MI300X FP8: M3 EAGLE vs Off Speculative Decoding
Speculative decoding comparison of M3 EAGLE versus Off on MI300X FP8 (AMD CDNA 3) running MiniMax M3 428B. 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.
M3 EAGLE posts 1176 tok/s/GPU for $0.26 per million tokens at 27 tok/s/user on MiniMax M3 428B (MI300X FP8); Off posts 860 tok/s/GPU for $0.36. M3 EAGLE is 37% cheaper per token; M3 EAGLE delivers 37% more tok/s/GPU. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 49 tok/s/user on MiniMax M3 428B (MI300X FP8): M3 EAGLE hits 689 tok/s/GPU, Off hits 531. Per-million costs land at $0.45 and $0.59 respectively. M3 EAGLE is 30% cheaper per token; M3 EAGLE delivers 30% more tok/s/GPU. 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 6–93 tok/s/user interactivity band, at 71 tok/s/user on MiniMax M3 428B (MI300X FP8): M3 EAGLE runs 642 tok/s/GPU at $0.49/M tokens, Off runs 353 at $0.86/M. M3 EAGLE is 78% cheaper per token; M3 EAGLE delivers 82% more tok/s/GPU. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | M3 EAGLE:1175.7Off:859.7 | M3 EAGLE:688.8Off:531.0 | M3 EAGLE:642.4Off:353.2 |
| Cost ($/M tok) | M3 EAGLE:$0.264Off:$0.363 | M3 EAGLE:$0.453Off:$0.591 | M3 EAGLE:$0.486Off:$0.865 |
| tok/s/MW | M3 EAGLE:656810Off:480275 | M3 EAGLE:384790Off:296637 | M3 EAGLE:358867Off:197321 |
| Concurrency | M3 EAGLE:~47Off:~29 | M3 EAGLE:~14Off:~10 | M3 EAGLE:~10Off:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.