mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2016-02-15, 20:45   #1
Fred
 
Fred's Avatar
 
"Ron"
Jan 2016
Fitchburg, MA

97 Posts
Default Testing Types

Ok, this is probably going to sound like a dumb question, but please be gentle, I'm new. Asking it in this sub-forum because I wasn't sure where else to put it.

It seems the holy grail of GIMPS is the discovery of new Mersenne Primes. I've read up on the different types of tests (namely TF and P-1) that occure before LL first time testing. Although these preliminary tests are obviously necessary to avoid unnecessary and costly LL tests, it seems the natural tendency would be for folks to want to only run LL first time tests (with the exception, perhaps, of of an occasional DC to confirm hardware is good).

All of this to ask, what's the motivation of those who perform mainly TF testing (and have clearly spent a LOT of time/resources doing so)? Is it the satisfaction of seeing how many potential Mersenne primes they can? Is it the satisfaction of helping the project as a whole along? Am I an ass for wanting to run LL first time tests exclusively, because (no matter how small the odds) I would love to "discover" the next one?
Fred is offline   Reply With Quote
Old 2016-02-15, 21:26   #2
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

17×251 Posts
Default

The probability of finding a new prime during an LL test is tiny. The monetary prize (for a wavefront test, not 100M digits) is pretty small, too. So really, you're just looking at the excitement and fame of finding a new Mersenne prime. That'd be great, but multiplied by the small probability...it's only a small incentive.

I like the faster run-times, and higher success rates of TF, P-1, and even DC (lower chance of "success" as in "prime", but high chance of "success" as in "I ran a good test, and can immediately prove that"). I have hardware that is well-suited to TF (a GPU) and P-1 (a machine with loads of fast memory), moreso than your average PC. I like contributing to GIMPS, even if someone else gets the fame. A first-time LL test, in contrast, is pretty boring in almost every case: you report a non-prime result after a long processing time, hope that your computer didn't make a mistake, and move on to the next number.

Besides, being one of the people who TF'd or P-1'd a prime before the LL gives its own measure of fame.

I don't look down on people who only want to run LL tests. We are all helping the project forward in our own way.
Mini-Geek is online now   Reply With Quote
Old 2016-02-15, 21:38   #3
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·3·5·11·29 Posts
Default

Quote:
Originally Posted by Fred View Post
All of this to ask, what's the motivation of those who perform mainly TF testing (and have clearly spent a LOT of time/resources doing so)?
Some are more comfortable behind the cameras than in front of them, but still want to make an impact.
chalsall is online now   Reply With Quote
Old 2016-02-15, 23:09   #4
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

22·7·11·31 Posts
Default

I am in complete agreement with the fine gentlemen above.

Knocking out candidates via TF gives some of us a 'special thrill', particularly in the 100M digit range. Doing TF on borged machines is less of an issue.

Much of my effort in my work life involves generating null results (this is good, detections above background are bad). Getting only null results in for GIMPS is boring. TF allows many more 'hits' than LL ever would (except doing dedicated triple checks on mismatches.)
Uncwilly is online now   Reply With Quote
Old 2016-02-15, 23:44   #5
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2×3×1,693 Posts
Default

Performing factoring to optimal levels reduces the amount of needless LL work. Optimal levels are determined by where prefactoring starts requiring more work than the LL. (DCs figure here, too.)

Then there is a competitive subtext: what LaurV calls "credit whoring." I won't deny taking pleasure in my standings in different categories, both on GPU72 and PrimeNet. My setup is very power-inefficient in present-day terms, but I get a substantial output from an FX-8350, and two GTX 580s.

In terms of motivation, I still remember the feeling when I found my first TF factor. That is when one can have a definitive impact on the course of the project, if only at the capillary level of the grand circulation of numbers. Even with considerable processing power, those thrills are pretty sparse for me these days. Given the range and bit-level needed for the project, I'm doing 68M exponents from 74 to75 bits. That is 2h 40m of processing, with a factor or two showing up about monthly from two of them. The important thing is work saved.

When a Mersenne Prime is announced, one of the first things I do is check who did the TF and P-1. It would be gratifying to find that I had had some direct hand in the discovery.
kladner is offline   Reply With Quote
Old 2016-02-15, 23:54   #6
Fred
 
Fred's Avatar
 
"Ron"
Jan 2016
Fitchburg, MA

97 Posts
Default

Thanks all for your replies. I hope my original post didn't imply anything that wasn't intended. I see now how the preliminary tests have their own (and much more realistic odd of having a) "A HA!" moment. I mainly wanted to understand the satisfaction reached from the other forms of testing, and I now have a much better understanding of that.

Last fiddled with by Fred on 2016-02-15 at 23:55
Fred is offline   Reply With Quote
Old 2016-02-16, 06:38   #7
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

25·5·59 Posts
Default

We subscribe to everything other people answered.

