DeepSeek R1 · Speculative decoding
GB200 NVL72 FP8: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on GB200 NVL72 FP8 (NVIDIA Blackwell) 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.
Near the low end of the 47–130 tok/s/user interactivity band, at 68 tok/s/user on DeepSeek R1 (GB200 NVL72 FP8): MTP runs 6,008 tok/s/chip at $0.086 per million tokens, Off runs 1,533 at $0.34. MTP delivers 3.9× the tok/s/chip on the same chip, which makes it 74% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 89 tok/s/user on DeepSeek R1 (GB200 NVL72 FP8), MTP delivers 5,149 tok/s/chip at $0.10 per million tokens; Off delivers 336 tok/s/chip at $1.54. MTP delivers 15× the tok/s/chip on the same chip, which makes it 93% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 4,430 tok/s/chip for $0.12 per million tokens at 110 tok/s/user on DeepSeek R1 (GB200 NVL72 FP8); Off posts 283 tok/s/chip for $1.83. MTP delivers 16× the tok/s/chip on the same chip, which makes it 94% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP6,008Off1,533 | MTP5,149Off336 | MTP4,430Off283 |
| Cost ($/M tok) | MTP$0.086Off$0.34 | MTP$0.10Off$1.54 | MTP$0.12Off$1.83 |
| tok/s/MW | MTP3,212,713Off819,566 | MTP2,753,252Off179,847 | MTP2,368,862Off151,379 |
| Concurrency | MTP~666Off~128 | MTP~368Off~10 | MTP~333Off~17 |
