DeepSeek R1 · Speculative decoding
GB300 NVL72 FP4: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on GB300 NVL72 FP4 (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.
At 59 tok/s/user on DeepSeek R1 (GB300 NVL72 FP4), MTP delivers 13,357 tok/s/chip at $0.048 per million tokens; Off delivers 6,821 tok/s/chip at $0.094. MTP delivers 96% more tok/s/chip on the same chip, which makes it 49% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 8,316 tok/s/chip for $0.077 per million tokens at 98 tok/s/user on DeepSeek R1 (GB300 NVL72 FP4); Off posts 560 tok/s/chip for $1.14. MTP delivers 15× the tok/s/chip on the same chip, which makes it 93% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 138 tok/s/user on DeepSeek R1 (GB300 NVL72 FP4): MTP hits 3,821 tok/s/chip, Off hits 180. Costs per million tokens land at $0.17 and $3.56 respectively. MTP delivers 21× the tok/s/chip on the same chip, which makes it 95% cheaper per token. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size. (Numbers reflect this URL's pinned 1k/1k · fp4 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP13,357Off6,821 | MTP8,316Off560 | MTP3,821Off180 |
| Cost ($/M tok) | MTP$0.048Off$0.094 | MTP$0.077Off$1.14 | MTP$0.17Off$3.56 |
| tok/s/MW | MTP6,300,484Off3,217,505 | MTP3,922,554Off264,382 | MTP1,802,447Off84,925 |
| Concurrency | MTP~2,484Off~3,022 | MTP~1,651Off~113 | MTP~702Off~8 |
