mersenneforum.org  

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

Reply
 
Thread Tools
Old 2015-09-15, 22:22   #12
wombatman
I moo ablest echo power!
 
wombatman's Avatar
 
May 2013

23·7·31 Posts
Default

Quote:
Originally Posted by chalsall View Post
I'm a little more paranoid...

Until I get the source code, I don't run it. Or, I put it in a "jail" and sniff the hell out of it.
Never said I was trusting enough to run it ;) Don't have 5 GPUs anyway (oh how I wish), so I wouldn't get to play with the apparent big guns anyway.
wombatman is offline   Reply With Quote
Old 2015-09-15, 22:38   #13
tha
 
tha's Avatar
 
Dec 2002

11000010102 Posts
Default

Quote:
Originally Posted by chalsall View Post
I'm a little more paranoid...
George cannot afford to do anything else than welcome a 'competitor' and just point out the minimum requirements that are needed to be accepted in the field as a trustworthy person or organisation.

I, like you, am very sceptical of this as well. This guy in my opinion has no clue of what kind of community this is, maybe he does not even understand the concept of a community at all.

Some statements in his post are obvious hyping whilst most of the forum members are proficient enough in programming to recognise such. Also the text has some social engineering in it that hardly is above the level of 419 scams.

We are not going to hear much of this in the years ahead.
tha is offline   Reply With Quote
Old 2015-09-15, 23:44   #14
Gordon
 
Gordon's Avatar
 
Nov 2008

7×71 Posts
Default

Quote:
Originally Posted by CEMPLLA Author View Post
Hello All,

I'm brand new to this forum, so please forgive any ignorance I may display.

I'm the author of a newly released software program that took me a little over four years to complete (on & off, in my spare time).

[snip]
Doesn't time fly by, 1st April again already (I can fly, honest!)
Gordon is offline   Reply With Quote
Old 2015-09-16, 00:54   #15
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2·11·353 Posts
Default

Quote:
Originally Posted by Prime95 View Post
5) Promise to share your results with GIMPS. GIMPS will happily give you a ton of factoring data on 100 million digit candidates. GIMPS does not "own" the exponents -- using GIMPS factoring data will not affect your ability to claim the EFF monetary awards.
There has been a vast amount of factoring work to eliminate potential candidates (~1,800 THz days). Currently in the million digit range that includes the first mersennes above 100M digits, well over 60% of the candidates have been eliminated. And looking at the numbers above the first 100M digit number, the percentage is even higher.
Uncwilly is offline   Reply With Quote
Old 2015-09-16, 01:23   #16
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

3·5·7·31 Posts
Default

Quote:
Originally Posted by Prime95 View Post
1) An absolute must: Your program must print out the exponent you are testing and when finished, the final 64-bits of the last LL iteration. This is how end users can validate the accuracy of your program. Without it participation will be very, very low.
2) There must be an interface where the user can select his own exponent to verify the correctness of your program and compare its performance with similar programs (e.g. CudaLucas).
3) Source code is nice, but if you do the above it is not necessary.
4) Share your algorithms and ideas here and learn from the very knowledgeable folks here on how to make it better. From experience I can guarantee that version 1.0 of your program pales in comparison to what it can be using the vast knowledge and experience available here.
5) Promise to share your results with GIMPS. GIMPS will happily give you a ton of factoring data on 100 million digit candidates. GIMPS does not "own" the exponents -- using GIMPS factoring data will not affect your ability to claim the EFF monetary awards.
Ditto all that. Unless the program provides some way of verifying the result, then who knows what it's really doing? I mean, that's really nice that it will do an "assessment" of any *known* Mersenne prime above 2^37-1, but I could do the same thing from my brain and that doesn't prove anything except I've memorized some numbers.

Being able to cross compare something like the 64-bit residue like what other programs do, including Prime95, would go a long way towards gaining some trust.

It's a pretty big ask for someone to come along with no reputation and try to get some hefty systems (5 GPUs?) running an unknown program with no source available...and did I mention no reputation?

Personally I'm leaning towards the conspiracy theories here, that it's probably some stealth bitcoin mining app, or maybe even something more nefarious. Otherwise, why the secrecy? At least show some portions of the code related to the core LL algorithm itself? Unless you've found some clever way to improve on it or make it run faster on the GPU than any other app out there, I don't get the need for secrecy.

