TF credit granting bug [not that I'm complaining]

[axn]

So this happened:

M69502099 has a factor: 14629490533170662441207 [TF:73:74:mfakto 0.13-Win cl_barrett15_82_gs_2]
M69502099 has a factor: 15564277726929917712817 [TF:73:74:mfakto 0.13-Win cl_barrett15_82_gs_2]
found 2 factors for M69502099 from 2^73 to 2^74 [mfakto 0.13-Win cl_barrett15_82_gs_2]

And then, this happened:

Factors found: 2
Processing result: M69502099 has a factor: 14629490533170662441207 [TF:73:74:mfakto 0.13-Win cl_barrett15_82_gs_2]
CPU credit is 17.3762 GHz-days.
Processing result: M69502099 has a factor: 15564277726929917712817 [TF:73:74:mfakto 0.13-Win cl_barrett15_82_gs_2]
CPU credit is 47.3579 GHz-days.

For comparison, the nominal credit for a no factor result in that range would be 27.5 GHz-days

[LaurV]

Yarrrr!
Very good example found! This is what some people here complained for a while. The given credit is wrong, no matter how you put it.

If the credit is given according with the factor size (if you search them in order), than the first factor is at 54.9% of the way, and the second is at 64.8% of the way. So, the given credit should be:
- for the first factor: 15.09+x GHzDays, where x is a "small bonus for finding a factor"
- for the second factor: 17.81+x GHzDays, where x is the same "small bonus for finding a factor"

If the credit is given by class, and assuming 4620 classes, then
- for the first factor (which is in class 437) a credit of 2.6 GHzDays should be given (plus the small bonus, and yes, this factor is found instantly when 4620 classes are used, so no time is spent for it!)
- for the second factor (which is in class 2292) a credit of 13.64 GHzDays should be given (plus the small bonus for finding a factor)

Assuming 420 classes (the less classes version, but this is never used for 74 bits!)
- first factor is in class 17, so you find it after 17/420 of the time, there is only 1.12 G credit
- second factor is in class 192, with only 12.57 G credit.

Under no circumstance one should get a higher credit than 27.5, except for the "bonus for finding a factor", which can be anything, at GIMPS' discretion, from zero to whatever, to stimulate factoring in some ranges, etc. This is how I see it. The most time you invest when you DON'T find a factor.

[snme2pm1]

Originally Posted by LaurV

If the credit is given by class...

Well, it's a curious thing, that credit is based on the Prime95 way of doing business, yet the notes accompanying at least the v28 distribution state explicitly that "Prime95 should not be used for trial factoring".
I have the notion that both mfakto and mfaktc progress in similar fashion, and the result files plainly state the work agent anyway, so isn't the order of work conducted determinable?

Here's another snippet that I conveyed to George last year in a message captioned expensive results. I didn't investigate deeply, it's similar I guess.

snme2pm1   Manual testing   78010451  F  Nov 29 2013 11:52PM   0.0  46.7813  17719675001845487933567                                                                                                                                                                                 
snme2pm1   Manual testing   78010451  F  Nov 29 2013 11:52PM   0.0  24.5924  9463974117042984423449

I suppose that one day the arrangements will be much less generous on the assumption that GPGPU equipment is commonplace.