DeepSeek R1 · Speculative decoding
H200 FP8: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on H200 FP8 (NVIDIA Hopper) running DeepSeek R1. Throughput, cost, and interactivity differences across LLM workloads. Use the chart controls below to switch sequences and metrics — same interactions as the main inference chart.
MTP acceptance-rate implementations differ across inference engines. Points from different engines are not directly comparable on the same curve — throughput and cost at matched interactivity may reflect engine-level differences rather than pure speculative decoding gains. Interpret cross-engine comparisons with caution.
MTP posts 1,888 tok/s/chip for $0.18 per million tokens at 39 tok/s/user on DeepSeek R1 (H200 FP8); Off posts 977 tok/s/chip for $0.35. MTP delivers 93% more tok/s/chip on the same chip, which makes it 48% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 62 tok/s/user on DeepSeek R1 (H200 FP8): MTP hits 1,124 tok/s/chip, Off hits 734. Costs per million tokens land at $0.30 and $0.46 respectively. MTP delivers 53% more tok/s/chip on the same chip, which makes it 35% cheaper per token. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size.
Toward the upper edge of the 18–106 tok/s/user interactivity band, at 84 tok/s/user on DeepSeek R1 (H200 FP8): MTP runs 836 tok/s/chip at $0.41 per million tokens, Off runs 508 at $0.67. MTP delivers 64% more tok/s/chip on the same chip, which makes it 39% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP1,888Off977 | MTP1,124Off734 | MTP836Off508 |
| Cost ($/M tok) | MTP$0.18Off$0.35 | MTP$0.30Off$0.46 | MTP$0.41Off$0.67 |
| tok/s/MW | MTP1,378,310Off713,435 | MTP820,285Off535,927 | MTP610,172Off371,130 |
| Concurrency | MTP~115Off~23 | MTP~160Off~11 | MTP~29Off~5 |
