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 18–150 tok/s/user interactivity band, at 51 tok/s/user on MiniMax M3 428B (MI355X FP8): M3 EAGLE runs 2709 tok/s/GPU at $0.15/M tokens, Off runs 1248 at $0.33/M. M3 EAGLE is 118% cheaper per token; M3 EAGLE delivers 117% more tok/s/GPU. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 84 tok/s/user on MiniMax M3 428B (MI355X FP8), M3 EAGLE delivers 1741 tok/s/GPU at $0.24 per million tokens; Off delivers 576 tok/s/GPU at $0.72. M3 EAGLE is 203% cheaper per token; M3 EAGLE delivers 202% more tok/s/GPU. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
M3 EAGLE posts 1135 tok/s/GPU for $0.36 per million tokens at 117 tok/s/user on MiniMax M3 428B (MI355X FP8); Off posts 290 tok/s/GPU for $1.37. M3 EAGLE is 278% cheaper per token; M3 EAGLE delivers 291% more tok/s/GPU. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 1k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | M3 EAGLE:2709.0Off:1247.5 | M3 EAGLE:1741.4Off:576.0 | M3 EAGLE:1134.6Off:290.1 |
| Cost ($/M tok) | M3 EAGLE:$0.152Off:$0.330 | M3 EAGLE:$0.236Off:$0.716 | M3 EAGLE:$0.363Off:$1.369 |
| tok/s/MW | M3 EAGLE:1022272Off:470760 | M3 EAGLE:657124Off:217376 | M3 EAGLE:428168Off:109470 |
| Concurrency | M3 EAGLE:~112Off:~49 | M3 EAGLE:~44Off:~14 | M3 EAGLE:~21Off:~5 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.