P-1 factoring anyone? - Page 5

Bent · 2008-12-15, 14:30

Hmm, that did it. Apparently giving it 128mb is too low now? I never saw a warning that said it didn't have enough.

petrw1 · 2008-12-15, 15:28

Quote:

Originally Posted by Bent

Apparently giving it 128mb is too low now?

I am currently using 20 Times that amount.
The more memory I give it the bigger the B2; the higher the chance it finds a factor; and I believe the longer it takes (a little).

garo · 2008-12-15, 16:40

Give about 500 to P-1. Another possibility is that you have lower RAm settings for the day time so it is waiting for night time to start Stage 2.

Bent · 2008-12-15, 18:18

This is a little bit older machine, 3.2 GHz P4 that has only a gig of ram. If I give it too much ram the performance will suffer. I wanted it to be as invisible as possible. I am aware of the daytime/nighttime settings, and I may have to set it up to only work on stage 2 at night.

garo · 2008-12-15, 22:24

128 should be enough for P-1 so perhaps you have given it even less memory during daytime. Do check that setting and let us know. Otherwise you can always switch to doublechecking. P-1 is best suited for machines with lots of RAM.

James Heinrich · 2008-12-16, 04:16

Quote:

Originally Posted by Bent

3.2 GHz P4 that has only a gig of ram. If I give it too much ram the performance will suffer. I wanted it to be as invisible as possible.

P-1 is one of the most system-impacting worktypes available; it's probably not a good choice on a system of limited resources where you want it to be "invisible" (garo's suggestion of double-check assignments is probably a good one). If you do want to do P-1, I would suggest at least 256MB allocated, preferably more, but as George said too much more than 500MB isn't necessary for now (but will be in the future years as exponents get larger). If you can let Prime95 use lots of memory at night and little during the day, that works well too, as long as you have as-many-or-more night hours as day hours, otherwise you'll get a lot of stage2 work queued up and no time to finish it when memory is available.

Bent · 2008-12-16, 13:46

Quote:

Originally Posted by garo

128 should be enough for P-1 so perhaps you have given it even less memory during daytime. Do check that setting and let us know. Otherwise you can always switch to doublechecking. P-1 is best suited for machines with lots of RAM.

The amount was set the same for the day as it was for the night. I did up the amount of both to 256, I should be able to spare that, and everything seems to work now.

db597 · 2008-12-17, 08:21

Thinking it was enough, I set one of my PCs to use 400MB RAM for both day and night... but being a dual core, this has to be shared by 2 workers. The P-1 still went ahead, and both workers went into stage 2 around the same time. They both completed fine, but I'm not sure if that has had an impact on the results (lower chance of finding factors?).

James Heinrich · 2008-12-17, 13:18

I don't think it impacts the chance of finding a factor (bounds are chosed based on that "up to 400MB available"), but it's slightly slower doing P-1 to given bounds with less memory rather than more.
You can prevent both workers from fighting (and sharing) memory for stage2 by adding this line to local.txt:

Code:

MaxHighMemWorkers=1

Then when stage2 comes up, the first worker that asks will get all 400MB and the other worker will wait until the first one's done before doing stage2 (it will go on to work on stage1 of the next exponent until RAM is available).

petrw1 · 2008-12-17, 19:33

When I reduced my P-1 RAM from 2500 to 1000 the chances of finding a factor reduced from 7.66% to 7.55%.

I don't have enough data yet to determine if this also impacts run time of either Stage.

HOWEVER, I have noticed that with 3 cores doing LL and 1 core doing P-1, Stage 1 has little impact on the other 3 cores but Stage 2 increases the per iteration time on the other cores from .064-0.66 to .066-.068 (about 3%). Stage 1 runs about 20 hours and Stage 2 about 34 hours.

db597 · 2008-12-18, 02:39

Quote:

Originally Posted by James Heinrich

You can prevent both workers from fighting (and sharing) memory for stage2 by adding this line to local.txt:

Code:

MaxHighMemWorkers=1

Then when stage2 comes up, the first worker that asks will get all 400MB and the other worker will wait until the first one's done before doing stage2 (it will go on to work on stage1 of the next exponent until RAM is available).

Thanks. I'll try out MaxHighMemWorkers=1. It sounds very useful especially for quad cores.

1 question - when one core is running stage 2, is prime95 smart enough to switch the other idle core to another exponent's stage 1?

2008-12-17, 08:21	#52
db597 Jan 2003 7·29 Posts	Thinking it was enough, I set one of my PCs to use 400MB RAM for both day and night... but being a dual core, this has to be shared by 2 workers. The P-1 still went ahead, and both workers went into stage 2 around the same time. They both completed fine, but I'm not sure if that has had an impact on the results (lower chance of finding factors?). Last fiddled with by db597 on 2008-12-17 at 08:21

2008-12-17, 13:18	#53
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 11×311 Posts	I don't think it impacts the chance of finding a factor (bounds are chosed based on that "up to 400MB available"), but it's slightly slower doing P-1 to given bounds with less memory rather than more. You can prevent both workers from fighting (and sharing) memory for stage2 by adding this line to local.txt: Code: MaxHighMemWorkers=1 Then when stage2 comes up, the first worker that asks will get all 400MB and the other worker will wait until the first one's done before doing stage2 (it will go on to work on stage1 of the next exponent until RAM is available).

2008-12-15, 14:30	#45
Bent Jul 2007 1110₂ Posts	Hmm, that did it. Apparently giving it 128mb is too low now? I never saw a warning that said it didn't have enough.

2008-12-15, 16:40	#47
garo Aug 2002 Termonfeckin, IE 2²·691 Posts	Give about 500 to P-1. Another possibility is that you have lower RAm settings for the day time so it is waiting for night time to start Stage 2.

2008-12-15, 18:18	#48
Bent Jul 2007 2×7 Posts	This is a little bit older machine, 3.2 GHz P4 that has only a gig of ram. If I give it too much ram the performance will suffer. I wanted it to be as invisible as possible. I am aware of the daytime/nighttime settings, and I may have to set it up to only work on stage 2 at night.

2008-12-15, 22:24	#49
garo Aug 2002 Termonfeckin, IE 2²·691 Posts	128 should be enough for P-1 so perhaps you have given it even less memory during daytime. Do check that setting and let us know. Otherwise you can always switch to doublechecking. P-1 is best suited for machines with lots of RAM.

2008-12-17, 19:33	#54
petrw1 1976 Toyota Corona years forever! "Wayne" Nov 2006 Saskatchewan, Canada 2²·3·17·23 Posts	When I reduced my P-1 RAM from 2500 to 1000 the chances of finding a factor reduced from 7.66% to 7.55%. I don't have enough data yet to determine if this also impacts run time of either Stage. HOWEVER, I have noticed that with 3 cores doing LL and 1 core doing P-1, Stage 1 has little impact on the other 3 cores but Stage 2 increases the per iteration time on the other cores from .064-0.66 to .066-.068 (about 3%). Stage 1 runs about 20 hours and Stage 2 about 34 hours.