Compute · Transform · Data & Research
Compute Amdahl's-Law parallel speedup and efficiency
A pure function that, given a parallelizable fraction and processor count, returns Amdahl's-Law speedup, efficiency, and the asymptotic ceiling, graded on hidden numeric cases.
You receive: A pure function taking {parallelFraction, processorCount, roundingDp} and returning { speedup:number, efficiency:number, maxSpeedup:number }, where parallelFraction p is in [0,1] and processorCount n has an integral value >= 1; invalid inputs return the whole-output error sentinel {"error": "invalid-input"}.
Part of Ship MVP
What's verified: STUD verifies the function reproduces Amdahl's-Law speedup, efficiency, and asymptotic ceiling on hidden input cases to the requested precision, including the invalid-input and infinite-ceiling sentinels. It does NOT verify the input p was measured correctly for a real program or that the workload's serial fraction is actually constant.
Opens soon
This play is verified and ready. It opens soon, once sign-in and payments are live.
Example
A sample of what this play produces. Your result is generated for your inputs.
Inputs
Illustrative scenario: STUD.com's multi-agent integration harness, where roughly 85 percent of wall time is parallelizable per-play authoring and the remaining 15 percent stays serial (assembly plus the gate run), fanned across 8 concurrent agents. Arithmetic recomputed against the tasks.py compute_amdahl_parallel_speedup oracle, 2026-08-08: serial fraction s = 1 - 0.85 = 0.15, raw speedup 1 / (0.15 + 0.85/8) = 1 / 0.25625 = 3.9024390244 rounds to 3.9024; efficiency is taken from the raw speedup, 3.9024390244 / 8 = 0.4878048780, and rounds to 0.4878; the asymptotic ceiling 1 / 0.15 = 6.6666666667 rounds to 6.6667. The 6.6667x ceiling means adding agents past 8 buys little here; shrinking the serial 15 percent moves the ceiling itself.
Get early access to STUD the day it goes live.