The more CPU cores a machine has, the more active running threads you can have. If you manage to split a sequential CPU-bound task into a parallel version, you can gain performance by having multiple cores working on the task in parallel.
Going from a single-threaded algorithm to an algorithm that can be run by two threads can, in the best case, double the performance. But, after adding more and more threads, you will eventually reach a limit when there is no more performance gain. Adding more threads beyond that limit will actually degrade performance since the overhead caused by context switching becomes more significant the more threads you add.
For I/O intensive tasks, for example, a web crawler that will spend ...
