GB300 NVL72 FP4: MTP vs Off Speculative Decoding
Speculative decoding comparison of MTP versus Off on GB300 NVL72 FP4 (NVIDIA Blackwell) running DeepSeek V4 Pro 1.6T. 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.
MTP posts 11812 tok/s/GPU for $0.06 per million tokens at 41 tok/s/user on DeepSeek V4 Pro 1.6T (GB300 NVL72 FP4); Off posts 6947 tok/s/GPU for $0.11. MTP is 71% cheaper per token; MTP delivers 70% more tok/s/GPU. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.
Throughput at 60 tok/s/user on DeepSeek V4 Pro 1.6T (GB300 NVL72 FP4): MTP hits 9971 tok/s/GPU, Off hits 1293. Per-million costs land at $0.07 and $0.57 respectively. MTP is 667% cheaper per token; MTP delivers 671% 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 22–98 tok/s/user interactivity band, at 79 tok/s/user on DeepSeek V4 Pro 1.6T (GB300 NVL72 FP4): MTP runs 8434 tok/s/GPU at $0.09/M tokens, Off runs 561 at $1.31/M. MTP is 1398% cheaper per token; MTP delivers 1403% more tok/s/GPU. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less. (Numbers reflect this URL's pinned 8k/1k · fp4 workload — changing sequence or model updates both the table and chart; the table stays pinned to this page's precision, so precision toggles in the controls affect the chart only.)
| Metric | Interactivity (tok/s/user) | Interactivity (tok/s/user) | Interactivity (tok/s/user) |
|---|---|---|---|
| Throughput (tok/s/gpu) | MTP:11812.5Off:6947.0 | MTP:9970.9Off:1293.3 | MTP:8434.2Off:561.1 |
| Cost ($/M tok) | MTP:$0.062Off:$0.106 | MTP:$0.074Off:$0.567 | MTP:$0.088Off:$1.312 |
| tok/s/MW | MTP:5624981Off:3308104 | MTP:4748071Off:615854 | MTP:4016282Off:267203 |
| Concurrency | MTP:~5450Off:~2254 | MTP:~1056Off:~112 | MTP:~1132Off:~25 |
Inference Performance
Inference performance metrics across different models, hardware configurations, and serving parameters.