DeepSeek R1 · Speculative decoding
MI355X FP4: MTP vs Off speculative decoding
Speculative decoding comparison of MTP versus Off on MI355X FP4 (AMD CDNA 4) running DeepSeek R1. 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 49 tok/s/user on DeepSeek R1 (MI355X FP4), MTP delivers 2,089 tok/s/chip at $0.20 per million tokens; Off delivers 1,589 tok/s/chip at $0.26. MTP delivers 32% more tok/s/chip on the same chip, which makes it 24% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
MTP posts 1,682 tok/s/chip for $0.25 per million tokens at 85 tok/s/user on DeepSeek R1 (MI355X FP4); Off posts 512 tok/s/chip for $0.81. MTP delivers 3.3× the tok/s/chip on the same chip, which makes it 70% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 121 tok/s/user on DeepSeek R1 (MI355X FP4): MTP hits 461 tok/s/chip, Off hits 132. Costs per million tokens land at $0.90 and $3.15 respectively. MTP delivers 3.5× the tok/s/chip on the same chip, which makes it 71% cheaper per token. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size. (Numbers reflect this URL's pinned 1k/1k · fp4 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | MTP2,089Off1,589 | MTP1,682Off512 | MTP461Off132 |
| Cost ($/M tok) | MTP$0.20Off$0.26 | MTP$0.25Off$0.81 | MTP$0.90Off$3.15 |
| tok/s/MW | MTP999,616Off760,051 | MTP804,804Off245,078 | MTP220,696Off63,363 |
| Concurrency | MTP~153Off~223 | MTP~256Off~61 | MTP~16Off~10 |
