I just checked the dtime() function of BOINC which is used by the benchmark, quite a surprising function :

It uses the wallclock time to measure the duration instead of measuring the elapsed CPU time.

Shouldn't the benchmark use the clock() function instead, in order to measure the CPU time needed for the benchmark, independant from other programs using up CPU time?

Now it is not so surprising anymore that the benchmark results are so unreliable and not much better than a randomizer on some systems :-/

---

��u�

Sure I do not know for sure, but I suppose it is because of taking the overall computer's activity into account.

Peter

Pepo:

Sure I do not know for sure, but I suppose it is because of taking the overall computer's activity into account.

Then Boinc would have to compute the credits as walltime * benchmark too. It doesn't - so walltime is wrong in the benchmark measurement.

---

Pepo:

Sure I do not know for sure, but I suppose it is because of taking the overall computer's activity into account.

Then Boinc would have to compute the credits as walltime * benchmark too. It doesn't - so walltime is wrong in the benchmark measurement.

OK, then the CPU efficiency in computing over time (% of time running, % of time work allowed, etc.) is taken into account in other calculations, when requesting the work.

(I knew I'm possibly wrong. The running characteristics should off course be sampled for much longer period of time, not only during short benchmark.)

Peter