Drag the sliders and watch the speedup curve bend toward its ceiling. The dashed red line is
S_max = 1/f — the wall no number of cores can cross. This is the whole law, in one picture.
How much of the program can be parallelized. The rest (serial fraction f) sets the ceiling.
Slides across powers of two: 1, 2, 4, … up to 1024 cores.