DeepSeek R1 · Chip comparison
DeepSeek R1 — B300 vs H200
Head-to-head AI inference benchmark comparison of B300 (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 51 tok/s/user as the target on DeepSeek R1, B300 produces 5,802 tok/s/chip ($0.11 per million tokens) and H200 produces 1,499 ($0.23). B300 is 52% cheaper per token; B300 delivers 3.9× the tok/s/chip.
At 90 tok/s/user interactivity on DeepSeek R1, B300 delivers 2,231 tok/s/chip at $0.28 per million tokens; H200 delivers 785 tok/s/chip at $0.43. B300 is 35% cheaper per token; B300 delivers 184% more tok/s/chip at this point.
B300 posts 1,079 tok/s/chip for $0.58 per million tokens at 129 tok/s/user on DeepSeek R1; H200 posts 523 tok/s/chip for $0.65. B300 is 10% cheaper per token; B300 delivers 106% more tok/s/chip. (Numbers reflect the default 8k/1k · fp8 selection for this URL — table and chart below update if you change sequence, precision, or model in the controls.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | B3005,802H2001,499 | B3002,231H200785 | B3001,079H200523 |
| Cost ($/M tok) | B300$0.11H200$0.23 | B300$0.28H200$0.43 | B300$0.58H200$0.65 |
| tok/s/MW | B3003,053,886H2001,094,512 | B3001,174,362H200573,273 | B300568,060H200381,794 |
| Concurrency | B300~277H200~54 | B300~55H200~8 | B300~8H200~4 |