DeepSeek V4 Pro 0813 1.6T · Speculative decoding

H200 FP8: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on H200 FP8 (NVIDIA Hopper) running DeepSeek V4 Pro 0813 1.6T. 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 29 tok/s/user on DeepSeek V4 Pro 0813 1.6T (H200 FP8), MTP delivers 270 tok/s/chip at $1.25 per million tokens; Off delivers 280 tok/s/chip at $1.21. Off delivers 3% more tok/s/chip on the same chip, which makes it 3% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 214 tok/s/chip for $1.58 per million tokens at 50 tok/s/user on DeepSeek V4 Pro 0813 1.6T (H200 FP8); Off posts 130 tok/s/chip for $2.62. MTP delivers 65% more tok/s/chip on the same chip, which makes it 39% cheaper per token. Draft-token acceptance rates determine whether speculative decoding helps or hurts at a given concurrency level.

Throughput at 70 tok/s/user on DeepSeek V4 Pro 0813 1.6T (H200 FP8): MTP hits 133 tok/s/chip, Off hits 68. Costs per million tokens land at $2.54 and $4.96 respectively. MTP delivers 95% more tok/s/chip on the same chip, which makes it 49% 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 · 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)
MTP270Off280
MTP214Off130
MTP133Off68
Cost ($/M tok)
MTP$1.25Off$1.21
MTP$1.58Off$2.62
MTP$2.54Off$4.96
tok/s/MW
MTP197,442Off204,296
MTP156,265Off94,565
MTP97,430Off49,896
Concurrency
MTP~44Off~35
MTP~19Off~11
MTP~8Off~4
DeepSeek V4 Pro 0813 1.6T: H200 FP8 MTP versus Off speculative decoding comparison at matched interactivity levels
H200 FP8 MTP versus Off speculative decoding comparison for this page's canonical default workload
Open full-size image ↗