GLM 5/5.1 · Speculative decoding

B200 FP4: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on B200 FP4 (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 40 tok/s/user on GLM 5/5.1 (B200 FP4), MTP delivers 1,269 tok/s/chip at $0.38 per million tokens; Off delivers 1,089 tok/s/chip at $0.44. MTP delivers 17% more tok/s/chip on the same chip, which makes it 14% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 934 tok/s/chip for $0.51 per million tokens at 57 tok/s/user on GLM 5/5.1 (B200 FP4); Off posts 577 tok/s/chip for $0.83. MTP delivers 62% more tok/s/chip on the same chip, which makes it 38% 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 GLM 5/5.1 (B200 FP4): MTP hits 687 tok/s/chip, Off hits 281. Costs per million tokens land at $0.70 and $1.71 respectively. MTP delivers 145% 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. (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)
MTP1,269Off1,089
MTP934Off577
MTP687Off281
Cost ($/M tok)
MTP$0.38Off$0.44
MTP$0.51Off$0.83
MTP$0.70Off$1.71
tok/s/MW
MTP742,326Off636,858
MTP546,308Off337,384
MTP401,758Off164,159
Concurrency
MTP~64Off~55
MTP~34Off~21
MTP~19Off~8
GLM 5/5.1: B200 FP4 MTP versus Off speculative decoding comparison at matched interactivity levels
B200 FP4 MTP versus Off speculative decoding comparison for this page's canonical default workload
Open full-size image ↗