DeepSeek R1 — GB200 NVL72 vs H200
Head-to-head AI inference benchmark comparison of GB200 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.
At 59 tok/s/user interactivity on DeepSeek R1, GB200 NVL72 delivers 4972 tok/s/chip at $0.10 per million tokens; H200 delivers 421 tok/s/chip at $0.80. GB200 NVL72 is 676% cheaper per token; GB200 NVL72 delivers 1082% more tok/s/chip at this point.
GB200 NVL72 posts 2445 tok/s/chip for $0.21 per million tokens at 98 tok/s/user on DeepSeek R1; H200 posts 152 tok/s/chip for $2.30. GB200 NVL72 is 993% cheaper per token; GB200 NVL72 delivers 1507% more tok/s/chip.
Throughput at 136 tok/s/user on DeepSeek R1: GB200 NVL72 hits 494 tok/s/chip, H200 hits 40. Per-million costs land at $1.58 and $8.60 respectively. GB200 NVL72 is 445% cheaper per token; GB200 NVL72 delivers 1150% more tok/s/chip. (Numbers reflect the default 1k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL72:4972.2H200:420.5 | GB200 NVL72:2444.8H200:152.1 | GB200 NVL72:494.4H200:39.6 |
| Cost ($/M tok) | GB200 NVL72:$0.104H200:$0.805 | GB200 NVL72:$0.211H200:$2.304 | GB200 NVL72:$1.579H200:$8.600 |
| tok/s/MW | GB200 NVL72:2658923H200:306952 | GB200 NVL72:1307377H200:111034 | GB200 NVL72:264379H200:28878 |
| Concurrency | GB200 NVL72:~1335H200:~133 | GB200 NVL72:~627H200:~6 | GB200 NVL72:~108H200:~14 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.