Saying the source would be released *after* some record prime is found is silly because if it's not a legit program, it'll never find a prime and you'll never have to release the source.

Yeah, my baloney detector is pegged at 11 right now.
Madpoo is offline   Reply With Quote
Old 2015-09-16, 01:27   #17
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

CB716 Posts
Default

Quote:
Last but not least, CEMPLLA is not affiliated in any way, shape, or form with "GIMPS", which is an organization that is involved in, among other things, achieving the same goals as CEMPLLA. They are using software that is twenty (20) years old, and their software, unlike CEMPLLA, does not employ GPU technology, and is not hardware-extensible - i.e. adding more GPUs will not make their software run any faster. CEMPLLA, on the other hand, does employ GPU technology (albeit, NVidia only), and is hardware-extensible (i.e. the more NVidia GPU devices you add, the faster it will run).
Well there you go.

Sorry mfaktc / mfakto.

And sorry to George too. Funny how it seems like Prime95 v28.7 only came out a month ago, but really it's been 20 years. My how time flies.
Madpoo is offline   Reply With Quote
Old 2015-09-16, 03:01   #18
CEMPLLA Author
 
Sep 2015

716 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Hunting big prime numbers has been known to inspire eager contributors. Until proven otherwise, I tend to believe the OP's intent.

My suggestions to OP based on two decades in the field:

1) An absolute must: Your program must print out the exponent you are testing and when finished, the final 64-bits of the last LL iteration. This is how end users can validate the accuracy of your program. Without it participation will be very, very low.
2) There must be an interface where the user can select his own exponent to verify the correctness of your program and compare its performance with similar programs (e.g. CudaLucas).
3) Source code is nice, but if you do the above it is not necessary.
4) Share your algorithms and ideas here and learn from the very knowledgeable folks here on how to make it better. From experience I can guarantee that version 1.0 of your program pales in comparison to what it can be using the vast knowledge and experience available here.
5) Promise to share your results with GIMPS. GIMPS will happily give you a ton of factoring data on 100 million digit candidates. GIMPS does not "own" the exponents -- using GIMPS factoring data will not affect your ability to claim the EFF monetary awards.

BTW, welcome to the forums! Hopefully you have created a fast program that can make some serious contributions to the field of Mersenne prime research.
Thanks very much for your reply, and your welcome to this forum, which I am sure is sincere. Not only did the former save me from being burned at the stake in this forum for being some kind of warrantless "sleeper" terrorist (which I'm still reeling over), but it was highly edifying. I'd like to address the issues you've raised in your post, but first, I'd like to address some of the replies I've received thus far (other than yours of course):

//-----[ Justifiable Outrage begins here ]-----//

I don't think I've ever had to face so much vitriol from so many for doing something so benign as attempting to distribute my own well-intentioned software. So a couple of questions are in order here:

When did every anti-virus program in the world become completely useless for its intended purpose? If you think I've programmed a virus, run it through a virus-checker, or better yet, send it to Comodo, or Norton, or whoever you want. Sure, it's easier just to accuse an innocent person of being a virus writer, playing on people's fears just to make yourself sound like you're some kind of munificent "protector of the people", but someone is going to figure out that you know less than nothing about the program itself, because you haven't even looked at it, much less downloaded it or examined it in any way, shape or form.

To "tha": Everything I said in my original post is true. Absolutely none of it is "social engineering" "hardly above the level of 419 scams". I mean, really, get a grip. Did you download the program? No. Did you check it with a virus-checker? No. Did you do anything other than skim the text of my original post for words you don't happen to like? No. You just jumped in to accuse me of being some kind of despicable low-life just itching to prey on gullible people for monetary gain. And, oh yeah, in the process, attempting to prop yourself up as someone who knows what the heck they're talking about. Guess what? You don't.

BTW, "sceptical" (sic) is spelled with a "k". Of course, you'd know that if you spent more than five minutes attempting to hang me out to dry for your own self-aggrandizement.

When did being a software author automatically make a person a despicable villain, worthy of being called a liar, "social engineer", "shady as all get-out", etc. without a single person making these accusations even bothering to look at the program in question?
When did it become a crime not to release source code for a Windows program? Is Corel a virus? Is Mathematica a virus? Is Microsoft Office a virus? They must all be viruses, because their makers haven't released their source code either. Oh, but they're corporations - corporations would never do that - only people would do that, right? Because people are inherently bad, and corporations are always good. Yeah, that makes sense..