Our main motivation was the efficiency of the hardware. Our computers were always CPU-limited, because we do a lot of CAD (mechanical, electronics) in real life, and we always had our computers with GPUs being more efficient that the CPUs. Like hundreds of times more computing power. Also, we were for a while a bitcoin miner (using the GPU too) before the ASICs come into the fray, killing our motivation. We were proud five years ago to be the first "gimpster" having 1.3 Teraflops in a single rig. If you look to this video, at minute 0:45, 2012-03-xx, you can see exactly where we came into it, and the difference we made (yes, that was me only, and you can also zoom out, because that wasn't the only range I was working). That was before mfaktX was able to sieve on the GPU, we still remember running 3 or 4 instances in each GPU, to maximize our two Teslas and two 580 cards (at that time), a hell of manual work! Our "anger" against "that manual work", in managing 15 instances of mfaktc or so, and later Jerry's (FlashJH) and Mike's (Xyzzy) was the "seed" that sprout the GPU sieving (which could maximize the GPU in a single instance, and let the CPU free in the same time), and later Misfit (management of mfaktX instances and work scheduling) - in fact two ways to solve the same problem: getting rid of manual management of hundreds of windows. They ended completing each-other, but that is a different story.

The "credit whore" part came later, when we started teasing each-other about our computing power. It was, and still is, more like a joke, between people who know what a joke is, for example I was teasing Peter ("Sid and Andy" on Gimps and GPU72) and he was teasing me, as we follow and overtake one another in tops for ages, but this is only a joke...

[no, it is not a joke, look to him! Man, how the hack did you get 3700 GHzDays in advance at DC Lifetime? Click image for larger version

Name:	sid.PNG
Views:	72
Size:	2.4 KB
ID:	13899 Now you just destroyed my sleeping nights for the next two weeks! But don't tell to these guys here about my sleepless nights, please...]

... and I don't think that somebody imagined that we ever took the "credit part" seriously. It is fun, however, and sometime it gives you quite a motivation. People here did a lot of nice things together, as a team, and they have a lot of achievements as a community, sometimes motivated only by these little teasing.

Meantime, people got more powerful hardware, and some people enjoy finding factors, or enjoy helping others. Many are here for the community, for talks, learning new things, spending time, etc. I don't do anything more "social" online, than the mersenneforum. I even don't have a facebook account... I am neither the "altruistic" type. But I learned a lot from the people here, and I am proud to be part of such a "select" group. Whatever one can say.

Doing LL is like lottery. You can find a prime and win. But same as lottery, this is a losing business. From the statistical/probabilistic point of view, playing lottery is a losing activity, because only a small part from all the money put in are returned back to the investors. Here is exactly the same, if you want to be rich or famous, I mean, if this is your only motivation, then don't join GIMPS. This is a perdant activity. You will invest more money in electricity and hardware, and you will consume more time than the gain you can make finding a prime, being either money, which are very few, or fame, which last very short - you discover a number which is no use for anybody, people will talk about you for a week, that's it. Cracking RSA numbers will make you more famous, for example.

From this point of view, doing LL or TF/P-1 makes very little a difference. Therefore, we do what we are best doing. Some, as I said, feel fine helping other people finding prime numbers.

[no, don't believe me, I am here for credit, I want to show to these guys that my computers are better than their computers, but don't tell anyone that]

Another motivation is related to the fact that the hardware we use is also used for real life work, or it is "movable", and sometimes having it locked for days in a LL test is not the best use for it. But doing few TF jobs, especially since we have mfaktX and Misfit working so wonderful together, well, that is appealing. DC is also appealing, to keep the hardware in check. But this was said already.

Note that from the "probabilistic" point of view, the best way to find primes (this was said many times, by "heavier" guys here, like RDS, or axn, etc) is to run LL tests. If the goal is to find primes, we should not care about DC-ing at all, the DC job is not profitable, assuming 10% of the LL tests are wrong (a number much higher than the real one!), and it takes 4 times the DC time to do a LL test, we would be about 2.5 times faster in finding a prime if we just ignore the DC. That is because we do 10 DC tests to find one which was bad, which may turn out to be prime, but in the same time we could finish 2.5 or 3 LL tests, having 2.5 or 3 times more chances to find a prime.

However, DC is good because it keeps the HW in check, and because it helps with the other goal of gimps, i.e. making sure we did not miss any primes.

What kind of job you do, or, say it like in matrix, where do you go from here, it is up to you...

Last fiddled with by LaurV on 2016-02-16 at 06:57 Reason: fixing typos and adding "s" to the third person singular verbs... grrr
LaurV is offline   Reply With Quote
Old 2016-02-16, 08:52   #8
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

25·5·59 Posts
Default

Quote:
Originally Posted by LaurV View Post
and later Jerry's (FlashJH) and Mike's (Xyzzy)
We forgot Craig (nucleon) who was actually the largest "producer" on the "eastern" (or should we say southern?) TF front, at the time. Sorry for the other people we forgot.
LaurV is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Anti-poverty drug testing vs "high" tax deduction testing kladner Soap Box 3 2016-10-14 18:43
New work types Unregistered Information & Answers 0 2011-07-25 10:19
v5 work types S00113 PrimeNet 14 2008-12-10 00:26
Primality-testing program with multiple types of moduli (PFGW-related) Unregistered Information & Answers 4 2006-10-04 22:38
Speed of P-1 testing vs. Trial Factoring testing eepiccolo Math 6 2006-03-28 20:53

All times are UTC. The time now is 13:40.

Fri May 7 13:40:19 UTC 2021 up 29 days, 8:21, 0 users, load averages: 2.64, 2.65, 2.53

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.