Lesson 13

Date: 4/19/2017
High Performance Computing (part I: OpenMP)
Linux System Administration

Parallel code speedup

  • In practice, usually only a part of a code can be efficiently parallelized.
  • Amdahl's law imposes a theoretical limit on a parallel code speedup:

  • For example, if only 50% of a code can be parallelized, its theoretical speedup can't exceed the factor of two, according to the Amdahl's law.

  • As you can see on the graph above, it doesn't make much sense utilizing more than 12 CPU cores for running the code since we may be able to reach about 90% of the maximum theoretical speedup on just 12 cores.

  • Take me to the Course Website