To "chalsall": "This would be funny if it wasn't so sad.." What's funny? What's sad? The fact that you accused me of being a "sleeper" virus writer (akin to being a "sleeper spy", I'd imagine?) That would be funny, if it wasn't so sad..

To "wombatman": I'm glad you were able to reconsider your opinion and apologize. Enough said.

To "Batalov": Including a "readme.txt" file inside a freely distributed .zip file is standard protocol. You can't get a virus from unzipping a .zip file, or opening a text file in Notepad. Besides which, every virus checking software in the world can scan a .zip file, even before its downloaded. And the entire CEMPLLA System Install program takes about three seconds to download, BTW.

One last thing to all: Just because a person can form grammatically correct sentences, and eloquently elucidate one or more connected trains of thought, does not make him, by default, a "scam artist" - or a virus writer.

//-----[ Justifiable Outrage ends ]-----//

Prime95: Apologies for tainting my reply to your very welcome post with the above, but I don't take being bullied, or being unjustly accused of criminality, lightly - never have, never will..

Anyway, on to the wonderful world of creation and discovery:

I'd like to address your points by number, if that's okay. It might read as a little.. disconnected at times though..

5) Here, you hit the nail on the head when you mention the EFF monetary awards - this is why CEMPLLA exists at all, and I make no apologies for it, nor do I believe that I should. If you want to hear me state it explicitly, then here it is: I want to win the EFF monetary awards for both types of primes. I also believe that CEMPLLA is the best way to do that, because CEMPLLA is hardware-extensible - the more hardware (aka: GPU tech) you add, the faster the discoveries get made. Another way of saying the same thing is to say that CEMPLLA enables people to "throw hardware" at the problem - something I've noticed is quite an enjoyable endeavor for some people - maybe even most people, when it's possible to do so..

I am a little concerned about your phraseology pertaining to GIMPS - i.e. "Promise to share your results with GIMPS". For what it's worth, they are free to download and install the CEMPLLA System themselves, which shares all results with all participants as they occur. However, your use of the word "promise" seems a bit odd to me in this context, especially in light of the fact that it would seem to further the idea that GIMPS has some kind of unspoken "lock" on the EFF monetary awards, and may even be counting on them for future funding. My concern is for appearing to cooperate with GIMPS to the point where they could claim that I was working with them, which I'm not, and have no wish to. They are more than welcome to use whatever CEMPLLA discovers however, as that is public knowledge by definition.

Anyway, I'm probably reading too much into it. I've already incorporated all of the published (on the internet) known factors for both types of primes into the CEMPLLA database, and will continue to do so as they are published. Some are actually larger than CEMPLLA is even capable of finding, because CEMPLLA is limited to 64-bit modulus arithmetic. But they're in there nonetheless, appropriately marked as "too large" (i.e. "too large to display", but there was no room in the text box to fit all that in)..

BTW, I hadn't actually planned on writing this much, so I'll try to be brief from here on in (don't want to write a damn book)..

1) "Your program must print out".. "the final 64-bits of the last LL iteration". I'm not sure I understand this. I do not remember reading that in the EFF rules, but more to the point, the final 64-bits of the final iteration would have to be, by definition, zero, or it wouldn't have passed the LLT, right? I'm probably missing something here, so by all means, enlighten me if at all possible..

2) There is. Exactly as you described. Great minds think alike?

3) By design, the CEMPLLA System does not retain the last iteration of the LLT for a discovered prime (which again, by definition, must be zero), because it would just be a binary file full of zeros. It does retain the last four iterations that were saved either explicitly or implicitly by the running CEMPLLA program, which could be any number of iterations below the last one. Would that suffice for what you're referring to?

4) Uh.. hmm.. Yes, I've already had a taste of the aforementioned "knowledge and experience available here"..

