any teaching on doing such kind of work?

I got the required program and doing some test.
What I have done is that install Prime95,create a worktodo.ini and paste a line into it. Is that right?
(I am doing 3999831,+1,what a strange start! LOL)

What's that?
Doing 2jobs in same time?

No! Is just a mistake.
I thought you have a method to do two jobs at same time and so do I.

That's not my meaning!
I know running only one work is better.
Is just a "word-misunderstanding",LOL.
EDIT:where is the file and its name if a factor is found?

OH! Not your problem.
Actually, its my problem.(LOL)
I have to say, I am 16-yr-old only and my mother langauge isn't English.

What's the meaning?(from result.txt)


3999831*2^3999831+1 completed P-1, B1=500000, Wc1: 6792D8F5

No! Is just a mistake.
I thought you have a method to do two jobs at same time and so do I.

I don't think you gain anything to run 2 Task.

Lennart

What's the meaning?(from result.txt)


3999831*2^3999831+1 completed P-1, B1=500000, Wc1: 6792D8F5

Are P-1 factoring reservations for the 4M-5M range handled here, too? I only see the 3.75-4M file here.

Oh, and by the way, if you use Pfactor= instead of Pminus1= in the worktodo.ini file entry, you can enter B1 and B2 based on the sieving depth, and how many LLR tests a factor saves, and Prime95 will pick the B1 and B2 for you automatically. (This is how they do it over at Riesel Sieve, for example.)

Here's the format for that:

Pfactor=k,2,n,[+1 or -1],[Sieve depth in terms of 2^x],[How many LLR tests a factor saves]

An example of how that might look would be:

Pfactor=4000029,2,4000029,-1,42,2

If the candidate has been sieved to 2^42, and a factor saves 2 LLR tests.

Are there any advantages to picking the B1 and B2 using either method? Or, are the B1 and B2 values you specified simply the same as Prime95 would have picked for the values such as the ones above (tuned for the 3M-4M range, of course)?
____________

Does anyone know the odds of these numbers yielding a p-1 factor? Also, as the n-value gets larger, a p-1 test prevents more and more unneeded work, so it would make sense if that caused the B1/B2 values to go up. Which leads to my next question:

How were the B1/B2 values calculated?

Next question: Is there an easy way to track the progress of the primetesting? I want to know how much time I have to turn in my numbers, that will affect my reservations AND how often I send those emails.

Does anyone know the odds of these numbers yielding a p-1 factor? Also, as the n-value gets larger, a p-1 test prevents more and more unneeded work, so it would make sense if that caused the B1/B2 values to go up. Which leads to my next question:

How were the B1/B2 values calculated?

What's the meaning?(from result.txt)


3999831*2^3999831+1 completed P-1, B1=500000, Wc1: 6792D8F5

What's the meaning?(from result.txt)


3999831*2^3999831+1 completed P-1, B1=500000, Wc1: 6792D8F5

This means that P-1 factoring was completed on the number, but no factors were found.

Note: If you set Prime95 so that it's allowed to use more memory (preferably 512 MB or more if you've got plenty) then it has a better chance of finding a factor in stage 2, I've heard.

For those of you who have Intel quad-cores, or core2Duos, performance is absolutely horrendous when you're doing Stage 2 p-1 on cores 0 and 1 simultaneously, or cores 2 and 3 simultaneously. The solution, if doing 2 Stage 2s, is pick one core of 0 and 1, and one core of 2 and 3, and set the affinity for those cores. This is not a problem for the beta version of the program, just set the affinity for each core and don't worry about it. For version 24.x(x being any number from 1 to 14), it's a bit more of a hassle. You need to figure out how to set affinity AND how to make half the cores only do Stage 1. Then after the Stage 1s are done, you have to manually switch them to the Prime95 directory of the instance that is doing the Stage 2s, though that instance won't necessarily be doing Stage 2s at that very second. After you switch them over, stop the program, bring up the status window, close the status window, and start the program up again. I'm not sure if stopping the program is necessary, but the status window part is DEFINITELY necessary if you're adding work to Prime95 after it's already started. Bringing up the status window guarantees the program is aware of the extra stuff.

Back to the performance numbers. When I'm talking about performance, the situation is EXTREME. When I run p-1 Stage 2 on both cores 0 and 1, the total time for a candidate is about 30 hours, when I run just one Stage 2, the total time is a little under 5 hours. So a difference of a factor of about 6(a little less). Btw, Stage 2s, when core assignment is done as I've instructed, is fast enough that the Stage 1s don't build up faster than the computer can complete the Stage 2 for them.

From Philmoore at Mersenne Forum(thanks dude :) : Set B2 equal to 1 to skip stage 2. Setting it equal to 0 causes mprime to take B2=100*B1 in most cases (B2 must be < 4.29e9). Setting B2 to any positive value < the B1 limit will cause B2 to be skipped.



