DeepSeek V4 Pro 0813 1.6T · 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 V4 Pro 0813 1.6T. 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 22–98 tok/s/user interactivity band, at 41 tok/s/user on DeepSeek V4 Pro 0813 1.6T (GB300 NVL72 FP4): MTP runs 11,812 tok/s/chip at $0.054 per million tokens, Off runs 8,636 at $0.074. MTP delivers 37% more tok/s/chip on the same chip, which makes it 27% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 60 tok/s/user on DeepSeek V4 Pro 0813 1.6T (GB300 NVL72 FP4), MTP delivers 9,971 tok/s/chip at $0.064 per million tokens; Off delivers 1,293 tok/s/chip at $0.50. MTP delivers 7.7× the tok/s/chip on the same chip, which makes it 87% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 8,434 tok/s/chip for $0.076 per million tokens at 79 tok/s/user on DeepSeek V4 Pro 0813 1.6T (GB300 NVL72 FP4); Off posts 561 tok/s/chip for $1.14. MTP delivers 15× the tok/s/chip on the same chip, which makes it 93% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 8k/1k · fp4 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP11,812Off8,636 | MTP9,971Off1,293 | MTP8,434Off561 |
| Cost ($/M tok) | MTP$0.054Off$0.074 | MTP$0.064Off$0.50 | MTP$0.076Off$1.14 |
| tok/s/MW | MTP5,571,916Off4,073,810 | MTP4,703,278Off610,044 | MTP3,978,393Off264,682 |
| Concurrency | MTP~5,450Off~2,390 | MTP~1,056Off~112 | MTP~1,132Off~25 |
