Amdahl's Law Speedup

The engineering

Amdahl's Law caps the speedup you can get by throwing more resources at a job when only a fraction p of the work benefits. The serial remainder (1 − p) runs at the same speed no matter how many cores, GPUs, or accelerator lanes you add, so it dominates as N grows. Reach for this before you buy a bigger box: it tells you the answer without the procurement.

The sobering rule of thumb: if 10% of your runtime is stubbornly serial, the most you'll ever get is 10× — infinite cores and all. Efficiency (speedup ÷ N) collapses fast; at 90% parallel and 32 cores you're already under 25% efficiency. Attack the serial fraction first, then scale.

Where this math comes from

Gene Amdahl was IBM's chief architect on the System/360 when he stood up at the 1967 AFIPS Spring Joint Computer Conference to argue against the then-fashionable belief that massively parallel machines were the obvious path forward. His paper wasn't a grand theorem — it was a pointed rebuttal, showing with simple arithmetic that the sequential housekeeping fraction of real programs would strangle parallel gains.

The relation became known as Amdahl's Law almost in spite of him; the paper never wrote it as the tidy formula students now memorize. Decades later John Gustafson pushed back with a scaled-workload counterpoint (1988), noting that bigger machines get used on bigger problems — but Amdahl's ceiling still governs any fixed-size job.

1. 1967Gene AmdahlArgues at AFIPS that the serial fraction limits parallel speedup.
2. 1988John GustafsonProposes scaled-speedup as the complementary case for growing problem sizes.
3. 2008Hill & MartyExtend Amdahl's Law to multicore chip design trade-offs.

See the full timeline of the math behind every calculator →

Runs entirely in your browser — nothing you enter leaves this page. Your recent runs are stored only on your device.