MI355X FP8: M3 EAGLE vs Off Speculative Decoding
Speculative decoding comparison of M3 EAGLE versus Off on MI355X FP8 (AMD CDNA 4) 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.
Near the low end of the 5–148 tok/s/user interactivity band, at 41 tok/s/user on MiniMax M3 428B (MI355X FP8): M3 EAGLE runs 6400 tok/s/GPU at $0.06/M tokens, Off runs 4713 at $0.09/M. M3 EAGLE is 35% cheaper per token; M3 EAGLE delivers 36% more tok/s/GPU. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 77 tok/s/user on MiniMax M3 428B (MI355X FP8), M3 EAGLE delivers 4450 tok/s/GPU at $0.09 per million tokens; Off delivers 2152 tok/s/GPU at $0.19. M3 EAGLE is 107% cheaper per token; M3 EAGLE delivers 107% more tok/s/GPU. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
M3 EAGLE posts 3387 tok/s/GPU for $0.12 per million tokens at 113 tok/s/user on MiniMax M3 428B (MI355X FP8); Off posts 1146 tok/s/GPU for $0.35. M3 EAGLE is 187% cheaper per token; M3 EAGLE delivers 196% more tok/s/GPU. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | M3 EAGLE:6400.0Off:4712.9 | M3 EAGLE:4450.4Off:2151.9 | M3 EAGLE:3387.3Off:1146.0 |
| Cost ($/M tok) | M3 EAGLE:$0.064Off:$0.087 | M3 EAGLE:$0.092Off:$0.191 | M3 EAGLE:$0.121Off:$0.348 |
| tok/s/MW | M3 EAGLE:3062180Off:2254984 | M3 EAGLE:2129380Off:1029601 | M3 EAGLE:1620699Off:548314 |
| Concurrency | M3 EAGLE:~83Off:~108 | M3 EAGLE:~31Off:~13 | M3 EAGLE:~16Off:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.