DeepSeek R1 — GB200 NVL72 vs H100
Head-to-head AI inference benchmark comparison of GB200 NVL72 (NVIDIA Blackwell) and H100 (NVIDIA Hopper) on DeepSeek R1. 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 posts 5383 tok/s/GPU for $0.11 per million tokens at 41 tok/s/user on DeepSeek R1; H100 posts 269 tok/s/GPU for $1.35. GB200 NVL72 is 1082% cheaper per token; GB200 NVL72 delivers 1898% more tok/s/GPU.
Throughput at 70 tok/s/user on DeepSeek R1: GB200 NVL72 hits 4419 tok/s/GPU, H100 hits 77. Per-million costs land at $0.14 and $4.65 respectively. GB200 NVL72 is 3251% cheaper per token; GB200 NVL72 delivers 5613% more tok/s/GPU.
GB200 NVL72 / H100 on DeepSeek R1 at 99 tok/s/user: 2387 / 23 tok/s/GPU, $0.26 / $15.6 per million tokens. GB200 NVL72 is 5974% cheaper per token; GB200 NVL72 delivers 10126% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB200 NVL72:5382.7H100:269.4 | GB200 NVL72:4419.4H100:77.4 | GB200 NVL72:2387.3H100:23.3 |
| Cost ($/M tok) | GB200 NVL72:$0.114H100:$1.351 | GB200 NVL72:$0.139H100:$4.655 | GB200 NVL72:$0.257H100:$15.632 |
| tok/s/MW | GB200 NVL72:2563185H100:155721 | GB200 NVL72:2104473H100:44718 | GB200 NVL72:1136786H100:13494 |
| Concurrency | GB200 NVL72:~3266H100:~584 | GB200 NVL72:~1006H100:~57 | GB200 NVL72:~612H100:~8 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.