GLM 5/5.1 · Speculative decoding
MI325X FP8: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on MI325X FP8 (AMD CDNA 3) running GLM 5/5.1. 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.
Throughput at 21 tok/s/user on GLM 5/5.1 (MI325X FP8): MTP hits 186 tok/s/chip, Off hits 185. Costs per million tokens land at $1.64 and $1.65 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.
Around the middle of the 17–34 tok/s/user interactivity band, at 26 tok/s/user on GLM 5/5.1 (MI325X FP8): MTP runs 100 tok/s/chip at $3.07 per million tokens, Off runs 100 at $3.06. 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.
At 30 tok/s/user on GLM 5/5.1 (MI325X FP8), MTP delivers 62 tok/s/chip at $4.95 per million tokens; Off delivers 62 tok/s/chip at $4.93. Cost per token is essentially tied; throughput per chip is essentially tied. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution. (Numbers reflect this URL's pinned 1k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP186Off185 | MTP100Off100 | MTP62Off62 |
| Cost ($/M tok) | MTP$1.64Off$1.65 | MTP$3.07Off$3.06 | MTP$4.95Off$4.93 |
| tok/s/MW | MTP110,051Off109,670 | MTP58,882Off59,083 | MTP36,512Off36,677 |
| Concurrency | MTP~39Off~38 | MTP~16Off~16 | MTP~9Off~9 |
