DeepSeek V4 Pro 0813 1.6T · Speculative decoding

MI355X FP4: MTP vs Off speculative decoding

Speculative decoding comparison of MTP versus Off on MI355X FP4 (AMD CDNA 4) 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.

Near the low end of the 4–75 tok/s/user interactivity band, at 21 tok/s/user on DeepSeek V4 Pro 0813 1.6T (MI355X FP4): MTP runs 5,202 tok/s/chip at $0.080 per million tokens, Off runs 4,487 at $0.093. MTP delivers 16% more tok/s/chip on the same chip, which makes it 14% cheaper per token. Gains from speculative decoding vary by workload; short-output prompts tend to benefit less.

At 39 tok/s/user on DeepSeek V4 Pro 0813 1.6T (MI355X FP4), MTP delivers 3,789 tok/s/chip at $0.11 per million tokens; Off delivers 1,330 tok/s/chip at $0.31. MTP delivers 185% more tok/s/chip on the same chip, which makes it 65% cheaper per token. Speculative decoding accepts draft tokens to reduce per-token latency — gains vary by workload and prompt distribution.

MTP posts 1,738 tok/s/chip for $0.24 per million tokens at 57 tok/s/user on DeepSeek V4 Pro 0813 1.6T (MI355X FP4); Off posts 593 tok/s/chip for $0.70. MTP delivers 193% more tok/s/chip on the same chip, which makes it 66% 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 8k/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)
MTP5,202Off4,487
MTP3,789Off1,330
MTP1,738Off593
Cost ($/M tok)
MTP$0.080Off$0.093
MTP$0.11Off$0.31
MTP$0.24Off$0.70
tok/s/MW
MTP2,488,936Off2,146,870
MTP1,813,116Off636,183
MTP831,629Off283,856
Concurrency
MTP~257Off~916
MTP~291Off~16
MTP~32Off~10
DeepSeek V4 Pro 0813 1.6T: MI355X FP4 MTP versus Off speculative decoding comparison at matched interactivity levels
MI355X FP4 MTP versus Off speculative decoding comparison for this page's canonical default workload
Open full-size image ↗