Qwen 3.5 397B-A17B · Speculative decoding
MI300X FP8: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on MI300X FP8 (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.
Off hits 346 tok/s/chip for $0.76 per million tokens at 43 tok/s/user on Qwen 3.5 397B-A17B (MI300X FP8). No MTP data at this operating point.
Off: 190 tok/s/chip, $1.39 per million tokens at 53 tok/s/user on Qwen 3.5 397B-A17B (MI300X FP8). MTP is unmeasured here.
At 62 tok/s/user on Qwen 3.5 397B-A17B (MI300X FP8), Off delivers 123 tok/s/chip at $2.15 per million tokens; MTP hasn't been benchmarked at this target. (Numbers reflect this URL's pinned 1k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP—Off346 | MTP—Off190 | MTP—Off123 |
| Cost ($/M tok) | MTP—Off$0.76 | MTP—Off$1.39 | MTP—Off$2.15 |
| tok/s/MW | MTP—Off248,748 | MTP—Off136,357 | MTP—Off88,306 |
| Concurrency | MTP—Off~34 | MTP—Off~15 | MTP—Off~8 |
