Another small question re: P-1

[WillowW]

Hi all,

I'd like to run PrP tests on the four remaining exponents of the form 123456xxx, which are:

123456583, 123456937, 123456967 and 123456979

I'm running P-1 factoring on these exponents right now, using assignments I got from:

https://www.mersenne.org/manual_assignment/

What surprised me is that B2 = B1 = 781000, so that there's no stage 2 testing. Perhaps that's normal, since it's supposed to forestall only 1.3 PrPs. But I thought I'd better ask, since for other exponents of the form 123456xxx, far more knowledgeable users have used, e.g., B2 = 35*B1.

Is this reasonable, or should I re-do the P-1? If so, how do I set B2 by hand?

Thanks for your help! :)

Willow


P.S. Ladies, please consider joining Team NerdyGirls! :D More people = more fun. Girls and women who don't identify as nerdy are also warmly welcome.

[axn]

You have not allocated sufficient RAM for P-1 stage 2. Change the setting from the default 300 MB to as much RAM as you can spare. That will cause it to do proper stage 2.

[WillowW]

Thank you very much, Axn!

I had forgotten the default 0.3GB from the readme.txt. I changed it to 50GB and it worked wonderfully for 123456979. The software gave me roughly B2 = 46*B1 and it took about 5 hours to finish. But now I'll have to go back and do it for the other three exponents, since they finished too quickly.

Until the next small question, ;)
Willow

[petrw1]

Originally Posted by WillowW

Thank you very much, Axn!

I had forgotten the default 0.3GB from the readme.txt. I changed it to 50GB and it worked wonderfully for 123456979. The software gave me roughly B2 = 46*B1 and it took about 5 hours to finish. But now I'll have to go back and do it for the other three exponents, since they finished too quickly.

Until the next small question, ;)
Willow

46xB1 is extremely low. Are you using version 30.8?
Someone may overrule me but I recommend it.

[kriesel]

Originally Posted by petrw1

46xB1 is extremely low.

Hi Wayne, please share the basis for that. Or maybe you meant for prime95 v30.8 & sufficient ram allocation it's low?

A very small random sample:
https://www.mersenne.ca/exponent/109420481 B1 548000 B2 21072000; B2/B1 = 38.45
https://www.mersenne.ca/exponent/94204543 B1 875000 B2 19906250; B2/B1 = 22.75
https://www.mersenne.ca/exponent/74204527 B1 735000 B2 15618750; B2/B1 = 21.25
Gpuowl at one point was set to 1M, 30M by default so B2/B1=30.

The OP is not wrong for letting the software (possibly v30.7b9 prime95 or similar) choose B1 and B2.
V30.8b14 is ok to use but is still not the production release version; v30.7b9 still is for now, per https://www.mersenne.org/download/

[petrw1]

Originally Posted by kriesel

Hi Wayne, please share the basis for that. Or maybe you meant for v30.8 it's small?

A very small random sample:
https://www.mersenne.ca/exponent/109420481 B1 548000 B2 21072000; B2/B1 = 38.45
https://www.mersenne.ca/exponent/94204543 B1 875000 B2 19906250; B2/B1 = 22.75
https://www.mersenne.ca/exponent/74204527 B1 735000 B2 15618750; B2/B1 = 21.25
Gpuowl at one point was set to 1M, 30M by default so B2/B1=30.

The OP is not wrong for letting the software (possibly v30.7b9 prime95 or similar) choose B1 and B2.
V30.8b14 is ok to use but is still not the production release version; v30.7b9 still is for now, per https://www.mersenne.org/download/

That's why I said you can overrule me.
I agree this B2 is fine for 30.7 and yes, 30.8 is still Beta but I would expect must are using it.

[slandrum]

Originally Posted by petrw1

That's why I said you can overrule me.
I agree this B2 is fine for 30.7 and yes, 30.8 is still Beta but I would expect must are using it.

I would not expect brand new people to be using it, and it doesn't get the same ratios when working on wavefront exponents that it does on much smaller exponents.

[petrw1]

ok...ok

I stand corrected

[axn]

Originally Posted by slandrum

it doesn't get the same ratios when working on wavefront exponents that it does on much smaller exponents.

True, but 30.8 still yields much larger ratios compared to 30.7 even at wavefront, especially if you throw something like 50GB at it. Maybe 200x-300x instead of 30x-40x?

[WillowW]

So.....

is it timely to upgrade to 30.8? I had understood that 30.8 was still in β-testing, so I'd been waiting until it was the standard release. But if you think it's safe and advisable to upgrade, then, well, I might be capable of accepting well-informed advice. ;) Right now I'm using 30.7b9, which was standard when I joined a few months ago.

PrimeNet just assigned me a wavefront exponent 109340867 to do prontississime, so my plan is finish that and perhaps one more LL-DC (65524619) with 30.7b9 before switching to 30.8. I'm trying to keep my % of LL-DC's around 20% as per point 14. in https://www.mersenneforum.org/showpo...18&postcount=7. (By the way, thank you, Kriesel, for those wonderful notes — they're super-helpful! :D)

Speaking of well-informed advice for a newbie: My little computer is an AMD 5600G, Radeon Vega 7, with 64 GB DDR4 memory. What tasks should she be doing? How might she be most helpful to the project, for what would she best suited? Would it be sensible to use gpuowl at the same time as mprime? I might be able to figure all this out myself by reading extensively and trying things out, but maybe you all could offer me a kind shortcut? I'm a full-time schoolteacher *and* a university student, so my time is really, really, limited. :P

Thank you, one and all,
Willow

[paulunderwood]

Originally Posted by WillowW

Speaking of well-informed advice for a newbie: My little computer is an AMD 5600G, Radeon Vega 7, with 64 GB DDR4 memory. What tasks should she be doing? How might she be most helpful to the project, for what would she best suited? Would it be sensible to use gpuowl at the same time as mprime? I might be able to figure all this out myself by reading extensively and trying things out, but maybe you all could offer me a kind shortcut? I'm a full-time schoolteacher *and* a university student, so my time is really, really, limited. :P

mprime works on the CPU whereas something like gpuOwl runs on the GPU -- I think your integrated GPU is not well suited to gpuOwl. Someone else might be able to tell you about a program that runs on your GPU to help the cause.

Mind you, running mprime and the GPU flat out will most likely cause your computer to overheat.

Furthermore, the iGPU shares system memory and running a heavy workload on it might actually slow down mprime which requires quite a bit of memory bandwidth.