tokens; H200 delivers 1797 tok/s/GPU at $0.22. GB200 NVL72 is 527% cheaper per token; GB200 NVL72 delivers 883% more tok/s/GPU at this point.
GB200 NVL72 posts 8801 tok/s/GPU for $0.07 per million tokens at 213 tok/s/user on gpt-oss 120B; H200 posts 846 tok/s/GPU for $0.46. GB200 NVL72 is 573% cheaper per token; GB200 NVL72 delivers 940% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:33470.6H200:3459.0 | GB200 NVL72:17665.1H200:1796.9 | GB200 NVL72:8801.0H200:846.5 |
| Cost ($/M tok) | GB200 NVL72:$0.018H200:$0.113 | GB200 NVL72:$0.035H200:$0.218 | GB200 NVL72:$0.069H200:$0.461 |
| tok/s/MW | GB200 NVL72:15938358H200:1999433 | GB200 NVL72:8411972H200:1038660 | GB200 NVL72:4190935H200:489288 |
| Concurrency | GB200 NVL72:~775H200:~64 | GB200 NVL72:~3072H200:~55 | GB200 NVL72:~107H200:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.