First of all, it usually does not work out to intensify the attempt to warm up the process to normal execution: this will only speed up the first 100 program cycles in your case, receiving a total of less than 20,000 ticks. This is much less than the 75,000 ticks that you have to invest in warming up.
Secondly, all of these benefits from warming up the process / cache / are somewhat fragile. There are a number of events that destroy the warming effect, which you usually do not control. This is mainly due to the fact that your process is not alone in the system. A process switch can behave just like an asynchronous cache flash, and whenever a kernel needs a memory page, it can remove a page from the disk cache.
Since factors make the computation time rather unpredictable, they need to be controlled when performing tests that should give results of any reliability. In addition, these effects are mostly ignored.
source share