GB300 NVL72 FP8: M3 EAGLE vs Off Speculative Decoding
Speculative decoding comparison of M3 EAGLE versus Off on GB300 NVL72 FP8 (NVIDIA Blackwell) 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 62–190 tok/s/user interactivity band, at 94 tok/s/user on MiniMax M3 428B (GB300 NVL72 FP8): M3 EAGLE runs 12034 tok/s/chip at $0.05/M tokens, Off runs 2598 at $0.25/M. M3 EAGLE is 362% cheaper per token; M3 EAGLE delivers 363% more tok/s/chip. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.
At 126 tok/s/user on MiniMax M3 428B (GB300 NVL72 FP8), M3 EAGLE delivers 4542 tok/s/chip at $0.15 per million tokens; Off delivers 1306 tok/s/chip at $0.48. M3 EAGLE is 218% cheaper per token; M3 EAGLE delivers 248% more tok/s/chip. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.
M3 EAGLE posts 2876 tok/s/chip for $0.25 per million tokens at 159 tok/s/user on MiniMax M3 428B (GB300 NVL72 FP8); Off posts 655 tok/s/chip for $1.05. M3 EAGLE is 314% cheaper per token; M3 EAGLE delivers 339% more tok/s/chip. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (Numbers reflect this URL's pinned 8k/1k · fp8 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/chip) | M3 EAGLE:12034.4Off:2598.3 | M3 EAGLE:4542.4Off:1306.4 | M3 EAGLE:2875.7Off:654.7 |
| Cost ($/M tok) | M3 EAGLE:$0.053Off:$0.246 | M3 EAGLE:$0.152Off:$0.482 | M3 EAGLE:$0.254Off:$1.052 |
| tok/s/MW | M3 EAGLE:5676591Off:1225595 | M3 EAGLE:2142652Off:616238 | M3 EAGLE:1356463Off:308827 |
| Concurrency | M3 EAGLE:~579Off:~67 | M3 EAGLE:~59Off:~24 | M3 EAGLE:~24Off:~10 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.