Okay, joking aside, I've had a little over four years to study the various and sundry algorithms that exist for fast and/or huge number multiplication - it actually amazes me that people think that I wouldn't have already done that - and without exception, they're all basically sequential. Writing the CEMPLLA System has been a journey - from the original LLT to fast sequential huge number multiplication techniques, i.e. Karatsuba, Toom-Cook, FFT, Schönhage–Strassen, etc., etc., blah de blah. Pretty much found them all lacking for the purpose, mostly because they are sequential algorithms, but also because the existing software for these techniques are full of comments like, "should this be here?", or "I'm not sure this is necessary.", which introduces a whole new level of ambiguity into the EFF award process that I knew I couldn't afford to abide. These sequential algorithms also fail miserably when it comes to something called "memory locality", which is important if your aim is to exploit the hardware as far as possible. Then I went on to develop my own DLP-based (Data-Level Parallelism) huge number multiplication algorithm, (which will be released to the GP when the source code is released), then by painful trial and error, figured out a way to implement same as a structured distributed computing effort, and with two decades experience as a professional Windows programmer, was able to dress it up nice for what I still hope to be the "general public"..

So okay, your comment, "I can guarantee that version 1.0 of your program pales in comparison to what it can be" might seem to me, at least at this point, having worked hard on this project for four whole years, to be little bit on the uninformed side of things. I really don't mean to offend you by saying that, I'm just being honest. This is not to say that CEMPLLA can't be improved. Of course it can be improved. All software can be improved. But it works now (MPCG has factored another 100 million decimal digit prime candidate since my last post), and I stand by it and all future releases.

As for sharing my algorithms with this forum, or anyone else for that matter, the danger there is that someone (GIMPS even?) will immediately use them to win the EFF award before I can even get CEMPLLA off the ground, based solely on my own hard work - and not theirs. If that sounds selfish, then so be it, It occurs to me however that the EFF awards are meant to be a competition, and it's in that spirit that I continue to work toward the goal of making computational history, and perhaps furthering the art and science of distributed computing. And, oh yeah, making a bit of money doing it as well..

To be fair to myself though, having worked on CEMPLLA for a little over four years now, even if I win both remaining EFF awards, my average pay for all that time would amount to about $10 an hour, if that.. Not exactly great..

BTW, are you the author of the software commonly referred to by your namesake?
CEMPLLA Author is offline   Reply With Quote
Old 2015-09-16, 03:53   #19
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2·11·353 Posts
Default

Quote:
Originally Posted by CEMPLLA Author View Post
5) Here, you hit the nail on the head when you mention the EFF monetary awards - this is why CEMPLLA exists at all, and I make no apologies for it, nor do I believe that I should. If you want to hear me state it explicitly, then here it is: I want to win the EFF monetary awards for both types of primes.
Based upon the search area, and what we know about the distribution of Mersenne Primes, you would be money ahead to mine bit-coins, IMO.

Quote:
My concern is for appearing to cooperate with GIMPS to the point where they could claim that I was working with them, which I'm not, and have no wish to. They are more than welcome to use whatever CEMPLLA discovers however, as that is public knowledge by definition.

Anyway, I'm probably reading too much into it. I've already incorporated all of the published (on the internet) known factors for both types of primes into the CEMPLLA database, and will continue to do so as they are published. Some are actually larger than CEMPLLA is even capable of finding, because CEMPLLA is limited to 64-bit modulus arithmetic. But they're in there nonetheless, appropriately marked as "too large" (i.e. "too large to display", but there was no room in the text box to fit all that in)..
If you have taken into account all the known factors, you have benefited from the work that GIMPS has done. If your program lacks the ability to work above the 64 bit level, it seems that you are missing a lot of valuable search space. You can submit found factors to PrimeNet (without having to have received assignments). This would help pay back the benefit that you have received from GIMPS. (The server will also check the factor to ensure that it is valid.) Also, it will help the competition. If you and GIMPS both don't have to do overlapping factoring, then more factors can be found and more candidates eliminated.
Quote:
1) "Your program must print out".. "the final 64-bits of the last LL iteration". I'm not sure I understand this. I do not remember reading that in the EFF rules, but more to the point, the final 64-bits of the final iteration would have to be, by definition, zero, or it wouldn't have passed the LLT, right? I'm probably missing something here, so by all means, enlighten me if at all possible..
The final 64-bits of the last LL iteration for non-primes, is non-zero. This is how candidates can be double checked by different hardware running different software to verify that a number is not prime. Also, having interim files allow for cross-checking as verification runs are being done. By having a 64 bit residue, users can make sure that the program is doing LL testing and not something else.

Quote:
3) By design, the CEMPLLA System does not retain the last iteration of the LLT for a discovered prime (which again, by definition, must be zero), because it would just be a binary file full of zeros. It does retain the last four iterations that were saved either explicitly or implicitly by the running CEMPLLA program, which could be any number of iterations below the last one. Would that suffice for what you're referring to?
Saving the last check point file if a prime is found allows a quick check (not a deterministic proof, rather a sanity check.) Also, saving a check point file will allow a quick check that somehow the program did not accidentally set the residue to zero earlier.

