DeepSeek R1 — GB300 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB300 NVL72 (NVIDIA Blackwell) and H200 (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.
Setting 59 tok/s/user as the target on DeepSeek R1, GB300 NVL72 produces 6244 tok/s/GPU ($0.12 per million tokens) and H200 produces 422 ($0.93). GB300 NVL72 is 686% cheaper per token; GB300 NVL72 delivers 1379% more tok/s/GPU.
At 98 tok/s/user interactivity on DeepSeek R1, GB300 NVL72 delivers 2579 tok/s/GPU at $0.29 per million tokens; H200 delivers 152 tok/s/GPU at $2.66. GB300 NVL72 is 832% cheaper per token; GB300 NVL72 delivers 1596% more tok/s/GPU at this point.
GB300 NVL72 posts 669 tok/s/GPU for $1.19 per million tokens at 136 tok/s/user on DeepSeek R1; H200 posts 40 tok/s/GPU for $9.94. GB300 NVL72 is 733% cheaper per token; GB300 NVL72 delivers 1591% more tok/s/GPU. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | GB300 NVL72:6244.3H200:422.2 | GB300 NVL72:2579.2H200:152.1 | GB300 NVL72:669.1H200:39.6 |
| Cost ($/M tok) | GB300 NVL72:$0.118H200:$0.929 | GB300 NVL72:$0.286H200:$2.663 | GB300 NVL72:$1.193H200:$9.940 |
| tok/s/MW | GB300 NVL72:2973478H200:244043 | GB300 NVL72:1228171H200:87929 | GB300 NVL72:318638H200:22869 |
| Concurrency | GB300 NVL72:~1402H200:~133 | GB300 NVL72:~578H200:~6 | GB300 NVL72:~134H200:~14 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.