B200 FP4: MTP vs Off Speculative Decoding
Speculative decoding comparison of MTP versus Off on B200 FP4 (NVIDIA Blackwell) running DeepSeek V4 Pro 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.
Throughput at 32 tok/s/user on DeepSeek V4 Pro 1.6T (B200 FP4): MTP hits 3139 tok/s/GPU, Off hits 1893. Per-million costs land at $0.17 and $0.29 respectively. MTP is 64% cheaper per token; MTP delivers 66% more tok/s/GPU. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size.
Around the middle of the 9–103 tok/s/user interactivity band, at 56 tok/s/user on DeepSeek V4 Pro 1.6T (B200 FP4): MTP runs 2402 tok/s/GPU at $0.22/M tokens, Off runs 286 at $1.88/M. MTP is 744% cheaper per token; MTP delivers 741% more tok/s/GPU. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 80 tok/s/user on DeepSeek V4 Pro 1.6T (B200 FP4), MTP delivers 499 tok/s/GPU at $1.09 per million tokens; Off delivers 88 tok/s/GPU at $6.32. MTP is 479% cheaper per token; MTP delivers 469% more tok/s/GPU. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution. (Numbers reflect this URL's pinned 1k/1k · fp4 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MTP:3139.2Off:1893.4 | MTP:2402.2Off:285.7 | MTP:499.4Off:87.8 |
| Cost ($/M tok) | MTP:$0.174Off:$0.287 | MTP:$0.223Off:$1.883 | MTP:$1.091Off:$6.315 |
| tok/s/MW | MTP:1446651Off:872519 | MTP:1106998Off:131645 | MTP:230150Off:40475 |
| Concurrency | MTP:~580Off:~247 | MTP:~197Off:~25 | MTP:~31Off:~4 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.