Quote:
But it works now (MPCG has factored another 100 million decimal digit prime candidate since my last post),
Give us a rough range for that number, the bit size of the factor, and how much time it took the program to run that exponent from the which bit level until it found the factor. I could set a Core2Duo off finding factors at a rather steep rate, if I chose to.

Quote:
It occurs to me however that the EFF awards are meant to be a competition, and it's in that spirit that I continue to work toward the goal of making computational history, and perhaps furthering the art and science of distributed computing. And, oh yeah, making a bit of money doing it as well..
The awards are meant to encourage cooperation.
From their website:
Quote:
EFF hopes to spur the technology of cooperative networking and encourage Internet users worldwide to join together in solving scientific problems involving massive computation. EFF is uniquely situated to sponsor these awards, since part of its mission is to encourage the harmonious integration of Internet innovations into the whole of society.

"The EFF awards are about cooperation," said John Gilmore, EFF co-founder and project leader for the awards.
Quote:
BTW, are you the author of the software commonly referred to by your namesake?
Yes he is.

BTW, based upon experience, parallelization only is good so far. If you want to test the most numbers the fastest, there is a point where testing multiple numbers at the same time generates more through-put. Also, running factoring in parallel to LL can be a good use of resources. Does your program run P-1? If not, why not?

Last fiddled with by Uncwilly on 2015-09-16 at 03:54
Uncwilly is offline   Reply With Quote
Old 2015-09-16, 04:07   #20
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

838410 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Hunting big prime numbers has been known to inspire eager contributors. Until proven otherwise, I tend to believe the OP's intent.

My suggestions to OP based on two decades in the field:

1) An absolute must: Your program must print out the exponent you are testing and when finished, the final 64-bits of the last LL iteration. This is how end users can validate the accuracy of your program. Without it participation will be very, very low.
2) There must be an interface where the user can select his own exponent to verify the correctness of your program and compare its performance with similar programs (e.g. CudaLucas).
3) Source code is nice, but if you do the above it is not necessary.
4) Share your algorithms and ideas here and learn from the very knowledgeable folks here on how to make it better. From experience I can guarantee that version 1.0 of your program pales in comparison to what it can be using the vast knowledge and experience available here.
5) Promise to share your results with GIMPS. GIMPS will happily give you a ton of factoring data on 100 million digit candidates. GIMPS does not "own" the exponents -- using GIMPS factoring data will not affect your ability to claim the EFF monetary awards.

BTW, welcome to the forums! Hopefully you have created a fast program that can make some serious contributions to the field of Mersenne prime research.
+1; +1; +1.
LaurV is offline   Reply With Quote
Old 2015-09-16, 04:10   #21
Madpoo
Serpentine Vermin Jar
 
Madpoo's Avatar
 
Jul 2014

3·5·7·31 Posts
Default

For the purposes of my reply, I'll give you the benefit of the doubt and assume your software is on the level.

Quote:
Originally Posted by CEMPLLA Author View Post
..."sceptical" (sic) is spelled with a "k".
Many of the participants on the forum hail from areas where "sceptical" is as acceptable a spelling as "colour". Of course it looks ridiculous to a yank like myself, or presumably you. LOL

Quote:
Originally Posted by CEMPLLA Author View Post
...5) Here, you hit the nail on the head when you mention the EFF monetary awards - this is why CEMPLLA exists at all, and I make no apologies for it...
It's as good a reason as any. A lot of participants are interested in the same thing. It is important to note though that GIMPS itself and a LOT of the old timers were doing all of this before the EFF or anyone else had offered rewards of any type. We did it for fun.

Quote:
Originally Posted by CEMPLLA Author View Post
I am a little concerned about your phraseology pertaining to GIMPS - i.e. "Promise to share your results with GIMPS".
I think he meant a general sharing of residues of the non-prime results. After all, it would be a bummer if two different projects were going on, each one doing LL tests and then not sharing the results with each other to avoid duplication of effort.

As George said, GIMPS doesn't "own" these numbers... anyone can test them, but in the end it's about scientific discovery and there's no harm in sharing the results of tests that didn't result in a prime being found.

