Following up on a question I posed earlier:
I ended up with a User CPU time and Total CPU time that was about 4% longer in duration than the elapsed real time. Based on the accepted answer to my earlier question, I don't understand how this could be the case. Could anyone explain this?
Multithreaded code on multiple cores can use more than 100% CPU time.
Because if I use two CPUs at 100% for 10 minutes, I've used 20 minutes worth of CPU time (i.e. were one of those CPUs disabled, it would take 20 minutes for my operation to complete)
One possibility to benchmarks being off by a small margin is due to incorrect timer resolution.
There are quite a few ways of determining those values (time, ticks, CPU frequency, OS API, etc) so not all benchmark routines are 100% reliable.