Amdahl's Law Calculator

Your details

Choose whether to calculate the speedup from a given processor count, or to find the processor count required to reach a target speedup.
The fraction of the total workload that can be parallelized, between 0 (fully serial) and 1 (fully parallel). For example, 0.8 means 80% can be parallelized.
The number of parallel processing units (CPU cores, threads, nodes) available.
cores
The baseline execution time before parallelization. Used to compute improved time; leave at 100 to see results as a percentage reduction.
s
SpeedupModerate scaling
3.333x

Theoretical overall speedup of the system

Parallel efficiency41.7%
Max speedup (infinite cores)5x
Improved execution time30s
Time saved70s
Serial fraction (1 - p)0.2%
41.7%
Poor<35%Moderate35%-60%Good60%-85%Near-linear85%+

Amdahl speedup: 3.333x (theoretical max: 5.00x)

  • The serial fraction (20%) caps your theoretical maximum speedup at 5.00x, no matter how many cores you add.
  • Parallel efficiency is only 41.7%. Most cores are idle waiting on the serial section. Reducing the serial fraction would help far more than adding cores.
  • At 8 cores you have used 67% of the theoretical maximum. Doubling cores gives diminishing gains.

Next stepTo close the gap to the theoretical limit, focus on reducing the serial fraction of the code rather than adding more processors.

 = Powered by OnlyCalculators