It's also extremely crucial to have those residues (partial is fine of course... GIMPS uses 64 bits of the final iteration) because double checks are vital. Especially on GPUs. This is currently a near and dear subject for me because I've been doing analysis of work that wasn't done right the first (or second, third, etc) time and running double checks on them well ahead of when they would be done in the natural course of things. The theory being that with an error rate of between 2-4%, there is a more than fair chance that a prime got missed the first time through.

You'd probably want to implement a setup like GIMPS does where the residue is masked (last couple hexits hidden) until it's been double checked, just to keep pranksters from submitting phony "matching" residues.

Prime95 also adds a check to the result line to keep casual pranksters from doing more silly things..submitting phony results or what not. Since your source code is unavailable I presume there's things in there that keep people from submitting bogus things?

Quote:
Originally Posted by CEMPLLA Author View Post
...your use of the word "promise" seems a bit odd to me in this context, especially in light of the fact that it would seem to further the idea that GIMPS has some kind of unspoken "lock" on the EFF monetary awards, and may even be counting on them for future funding. My concern is for appearing to cooperate with GIMPS to the point where they could claim that I was working with them, which I'm not, and have no wish to. They are more than welcome to use whatever CEMPLLA discovers however, as that is public knowledge by definition.
The legalese on www.mersenne.org spells out the terms and conditions. In a nutshell, if you use Prime95 (or I think it might apply to anything using the libraries themselves, not just the compiled Prime95 apps) then you split any reward with GIMPS and a charity.

I'm not sure if your "readme.txt" includes the terms and conditions of use and how any monetary prizes are to be shared between yourself and the participant. If not, DO add that, right away... (I am not a lawyer, but that's my advice)

Note that GIMPS is a registered 501(c)(3) non-profit. I don't think any of the board are looking to get rich from this... they've probably spent untold amounts of their own money (and time) to get this all going.

Also note that unlike the EFF award, GIMPS pays out a prize to non-record breaking discoveries. Money that GIMPS has shared in the past with the discoverer has gone into a trust and it pays for the server hosting and kept available for these non-record awards. (I think that's how it works? George will correct me if I'm way off base)

If there was a concern over whether merely sharing residues from a totally independent program would be construed as being a "participant in GIMPS" then I'm sure that could be clarified.

Quote:
Originally Posted by CEMPLLA Author View Post
Anyway, I'm probably reading too much into it. I've already incorporated all of the published (on the internet) known factors for both types of primes into the CEMPLLA database, and will continue to do so as they are published.
And that's the benefit of sharing... GIMPS has put a LOT of combined work into factoring. Does that mean that if you use that information to eliminate candidates that somehow GIMPS gets to share in any rewards your program makes? Nobody (hopefully) would make that claim because it's ridiculous, so use that as an example of how sharing can benefit everyone.

Quote:
Originally Posted by CEMPLLA Author View Post
1) "Your program must print out".. "the final 64-bits of the last LL iteration". I'm not sure I understand this. I do not remember reading that in the EFF rules, but more to the point, the final 64-bits of the final iteration would have to be, by definition, zero, or it wouldn't have passed the LLT, right? I'm probably missing something here, so by all means, enlighten me if at all possible..
See my comments above regarding using residues to double-check. You had planned to double-check, I hope? If not, and if you hadn't even planned to save partial residues for later comparisons for that purpose, I don't mind saying it's a fools errand because the error rate of GPU's is even higher than on a CPU doing LL tests (in my experience). Primes will be missed and you'll have no way of knowing. And without a way to prevent it, some prankster will submit bogus "not prime" results for everything. People are weird that way. They need better hobbies, but there it is.

Also, it would be nice to be able to compare residues from different applications for NON-prime numbers. Anyone can spit out a bunch of zeroes for the final residue of a Mersenne prime, but to get a matching non-prime residue it means everything has to be working. What better way to test how well your program is doing than by comparing the residue to Prime95 or mprime or whatever?

Consider saving those 64-bit residues (or whatever length...64 bit is what current Prime95 uses, but earlier versions used as little as 16 bits which wasn't that useful...you have to consider the odds of getting false matches in the partial residue by not making it large enough).

