DeepSeek R1 · Speculative decoding

B300 FP4: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on B300 FP4 (NVIDIA Blackwell) running DeepSeek R1. 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 comparability

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 59 tok/s/user on DeepSeek R1 (B300 FP4), MTP delivers 9,811 tok/s/chip at $0.064 per million tokens; Off delivers 1,349 tok/s/chip at $0.47. MTP delivers 7.3× the tok/s/chip on the same chip, which makes it 86% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 2,781 tok/s/chip for $0.23 per million tokens at 102 tok/s/user on DeepSeek R1 (B300 FP4); Off posts 461 tok/s/chip for $1.36. MTP delivers 6.0× the tok/s/chip on the same chip, which makes it 83% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.

Throughput at 144 tok/s/user on DeepSeek R1 (B300 FP4): MTP hits 676 tok/s/chip, Off hits 166. Costs per million tokens land at $0.93 and $3.79 respectively. MTP delivers 4.1× the tok/s/chip on the same chip, which makes it 76% cheaper per token. 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 · 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.)

Interpolated from real benchmark data. Edit target interactivity values below to compare at different operating points.

Metric
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Interactivity (tok/s/user)
Throughput (tok/s/chip)
MTP9,811Off1,349
MTP2,781Off461
MTP676Off166
Cost ($/M tok)
MTP$0.064Off$0.47
MTP$0.23Off$1.36
MTP$0.93Off$3.79
tok/s/MW
MTP5,163,529Off710,207
MTP1,463,630Off242,461
MTP356,029Off87,213
Concurrency
MTP~1,092Off~137
MTP~272Off~84
MTP~103Off~8
DeepSeek R1: B300 FP4 MTP versus Off speculative decoding comparison at matched interactivity levels
B300 FP4 MTP versus Off speculative decoding comparison for this page's canonical default workload
Open full-size image ↗