Accuracy of completion date estimates?

[kdq]

I'm not complaining, mind you, but on all my machines, version 25.6 seems to be making very pessimistic estimations. Almost every day, the estimated completion is nearly a full day sooner than previous estimate. Is this usual/common?

[Uncwilly]

Since you are relatively new (a new return), it is not uncommon for a machine to have low ball estimates. How many hours a day are you letting the machine run and how many did you tell it that you were going to run? That will effect it.

[cheesehead]

Originally Posted by kdq

Almost every day, the estimated completion is nearly a full day sooner than previous estimate. Is this usual/common?

I have a similar problem with V24 estimates. It gradually gets better, and at the end of a sequence of tests the elapsed time is not all that far from the sum of original estimates. If I have several tests queued, the future estimates are consistent, but the in-progress test's estimate starts low then increases.

Prime95 does try to correct its estimates by keeping track of the ratio of actual times to estimated times. In file local.ini, there's a pair of lines like:

RollingAverage=457
RollingStartTime=1222988357

RollingAverage = 1000 * (initial estimated time) / (actual time)
= apparent speed of progress where 1000 = nominal

RollingStartTime is when its calculation started.

457 means that actual times are a bit over twice the unadjusted estimated times. In future estimates, Prime95 multiplies its formula-calculated estimates by 1/.457 to produce displayed estimates. Apparently, from my observations, it does not do that correctly for the in-progress test.

I do not know whether V25's estimation procedure differs from V24's.

[S485122]

I have noted this as well for trial factoring : with Prime95 25.6, 1,5 days are added to the first expected expected date, the following intervals are not inflated though.

Jacob

[kdq]

Well, that does explain it: it's adaptive. (24 hours and 24 hours to the first question.)