mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Information & Answers (https://www.mersenneforum.org/forumdisplay.php?f=38)
-   -   What are the most exciting things that can happen in order of likelihood? (https://www.mersenneforum.org/showthread.php?t=25289)

EugenioBruno 2020-02-24 12:32

What are the most exciting things that can happen in order of likelihood?
 
Hi!

Two questions, in bold at the end.

I have a 3700X that runs at ~4.2GHz just fine, and if I'm not encoding or compiling something, that's a lot of spare cycles (and not an unreasonable amount of watts to keep it at that frequency)!

So what the heck, I'll just help out. I know nothing about mathematics, but prime95 doesn't care about that. It can deals with the mathematics by itself :P

I also understand that it's extremely unlikely for any one individual to find a mersenne prime - on the order of winning a lottery.

[B]1) In order of likelihood, what are the things my computer is likely to contribute/discover/prove/disprove?[/B] Right now, I've been assigned what I think I understand is double checking work. Reasonable, has to be done. Hopefully I'll also get assigned more exciting stuff, though?

[B]2) Is it OK if my computer is more or less running 12 hours per day, or will that disqualify me from more interesting work being assigned?[/B]

EugenioBruno 2020-02-24 14:24

Seems like I can't edit the previous message, but two bonus question:

3) I set one worker to do trial factorization. This is so I can have a long running job I check on for days, while still doing some useful work that gets submitted multiple time a day and seeing some progress more quickly.

Is it still roughly as useful to the project, or is it much better to leave both workers to whatever the server thinks is best?

4) If I have two computers running, and I forget to turn one on or otherwise that computer is less "trustworthy", does GIMPS consider it individually, or does it drag the "trustworthiness" of my account as a whole down?

Prime95 2020-02-24 15:18

Exciting is in the eye of the beholder.

Double-checking work can find a new Mersenne prime, chances are very very very low.
First time PRP testing can find a new Mersenne prime, chances are very very low.
DC and PRP can prove a Mersenne number composite. Chances are wonderful. Excitement level not much.

TF can find previously unknown factors. Chances are 1 in every 75 assignments. CPUs are not particularly good at this. TF is good in that it does not compete for memory bandwidth.

P-1 can find previously unknown factors and save someone else from running a first time PRP test.

ECM can find previously unknown factors some quite large. Chances are decent.

PRP-CF can find new Mersenne cofactor probable primes. Chances are low. We have no way to prove the probable prime is really prime.


Do whatever is the most fun for you.


P.S. "trust" has to do with errors occurring during a PRP or LL test, not how many hours a day you run.

P.P.S. In undoc.txt there are ways to automate when prime95 runs -- either by time of day or stopping when other programs are running.

EugenioBruno 2020-02-24 16:23

Possibly my last question: if I just set my preference to "whatever makes most sense", will I get a mix of of those that makes sense to advance the GIMPS project, or is it just saying "yeah, your machine is fast enough, so you're only going to get LL/DC tests" or something along those lines?

kriesel 2020-02-24 17:24

[QUOTE=EugenioBruno;538248]
[B]1) In order of likelihood, what are the things my computer is likely to contribute/discover/prove/disprove?[/B] Right now, I've been assigned what I think I understand is double checking work. Reasonable, has to be done. Hopefully I'll also get assigned more exciting stuff, though?

[B]2) Is it OK if my computer is more or less running 12 hours per day, or will that disqualify me from more interesting work being assigned?[/B][/QUOTE]
1) Most likely, a somewhat higher electrical bill, a good assessment of how reliable your computing equipment is, and determining a given Mersenne number is composite by primality testing, which is somewhat time consuming. Next most likely, finding a factor by P-1 testing. (Or if you run gpu factoring software too, by trial factoring.) Least likely, finding a new Mersenne prime; better odds than a lottery jackpot, but quite small. Like with a lottery ticket, it's the entertainment value, not the hypothetical financial gain.

2) As long as your computer reports in multiple times a week, and completes assigned work before the assignment expires, part time operation is fine.

[QUOTE=EugenioBruno;538256]Seems like I can't edit the previous message, but two bonus question:

3) I set one worker to do trial factorization. This is so I can have a long running job I check on for days, while still doing some useful work that gets submitted multiple time a day and seeing some progress more quickly.