(Please don't delete this post, people need the information here. But if you want to edit it, go for it)

I have made a slight error. Apparently, the cpu affinity isn't the issue, it's the RAM bandwidth. If you use 800MHz RAM(never tried the faster stuff, not ever) then you can only use one instance of p-1 without causing a major slowdown. The effect is that, if more than one core is running Stage 2, one core runs normally and the other(s) run dog slow.

At this point I'm not sure if one core can cope with the Stage 1 save files and complete all the necessary Stage 2s without the work being backlogged. Dual core owners need not be concerned.

I have made a slight error. Apparently, the cpu affinity isn't the issue, it's the RAM bandwidth. If you use 800MHz RAM(never tried the faster stuff, not ever) then you can only use one instance of p-1 without causing a major slowdown. The effect is that, if more than one core is running Stage 2, one core runs normally and the other(s) run dog slow.

At this point I'm not sure if one core can cope with the Stage 1 save files and complete all the necessary Stage 2s without the work being backlogged. Dual core owners need not be concerned.

I have made a slight error. Apparently, the cpu affinity isn't the issue, it's the RAM bandwidth. If you use 800MHz RAM(never tried the faster stuff, not ever) then you can only use one instance of p-1 without causing a major slowdown. The effect is that, if more than one core is running Stage 2, one core runs normally and the other(s) run dog slow.

At this point I'm not sure if one core can cope with the Stage 1 save files and complete all the necessary Stage 2s without the work being backlogged. Dual core owners need not be concerned.

You could always set the different instances to be allowed to use a bunch of RAM during different times of the day, and during the rest of the day, only a little, say 128 MB...this will force Prime95 to do Stage 2 only during the time of day where it can use more RAM. Set them all so they each have 1/4 of the time of day you'll be running them, and they'll all simply take turns (Prime95 will hold on to the Stage 1 save files until it can do Stage 2).

Suddenly my prime95 is saying not enough memory to do stage 2 which is bull because there is over 500Mb ram available and settings are for max use of 500Mb ram.

Also I've noticed when running Boinc LLR prime95 runs REALLY slow.. (only ~35% CPU usage)

Any Ideas?
____________

Suddenly my prime95 is saying not enough memory to do stage 2 which is bull because there is over 500Mb ram available and settings are for max use of 500Mb ram.

Also I've noticed when running Boinc LLR prime95 runs REALLY slow.. (only ~35% CPU usage)

Any Ideas?

Suddenly my prime95 is saying not enough memory to do stage 2 which is bull because there is over 500Mb ram available and settings are for max use of 500Mb ram.

Also I've noticed when running Boinc LLR prime95 runs REALLY slow.. (only ~35% CPU usage)

Any Ideas?

maybe over 512mb
my setting is 768mb

Strange message in results.txt...

3751392*2^3751392+1 does not need P-1 factoring.

????????????????????????????????
____________

Using Pfactor= and prime95 version 25.4 I am getting a factor every approx. 45,444s

core2 duo E6400 @2.6Ghz (OC'ed)
1Gb ram

I know that using the B1=500000 and B2=6000000 takes a lot longer than this on my PC.

Also I am doing the ranges using pfactor= and I found a factor that wasn't found in StephanoD's range. I also found the 2 that he did.
Pfactor selects the bounds of B1=90000 and B2=1440000

Anyone know why I found the extra factor? and also what are others seeing for the seconds per factor?

~BoB
____________

As a reminder to all P-1 Sievers The whole goal is to find a factor in less time than it takes to do an LLR test.

From my testing I have found it is better to run Pfactor= than Pminus1= (see P-1 discussion thread)

Please do your own tests to see which one finds factors better.

Since I found it better to use Pfactor= I have created a separate available candidates page for it located here

Also for multi core systems I recommend trying Prime95 version 25.4 available here ( I found better results using all cpu's (cores) than multiple worker threads but again test and see what works best for you.)

~Bob
____________

Has anyone thought of doing a bit of sieving on these numbers with newpgen? I tried a number and it was banging along at about 10G every minute and a half on a core of my 2.4GHz Intel quad-core.

If it's worthwhile, maybe someone could come up with a batch file that runs these numbers in succession up to a certain level. It would be slower than Prime95, but we might be able to kill a few more numbers.

Has anyone thought of doing a bit of sieving on these numbers with newpgen? I tried a number and it was banging along at about 10G every minute and a half on a core of my 2.4GHz Intel quad-core.

If it's worthwhile, maybe someone could come up with a batch file that runs these numbers in succession up to a certain level. It would be slower than Prime95, but we might be able to kill a few more numbers.

I talked to some people in Mersenne forum(a very good place for questions like this) and apparently it IS worthwhile, the question is how high to go.

I'm willing to do a bit of legwork, but are people wanting to add this project to the list, seeing as how it could get rid of some more numbers?

One other thing is that if we're going to add this, then it should be done BEFORE p-1, since it will probably increase the recommended parameters, at least in terms of B1, if not B2.

*bump*

That newpgen idea was a bust, I simply forgot to post that fact here.

Could we get an update on where things stand right now? As in, whether or not it's safe to just start at the top of the file that's supplied? (I mean the 3.75M-4M file)