Parallel Computing: Amdahl's Law

An interesting fact that we learn from this formula is that the number of CPU cores in the device is a limiting factor: creating more and more threads won't make code run faster. To actually reach the maximum speedup, the program must create a number of threads that is equal to the number of existing CPU cores (not more and not less) that will run that section of code.

Example: If a computer has 4 CPU cores, and the portion of the program that is supposed to manifest parallelism is 30%, then the overall speedup of the whole program is:

$$Speedup = \frac{1}{(1-0.3)+\frac{0.3}{4}} = \frac{1}{0.7+0.075} = \frac{1}{0.775} = 1.2903,$$

so the program will run about 1.3 times faster than if it were to run fully on a single core.

These notes by Miriam Briskman are licensed under CC BY-NC 4.0 and based on sources.