Is it still roughly as useful to the project, or is it much better to leave both workers to whatever the server thinks is best?

4) If I have two computers running, and I forget to turn one on or otherwise that computer is less "trustworthy", does GIMPS consider it individually, or does it drag the "trustworthiness" of my account as a whole down?[/QUOTE]For us ordinary forum users, editing a prior post is only possible for an hour after it's first posted. (It's different for forum moderators.)

3) Please do not use cpu cycles for trial factoring if you can avoid it. Gpus are far better at it. Please do use benchmarking, assuming you are using prime95 or mprime. The optimal number of workers for a given system and work combination depends on cpu type, installed ram, exponent (and consequently fft length) and possibly assignment type (P-1 versus primality testing). See [URL]https://www.mersenneforum.org/showpost.php?p=504218&postcount=4[/URL] and [URL]https://www.mersenneforum.org/showpost.php?p=504219&postcount=5[/URL] for some examples. I think right now the project could use some help keeping up with P-1 factoring, so if you would set one of your workers to do that, and adjust the allowed RAM to some large amount, that would help, and also complete in less time than a double check or first primality test. Occasionally you'll find a factor, which usually eliminates the need for both a first primality test and a double check (an an occasional triple check).

4) Don't worry about a little off time or one system affecting another. One of my computers has been down for nearly a month. It has had no noticeable effect on my account or other systems under my account.

You may find it useful to do some reading at [URL]https://mersenneforum.org/showthread.php?t=24607[/URL] Have a look.

EugenioBruno 2020-02-24 18:22

After your fantastic reply I'm mainly left with a couple of simple questions, but I'm going to address your points one by one first. First of all, thanks a lot for taking the time to answer! :D

[QUOTE=kriesel;538263]1) Most likely, a somewhat higher electrical bill[/QUOTE]
Good point - I've been looking around in my house to find my power meter to see, at the outlet, what the actual difference is (even though I get a reading from my processor I don't quite trust that). My ballpark estimate, however, considering that I have mixing monitors turned on and lots of gadgets plugged in and I do a lot of browsing heavy websites (unbelievable how much taxing that is these days - sounds so backwards), is that it won't be very significant in % terms.

[QUOTE=kriesel;538263]
and determining a given Mersenne number is composite by primality testing, which is somewhat time consuming.
[/QUOTE]
To be honest I don't see this as non-exciting. My computer has (kind of - will need verification) proved something about mathematics. Yeah, I haven't found a huge prime, but I have ruled one out. :)

[QUOTE=kriesel;538263]
2) As long as your computer reports in multiple times a week, and completes assigned work before the assignment expires, part time operation is fine.

4) Don't worry about a little off time or one system affecting another. One of my computers has been down for nearly a month. It has had no noticeable effect on my account or other systems under my account.
[/QUOTE]

Great.



[QUOTE=kriesel;538263]
3) Please do not use cpu cycles for trial factoring if you can avoid it.
[/QUOTE]
Got it. I've fiddled rearranging my worktodo.txt so the trial factoring I've been assigned (by asking for it) should be done fairly quickly and then I can start not wasting energy.

[QUOTE=kriesel;538263]
Please do use benchmarking, assuming you are using prime95 or mprime. The optimal number of workers for a given system and work combination depends on cpu type, installed ram, exponent (and consequently fft length) and possibly assignment type (P-1 versus primality testing).
[/QUOTE]
I've ran a benchmark before reading your post, so I'll do some again with the links you've given me in mind, but it looks like the sweet spot is 1 worker with 8 cores dedicated to it. 2 workers isn't very far behind it, however (on the order of 5%). I think the pattern is that more core I throw at a single thing, the better. Having 1 worker with 6 cores and 1 worker with two (for work that finishes fast, maybe "less exciting" but helpful) might work, but I'll benchmark it manually because I think my CPU is kind of divided into two parts so either doing 4/4 or 8 might be near optimal while other combinations not so much.

