DeepSeek R1 · Speculative decoding

B200 FP8: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on B200 FP8 (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.

MTP posts 5,274 tok/s/chip for $0.091 per million tokens at 44 tok/s/user on DeepSeek R1 (B200 FP8); Off posts 1,885 tok/s/chip for $0.25. MTP delivers 180% more tok/s/chip on the same chip, which makes it 64% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.

Throughput at 75 tok/s/user on DeepSeek R1 (B200 FP8): MTP hits 2,542 tok/s/chip, Off hits 1,042. Costs per million tokens land at $0.19 and $0.46 respectively. MTP delivers 144% more tok/s/chip on the same chip, which makes it 59% cheaper per token. 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 13–137 tok/s/user interactivity band, at 106 tok/s/user on DeepSeek R1 (B200 FP8): MTP runs 1,327 tok/s/chip at $0.36 per million tokens, Off runs 503 at $0.95. MTP delivers 164% more tok/s/chip on the same chip, which makes it 62% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less. (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.)

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)
MTP5,274Off1,885
MTP2,542Off1,042
MTP1,327Off503
Cost ($/M tok)
MTP$0.091Off$0.25
MTP$0.19Off$0.46
MTP$0.36Off$0.95
tok/s/MW
MTP3,084,218Off1,102,269
MTP1,486,530Off609,641
MTP776,046Off294,418
Concurrency
MTP~389Off~22
MTP~160Off~66
MTP~87Off~4
DeepSeek R1: B200 FP8 MTP versus Off speculative decoding comparison at matched interactivity levels
B200 FP8 MTP versus Off speculative decoding comparison for this page's canonical default workload
Open full-size image ↗