GB200 NVL72 FP8: MTP vs Off Speculative Decoding
Speculative decoding comparison of MTP versus Off on GB200 NVL72 FP8 (NVIDIA Blackwell) running Qwen 3.5 397B-A17B. 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 89 tok/s/user on Qwen 3.5 397B-A17B (GB200 NVL72 FP8), MTP delivers 10304 tok/s/chip at $0.05 per million tokens; Off delivers 3473 tok/s/chip at $0.15. MTP is 198% cheaper per token; MTP delivers 197% more tok/s/chip. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 7443 tok/s/chip for $0.07 per million tokens at 131 tok/s/user on Qwen 3.5 397B-A17B (GB200 NVL72 FP8); Off posts 1842 tok/s/chip for $0.28. MTP is 296% cheaper per token; MTP delivers 304% more tok/s/chip. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 173 tok/s/user on Qwen 3.5 397B-A17B (GB200 NVL72 FP8): MTP hits 3779 tok/s/chip, Off hits 935. Per-million costs land at $0.14 and $0.55 respectively. MTP is 304% cheaper per token; MTP delivers 304% more tok/s/chip. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP:10304.0Off:3472.9 | MTP:7443.2Off:1841.9 | MTP:3779.1Off:935.2 |
| Cost ($/M tok) | MTP:$0.050Off:$0.149 | MTP:$0.070Off:$0.278 | MTP:$0.137Off:$0.552 |
| tok/s/MW | MTP:5510151Off:1857146 | MTP:3980305Off:984954 | MTP:2020915Off:500090 |
| Concurrency | MTP:~1083Off:~38 | MTP:~401Off:~13 | MTP:~49Off:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.