[QUOTE]
I think right now the project could use some help keeping up with P-1 factoring, so if you would set one of your workers to do that, and adjust the allowed RAM to some large amount, that would help, and also complete in less time than a double check or first primality test. Occasionally you'll find a factor, which usually eliminates the need for both a first primality test and a double check (an an occasional triple check).
[/QUOTE]

[B]So here's my remaining questions (after you exhaustively answered all my previous ones): [/B]

1) If the project needs some help with P-1 factoring, will leaving my work preference to "Whatever makes the most sense" make the server send me P-1 factoring jobs? Or should I keep up with the project to determine what work is most useful at a given time to assign myself some of that work?

2) I'm confident the other couple of questions will be answered by the reference post you sent me.



Again, thanks for all the help you've already given me :)

Uncwilly 2020-02-24 23:02

[QUOTE=EugenioBruno;538266]1) If the project needs some help with P-1 factoring, will leaving my work preference to "Whatever makes the most sense" make the server send me P-1 factoring jobs? Or should I keep up with the project to determine what work is most useful at a given time to assign myself some of that work?[/QUOTE]If you have a good machine with a bunch of memory that you can allow it use (about 2 G or more of RAM, which you will have to tell Prime95 that it can use), then select P-1 as an assignment type. It will compete with any DC's or first time PRP (or LL) tests for memory bandwidth. if you decide to change the type, you don't have to unassign any work, just let it finish it and get new work of the new type.

EugenioBruno 2020-02-24 23:34

All right, P-1 it is.

I'll be reading about the memory amount searching the forum, doing benchmarks as well. I have 32GB of DDR4 that I rarely utilize (the reason I have that much is to run a relatively lightweight orchestral mockup template in audio production software). In the meantime I just assigned 20GB to it.

I'll also look at doing something about my RAM speed (verifying that it's stable, of course). Right now is running veeeeeery slowly at 2100mhz but I think it's because I have old bios + never even touched it.

Finally, although I doubt I'll ever reach the point where I understand the actual math behind number theory and the search for primes, I've been inspired by this to allocate a small amount of time daily to the study of mathematics. It will not allow me to do any interesting research or get cutting edge papers, but I'm confident it'll help expand my thinking in what I do with the rest of my time which is mostly software engineering.

Thanks everyone! Now that I have a feel for the basics, I can figure out everything from here on by searching the forums.

EugenioBruno 2020-03-07 14:32

Hi,

I've finally settled on using my GPU (about to upgrade to a 1650) for TF and setting my worker to "whatever makes most sense".

I'm curious about what "whatever makes most sense" really means. I've done a DC on an exponent that was tested in late 2013. In a sense, isn't the project "behind" in double checks, or is it better to leave double checks for a long time in the future as they become very cheap compared to the first time PRPs?

I set my worker to do 80% DC as work preference, in part because I like work that completes quicker. I don't remember what the default was, though. What would "make most sense" as that setting, if my goal is just to give spare cycles to the project?

chalsall 2020-03-07 16:03

[QUOTE=EugenioBruno;539083]I'm curious about what "whatever makes most sense" really means. I've done a DC on an exponent that was tested in late 2013. In a sense, isn't the project "behind" in double checks, or is it better to leave double checks for a long time in the future as they become very cheap compared to the first time PRPs?[/QUOTE]

Right now we're focusing on the LL TF wavefronts because that's where we're being chased /hard/ by the LL'ers.

The DC wavefronts have already been appropriately TF'ed for many years to come.

Welcome to the hunt! :tu:

EugenioBruno 2020-03-07 16:44

[QUOTE=chalsall;539090]Right now we're focusing on the LL TF wavefronts because that's where we're being chased /hard/ by the LL'ers.
[/QUOTE]
Is my interpretation of that sentence correct: "the people doing LL are going fast, so we're starting to fall behind in TFs that might avoid those LL checks from being done"?

[QUOTE=chalsall;539090]The DC wavefronts have already been appropriately TF'ed for many years to come.
[/QUOTE]
This is confusing to me. What is a DC wavefront in relation to TFs? If my understanding is correct, TFs are done *before* any LL (or DC) test is done. I don't unrestand, what do TFs have to do at all with DCs?

Also, my question about DC/LL percentage was in reference to my CPU worker.

[QUOTE=chalsall;539090]Welcome to the hunt! :tu:[/QUOTE]
Thanks! As you can see, I can contribute with cycles. Brain, not so much. :P


All times are UTC. The time now is 05:56.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.