GLM 5/5.1 · Speculative decoding

B300 FP4: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on B300 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.

Near the low end of the 27–92 tok/s/user interactivity band, at 43 tok/s/user on GLM 5/5.1 (B300 FP4): MTP runs 2,310 tok/s/chip at $0.27 per million tokens, Off runs 984 at $0.64. MTP delivers 135% more tok/s/chip on the same chip, which makes it 57% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.

At 59 tok/s/user on GLM 5/5.1 (B300 FP4), MTP delivers 1,606 tok/s/chip at $0.39 per million tokens; Off delivers 519 tok/s/chip at $1.21. MTP delivers 3.1× the tok/s/chip on the same chip, which makes it 68% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 1,128 tok/s/chip for $0.56 per million tokens at 76 tok/s/user on GLM 5/5.1 (B300 FP4); Off posts 259 tok/s/chip for $2.43. MTP delivers 4.4× the tok/s/chip on the same chip, which makes it 77% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level. (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)
MTP2,310Off984
MTP1,606Off519
MTP1,128Off259
Cost ($/M tok)
MTP$0.27Off$0.64
MTP$0.39Off$1.21
MTP$0.56Off$2.43
tok/s/MW
MTP1,216,013Off517,845
MTP845,018Off273,018
MTP593,870Off136,228
Concurrency
MTP~129Off~47
MTP~55Off~18
MTP~31Off~7
GLM 5/5.1: 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 ↗