GLM 5/5.1 · Speculative decoding

B200 FP8: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on B200 FP8 (NVIDIA Blackwell) running GLM 5/5.1. 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 27 tok/s/user on GLM 5/5.1 (B200 FP8), MTP delivers 1,757 tok/s/chip at $0.27 per million tokens; Off delivers 2,246 tok/s/chip at $0.21. Off delivers 28% more tok/s/chip on the same chip, which makes it 22% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 1,491 tok/s/chip for $0.32 per million tokens at 41 tok/s/user on GLM 5/5.1 (B200 FP8); Off posts 1,174 tok/s/chip for $0.41. MTP delivers 27% more tok/s/chip on the same chip, which makes it 21% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.

Throughput at 55 tok/s/user on GLM 5/5.1 (B200 FP8): MTP hits 1,249 tok/s/chip, Off hits 520. Costs per million tokens land at $0.38 and $0.92 respectively. MTP delivers 140% more tok/s/chip on the same chip, which makes it 58% 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 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)
MTP1,757Off2,246
MTP1,491Off1,174
MTP1,249Off520
Cost ($/M tok)
MTP$0.27Off$0.21
MTP$0.32Off$0.41
MTP$0.38Off$0.92
tok/s/MW
MTP1,027,437Off1,313,616
MTP871,839Off686,779
MTP730,383Off304,274
Concurrency
MTP~60Off~229
MTP~34Off~244
MTP~21Off~15
GLM 5/5.1: 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 ↗