Qwen 3.5 397B-A17B · Speculative decoding
B200 BF16: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on B200 BF16 (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.
Near the low end of the 103–168 tok/s/user interactivity band, at 119 tok/s/user on Qwen 3.5 397B-A17B (B200 BF16): MTP runs 1,058 tok/s/chip at $0.45 per million tokens, Off runs 270 at $1.78. 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 135 tok/s/user on Qwen 3.5 397B-A17B (B200 BF16), MTP delivers 828 tok/s/chip at $0.58 per million tokens; Off delivers 214 tok/s/chip at $2.25. MTP delivers 3.9× the tok/s/chip on the same chip, which makes it 74% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 732 tok/s/chip for $0.66 per million tokens at 152 tok/s/user on Qwen 3.5 397B-A17B (B200 BF16); Off posts 178 tok/s/chip for $2.70. MTP delivers 4.1× the tok/s/chip on the same chip, which makes it 76% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 1k/1k · bf16 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP1,058Off270 | MTP828Off214 | MTP732Off178 |
| Cost ($/M tok) | MTP$0.45Off$1.78 | MTP$0.58Off$2.25 | MTP$0.66Off$2.70 |
| tok/s/MW | MTP618,848Off158,129 | MTP484,300Off124,930 | MTP427,931Off103,964 |
| Concurrency | MTP~40Off~10 | MTP~27Off~7 | MTP~22Off~5 |
