New P-1 CPU credit formula

[Prime95]

P-1 was being under-credited. The new formula is:

GHz_days = timing * ( 1.45 * B1 + 0.079 * (B2 - B1) ) / 86400.0

[Prime95]

16753 past P-1 tests have been given a 30% boost in CPU credit to compensate for the under-crediting (only those that did stage 2 were eligible for a boost)

[Kevin]

Would the retroactive 30% boost for P-1 results be noticeable on the individual lines in the "results" page, or was it just added to the P-1 total?

[petrw1]

Originally Posted by Prime95

16753 past P-1 tests have been given a 30% boost in CPU credit to compensate for the under-crediting (only those that did stage 2 were eligible for a boost)

The previous points showed up (in the total only - not detail lines) ... however ... a result that completed only a few hours after this announcement received less than half the credit similar assignments got:

Woody 50756567 NF-PM1 2009-01-27 05:35 49.0 B1=610000, B2=19520000 1.8746 
Woody 50722993 NF-PM1 2009-01-24 23:02 46.8 B1=610000, B2=19520000 4.0079 
Woody 50756449 NF-PM1 2009-01-22 16:20 44.5 B1=610000, B2=19520000 4.0079 
Woody 50756369 NF-PM1 2009-01-20 08:51 42.1 B1=610000, B2=19520000 4.0079

[Prime95]

Originally Posted by Kevin

Would the retroactive 30% boost for P-1 results be noticeable on the individual lines in the "results" page, or was it just added to the P-1 total?

It was the total only. I'm working on the individual result lines now.

[Prime95]

Originally Posted by petrw1

... a result that completed only a few hours after this announcement received less than half the credit

Oops. Formula fixed. I'll nearly triple the credit for P-1 assignments in the buggy 10 hour window.

[petrw1]

Originally Posted by Prime95

Oops. Formula fixed. I'll nearly triple the credit for P-1 assignments in the buggy 10 hour window.

Thanks...I hate to be picky because in my case it is only fractions but if there are others who submitted a lot more in the window it may matter more:

All three listed below are recent and have the same B1/B2 and about the same Exponent. The first one is one in the 10 hour window you manually fixed the other two are before the credit change that were adjusted by your recent conversion.

Woody 50756567 NF-PM1 2009-01-27 05:35 49.0 B1=610000, B2=19520000 5.1553 
Woody 50722993 NF-PM1 2009-01-24 23:02 46.8 B1=610000, B2=19520000 5.2102 
Woody 50756449 NF-PM1 2009-01-22 16:20 44.5 B1=610000, B2=19520000 5.2102

[S485122]

Originally Posted by Prime95

P-1 was being under-credited. The new formula is:

GHz_days = timing * ( 1.45 * B1 + 0.079 * (B2 - B1) ) / 86400.0

I discussed this change with George, probably others did as well. Aside from the fact that P-1 factoring was undervalued my point was that I was under the impression that stage 2 credit was given even when no stage 2 had been done. In order to correct that it seemed logical to have a B2-B1 term. My argumentation convinced George...

What I failed to see is that the new formula is equivalent to :

GHz_days = timing * ( 1.371 * B1 + 0.079 * B2 ) / 86400.0

And that formula again gives the impression that stage 2 credit is awarded even if no stage 2 was done. This means there was no problem with the initial formula even if it seemed to award stage 2 credit for nothing. I welcome the change in coefficients though !

The only remaining problem is subtracting the previously attributed P-1 credit for the same exponent.

GHz_days = timing * ( 1.371 * ( B1 - Old_B1 )+ 0.079 * ( B2 - Old_B2 ) ) / 86400.0

This would be to cope with those like me that redo P-1 on exponents when they find the previously used bounds to low...

Jacob

[petrw1]

Originally Posted by Prime95

It was the total only. I'm working on the individual result lines now.

The NF-PM1 lines have been adjusted for me.
The F-PM1 lines have NOT ... I assume that is yet to be done?

[Prime95]

Originally Posted by petrw1

The NF-PM1 lines have been adjusted for me.
The F-PM1 lines have NOT ... I assume that is yet to be done?

Not enough information is lying around in the database to decide if stage 2 was used to find the P-1 factor. Thus, all old P-1 runs that found a factor were not adjusted.

[garo]

Originally Posted by S485122

I discussed this change with George, probably others did as well. Aside from the fact that P-1 factoring was undervalued my point was that I was under the impression that stage 2 credit was given even when no stage 2 had been done. In order to correct that it seemed logical to have a B2-B1 term. My argumentation convinced George...

What I failed to see is that the new formula is equivalent to :

GHz_days = timing * ( 1.371 * B1 + 0.079 * B2 ) / 86400.0

And that formula again gives the impression that stage 2 credit is awarded even if no stage 2 was done. This means there was no problem with the initial formula even if it seemed to award stage 2 credit for nothing. I welcome the change in coefficients though !

The only remaining problem is subtracting the previously attributed P-1 credit for the same exponent.

GHz_days = timing * ( 1.371 * ( B1 - Old_B1 )+ 0.079 * ( B2 - Old_B2 ) ) / 86400.0

This would be to cope with those like me that redo P-1 on exponents when they find the previously used bounds to low...

Jacob

I think your reasoning is incorrect. If the correct B1 coefficient should be 1.45, then in your equivalent formula, if no stage 2 is done the B2 term simply adds to the B1 term to result in the correct amount of credit.