H200 FP8: M3 EAGLE vs Off Speculative Decoding
Speculative decoding comparison of M3 EAGLE versus Off on H200 FP8 (NVIDIA Hopper) 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.
At 41 tok/s/user on MiniMax M3 428B (H200 FP8), M3 EAGLE delivers 1263 tok/s/GPU at $0.31 per million tokens; Off delivers 715 tok/s/GPU at $0.55. M3 EAGLE is 76% cheaper per token; M3 EAGLE delivers 77% more tok/s/GPU. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
M3 EAGLE posts 699 tok/s/GPU for $0.55 per million tokens at 76 tok/s/user on MiniMax M3 428B (H200 FP8); Off posts 331 tok/s/GPU for $1.18. M3 EAGLE is 114% cheaper per token; M3 EAGLE delivers 111% more tok/s/GPU. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 112 tok/s/user on MiniMax M3 428B (H200 FP8): M3 EAGLE hits 461 tok/s/GPU, Off hits 164. Per-million costs land at $0.86 and $2.33 respectively. M3 EAGLE is 171% cheaper per token; M3 EAGLE delivers 181% more tok/s/GPU. Speculative decoding trades extra compute on draft tokens for fewer decoding steps — the payoff depends on sequence length and batch size. (Numbers reflect this URL's pinned 1k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | M3 EAGLE:1262.7Off:715.0 | M3 EAGLE:699.4Off:331.3 | M3 EAGLE:461.2Off:164.1 |
| Cost ($/M tok) | M3 EAGLE:$0.310Off:$0.546 | M3 EAGLE:$0.551Off:$1.177 | M3 EAGLE:$0.859Off:$2.331 |
| tok/s/MW | M3 EAGLE:729887Off:413320 | M3 EAGLE:404301Off:191513 | M3 EAGLE:266616Off:94877 |
| Concurrency | M3 EAGLE:~69Off:~36 | M3 EAGLE:~20Off:~9 | M3 EAGLE:~9Off:~3 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.