Hi,
At 07:34 AM 4/13/00 +0200, Martijn Kruithof wrote:
>This happened on my machine while switching from beta 3 to beta 4 is
>this correct?
>
>Starting P-1 factoring on M33237229 with B1=330000, B2=2640000
>Chance of finding a factor is an estimated 4.08%
>P-1 on M33237229 with B1=330000, B2=2640000
>M33237229 stage 2 is 12.08% complete. Time: 1822.421 sec. (789108274292
>clocks)
>
>And after installing beta 4: (Stage 1 was not redone!)
>
>Starting P-1 factoring on M33237229 with B1=305000, B2=1525000
>Chance of finding a factor is an estimated 3.58%
>P-1 on M33237229 with B1=305000, B2=1525000
>M33237229 stage 2 is 20.62% complete. Time: 1767.363 sec. (765268138179
>clocks)
The above looks fine! As was mentioned in whatsnew.txt the program now
does a better job estimating its memory use. More precisely, beta 3 calculated
the amount of FFT data used, beta 4 adds in the sine/cosine and other data
used during the FFT process. For your HUGE exponent that's another 4MB or so.
So what happened is beta 4 decided it couldn't allocate as many temporary
variables for FFT data as beta 3 did. Therefore, stage 2 will be less
efficient
and thus it is not "worthwhile" to run stage 2 as deep as beta 3 wanted to.
Stage 1 was not redone since you had already done more stage 1 than beta 4's
new stage 1 limit. Also, since stage 2 bound decreased your percentage
complete
jumped from 12% to 20%.
One final note. Whatsnew.txt indicates a relatively rare bug in computing
optimal P-1 bounds was fixed. You probably ran into this case. When beta 3
only had enough memory for 4 temporary variables it computed P-1 bounds
assuming it could allocate 5 temporary variables.
>Furthermore a feature would be very nice that it does not recalculate
>the bounds
>without the users consent. (even if it violates the memory bounds)
I'm reluctant to do this. I don't want prime95 to get a reputation of
degrading
performance due to thrashing. Anyway, once you've set your memory parameters
and no longer download a new beta every week, then the computed P-1 bounds
will not change.
>The way to run the memory constraints on a dual machine is by the way
>configuring
>day/nighttime differently for the two cpu's as the stage 2 is only very
>short, it does
>not matter if that's only 4 hours (it can resume the next day anyway).
>(i.e.) one has daytime from 6 pm to 6 am, the other from 6 am to 6 pm.
>(real daytime from 19 to 23 pm might yield : daytime from 19 to 6 and
>one daytime from 6 to 23)
That's one way to do it. You should be aware however that if prime95 does
not have enough memory to run stage 2 and there is no other work to do, then
it will start the LL test. This LL work will be wasted if P-1 later finds
a factor.
Giving stage 2 only 4 hours a day increases the chance that this will happen -
not a major concern, but worth taking into consideration.
Best regards,
George
_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
