gpt-oss 120B — B200 vs GB200 NVL72
Head-to-head AI inference benchmark comparison of B200 (NVIDIA Blackwell) and GB200 NVL72 (NVIDIA Blackwell) 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.
B200 posts 10903 tok/s/chip for $0.04 per million tokens at 152 tok/s/user on gpt-oss 120B; GB200 NVL72 posts 17836 tok/s/chip for $0.03. GB200 NVL72 is 53% cheaper per token; GB200 NVL72 delivers 64% more tok/s/chip.
Throughput at 245 tok/s/user on gpt-oss 120B: B200 hits 4212 tok/s/chip, GB200 NVL72 hits 6536. Per-million costs land at $0.12 and $0.08 respectively. GB200 NVL72 is 48% cheaper per token; GB200 NVL72 delivers 55% more tok/s/chip.
B200 / GB200 NVL72 on gpt-oss 120B at 338 tok/s/user: 1769 / 2259 tok/s/chip, $0.27 / $0.23 per million tokens. GB200 NVL72 is 21% cheaper per token; GB200 NVL72 delivers 28% more tok/s/chip. (Numbers reflect the default 1k/1k · fp4 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B200:10903.3GB200 NVL72:17835.7 | B200:4211.5GB200 NVL72:6535.9 | B200:1769.3GB200 NVL72:2258.7 |
| Cost ($/M tok) | B200:$0.044GB200 NVL72:$0.029 | B200:$0.115GB200 NVL72:$0.078 | B200:$0.272GB200 NVL72:$0.225 |
| tok/s/MW | B200:6376211GB200 NVL72:9537828 | B200:2462878GB200 NVL72:3495151 | B200:1034655GB200 NVL72:1207849 |
| Concurrency | B200:~68GB200 NVL72:~3024 | B200:~37GB200 NVL72:~71 | B200:~11GB200 NVL72:~16 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.