Qwen 3.5 397B-A17B · Speculative decoding
MI325X BF16: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on MI325X BF16 (AMD CDNA 3) 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.
MTP posts 392 tok/s/chip for $0.78 per million tokens at 55 tok/s/user on Qwen 3.5 397B-A17B (MI325X BF16); Off posts 392 tok/s/chip for $0.78. Cost per token is essentially tied; throughput per chip is essentially tied. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 70 tok/s/user on Qwen 3.5 397B-A17B (MI325X BF16): MTP hits 274 tok/s/chip, Off hits 275. Costs per million tokens land at $1.11 and $1.11 respectively. Cost per token is essentially tied; throughput per chip is essentially tied. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size.
Toward the upper edge of the 41–98 tok/s/user interactivity band, at 84 tok/s/user on Qwen 3.5 397B-A17B (MI325X BF16): MTP runs 168 tok/s/chip at $1.82 per million tokens, Off runs 170 at $1.80. Cost per token is essentially tied; throughput per chip is essentially tied. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less. (Numbers reflect this URL's pinned 1k/1k · bf16 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP392Off392 | MTP274Off275 | MTP168Off170 |
| Cost ($/M tok) | MTP$0.78Off$0.78 | MTP$1.11Off$1.11 | MTP$1.82Off$1.80 |
| tok/s/MW | MTP232,183Off231,931 | MTP162,316Off162,727 | MTP99,399Off100,367 |
| Concurrency | MTP~29Off~29 | MTP~16Off~16 | MTP~8Off~8 |
