DeepSeek R1 · Chip comparison
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 65 tok/s/user interactivity on DeepSeek R1, GB200 NVL72 delivers 6,057 tok/s/chip at $0.085 per million tokens; H200 delivers 1,059 tok/s/chip at $0.32. GB200 NVL72 is 73% cheaper per token; GB200 NVL72 delivers 5.7× the tok/s/chip at this point.
GB200 NVL72 posts 4,759 tok/s/chip for $0.11 per million tokens at 99 tok/s/user on DeepSeek R1; H200 posts 712 tok/s/chip for $0.48. GB200 NVL72 is 77% cheaper per token; GB200 NVL72 delivers 6.7× the tok/s/chip.
Throughput at 133 tok/s/user on DeepSeek R1: GB200 NVL72 hits 2,941 tok/s/chip, H200 hits 489. Costs per million tokens land at $0.18 and $0.69 respectively. GB200 NVL72 is 75% cheaper per token; GB200 NVL72 delivers 6.0× the tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | GB200 NVL726,057H2001,059 | GB200 NVL724,759H200712 | GB200 NVL722,941H200489 |
| Cost ($/M tok) | GB200 NVL72$0.085H200$0.32 | GB200 NVL72$0.11H200$0.48 | GB200 NVL72$0.18H200$0.69 |
| tok/s/MW | GB200 NVL723,238,968H200772,633 | GB200 NVL722,544,727H200519,628 | GB200 NVL721,572,785H200356,982 |
| Concurrency | GB200 NVL72~975H200~155 | GB200 NVL72~333H200~6 | GB200 NVL72~242H200~4 |