Quote:
Originally Posted by CEMPLLA Author View Post
...I've had a little over four years to study the various and sundry algorithms that exist for fast and/or huge number multiplication - it actually amazes me that people think that I wouldn't have already done that - and without exception, they're all basically sequential. Writing the CEMPLLA System has been a journey - from the original LLT to fast sequential huge number multiplication techniques, i.e. Karatsuba, Toom-Cook, FFT, Schönhage–Strassen, etc., etc., blah de blah. Pretty much found them all lacking for the purpose, mostly because they are sequential algorithms, but also because the existing software for these techniques are full of comments like, "should this be here?", or "I'm not sure this is necessary.", which introduces a whole new level of ambiguity into the EFF award process that I knew I couldn't afford to abide.
Fair warning, upon any discovery by your software, the EFF will probably want to examine the source to make sure you're not using anything that could infringe upon, oh... for example, the Primenet terms and conditions of use. That is, nothing taken from the gwnum libraries or anything else.

I don't want to sound like I'm belittling your programming abilities because a) I don't know you and b) I haven't seen the program, but a lot of work has been put into getting Prime95, mprime, mfakt* etc all running really well. There are things I think we could agree it could do better as far as parallelism, but I would just caution you to make sure none of your code is taken from the work of others if you intend to claim sole rights to any prize money. Again, I am not a lawyer so add as much salt to that warning as you care.

Quote:
Originally Posted by CEMPLLA Author View Post
So okay, your comment, "I can guarantee that version 1.0 of your program pales in comparison to what it can be" might seem to me, at least at this point, having worked hard on this project for four whole years, to be little bit on the uninformed side of things. I really don't mean to offend you by saying that, I'm just being honest. This is not to say that CEMPLLA can't be improved. Of course it can be improved. All software can be improved. But it works now (MPCG has factored another 100 million decimal digit prime candidate since my last post), and I stand by it and all future releases.
Maybe the algorithm itself is solid. But you may just learn some valuable tips and techniques on the rest of the application and distributed computing side of things from others, things you might not have considered (like the aforementioned pranksters who take perverse delight in breaking things).

Quote:
Originally Posted by CEMPLLA Author View Post
As for sharing my algorithms with this forum, or anyone else for that matter, the danger there is that someone (GIMPS even?) will immediately use them to win the EFF award before I can even get CEMPLLA off the ground, based solely on my own hard work - and not theirs. If that sounds selfish, then so be it, It occurs to me however that the EFF awards are meant to be a competition, and it's in that spirit that I continue to work toward the goal of making computational history, and perhaps furthering the art and science of distributed computing. And, oh yeah, making a bit of money doing it as well..
All fair points. For the record, what is the prize sharing you plan to implement? 50/50 or some other strategy like the 1/3rd that GIMPS does?

Quote:
Originally Posted by CEMPLLA Author View Post
BTW, are you the author of the software commonly referred to by your namesake?
FYI, it's comments like that which make people suspect you're a "sleeper"...and by that I mean some people have created accounts on the forum a long time ago and are just lurkers. At least, I hope nothing untoward was implied. But anyway, if you're not familiar with the "Prime95" user then it makes it sound like you haven't really researched this all as much as you'd like us to think.

But yes, Prime95 = George Woltman. And I, for the record, am the doofus who got busted for running Prime95 on a bunch of computers without permission. LOL
Madpoo is offline   Reply With Quote
Old 2015-09-16, 04:10   #22
wombatman
I moo ablest echo power!
 
wombatman's Avatar
 
May 2013

23×7×31 Posts
Default

Quote:
Originally Posted by Madpoo View Post
Well there you go.

Sorry mfaktc / mfakto.

And sorry to George too. Funny how it seems like Prime95 v28.7 only came out a month ago, but really it's been 20 years. My how time flies.
As Madpoo notes, you'd do well to do a bit more research. There's plenty of GPU-based work going on related to the prime search...
wombatman is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Java applet alternative a1call Programming 19 2019-11-08 22:31
Alternative to LL paulunderwood Miscellaneous Math 36 2019-08-26 08:09
Fulsorials a1call Miscellaneous Math 41 2019-07-21 14:19
Is CEMPLLA 1.5 "the only software in the world capable of discovering" something? Not really. CRGreathouse Number Theory Discussion Group 51 2018-12-16 21:55
free alternative to EasyFit? ixfd64 Software 1 2008-04-26 21:28

All times are UTC. The time now is 00:17.

Thu Apr 2 00:17:09 UTC 2020 up 7 days, 21:50, 3 users, load averages: 2.83, 1.50, 1.30

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