Qwen 3.5 397B-A17B · Speculative decoding
MI355X BF16: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on MI355X BF16 (AMD CDNA 4) 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 78 tok/s/user on Qwen 3.5 397B-A17B (MI355X BF16), MTP delivers 786 tok/s/chip at $0.53 per million tokens; Off delivers 339 tok/s/chip at $1.23. MTP delivers 132% more tok/s/chip on the same chip, which makes it 57% 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 $0.57 per million tokens at 86 tok/s/user on Qwen 3.5 397B-A17B (MI355X BF16); Off posts 254 tok/s/chip for $1.64. MTP delivers 188% more tok/s/chip on the same chip, which makes it 65% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 95 tok/s/user on Qwen 3.5 397B-A17B (MI355X BF16): MTP hits 648 tok/s/chip, Off hits 164. Costs per million tokens land at $0.64 and $2.54 respectively. MTP delivers 3.9× the tok/s/chip on the same chip, which makes it 75% cheaper per token. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size. (Numbers reflect this URL's pinned 1k/1k · bf16 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP786Off339 | MTP732Off254 | MTP648Off164 |
| Cost ($/M tok) | MTP$0.53Off$1.23 | MTP$0.57Off$1.64 | MTP$0.64Off$2.54 |
| tok/s/MW | MTP375,938Off162,240 | MTP350,044Off121,455 | MTP309,875Off78,503 |
| Concurrency | MTP~87Off~18 | MTP~50Off~12 | MTP~29Off~7 |
