|
[QUOTE=Dubslow;291652]The problem is his site doesn't let you enter the memory to use, which plays a not-minor role in bounds choosing. You're allocating ~ [I]100GiB[/I](!!!), which is why Prime95 chose higher bounds and thus a better factor chance. If in doubt, always go with Prime95's bounds. James' site is meant for exploration, i.e. not having to put a Pfactor line in worktodo.txt and then stopping the test to get bounds calculated.
Edit: Whoops, cross post. Addendum: Why the hell are you doing P-1 that high? :P[/QUOTE] Somewhere along the way I was curious about doing high # P-1s. It was discussed [URL="http://www.mersenneforum.org/showthread.php?t=16473"]a while back[/URL] and so I'm working on finishing what I started. I've wondered about Prime95 and James' site for a while with the P-1 calculation, but with such a big exponent, I didn't want to run it without asking. [QUOTE=James Heinrich;291656]Because he has 128GB of RAM :smile: My current 32GB now feels strangely inadequate for my even-crazier attempt at [URL="http://mersenne-aries.sili.net/M595900037"][COLOR=#0066cc]M595,900,037[/COLOR][/URL] (3.5 months and I'm not done stage1 yet, although that's because it's sharing a single core with mfaktc on my i7-920; I'll put 4x 3930K cores on it once it's time for stage2).[/QUOTE] Is 595900037 the highest Prime95 can go right now? I have 4 cores of my 6272 running it. Prime95 says it will complete 25 May 2012. We'll see. If it tries to take all 88M of ram in stage 2, I'll have to limit it so the other workers can continue. |
[QUOTE=flashjh;291657]Is 595900037 the highest Prime95 can go right now?[/QUOTE]No, there's room for you to beat me: the largest FFT in Prime95 (32M) is capped at M596,000,000 so that means [url=http://mersenne-aries.sili.net/M595999993]M595,999,993[/url] is the largest candidate. But, being the largest candidate, it's already [url=http://v5www.mersenne.org/report_exponent/?exp_lo=595999993]had some P-1 attention[/url] from at least 3 users, most recently [i]B1=4479490, B2=107507760 by "Åke Tilander" on 2011-12-03[/i].
|
Actually James, if your bounds are in general a little low, Prime95 calculates bounds using LL tests saved slightly higher than 2 for erroneous tests.
|
[QUOTE=James Heinrich;291659]No, there's room for you to beat me: the largest FFT in Prime95 (32M) is capped at M596,000,000 so that means [URL="http://mersenne-aries.sili.net/M595999993"]M595,999,993[/URL] is the largest candidate. But, being the largest candidate, it's already [URL="http://v5www.mersenne.org/report_exponent/?exp_lo=595999993"]had some P-1 attention[/URL] from at least 3 users, most recently [I]B1=4479490, B2=107507760 by "Åke Tilander" on 2011-12-03[/I].[/QUOTE]
Not looking to beat you :smile: Just exploring. Some day a P-1 like this will be easy! Anyway, Åke Tilander has that assignment out for LL testing as of 29 Feb 2012. I would hate to P-1 it and find a factor while it's assigned ;) Back on subject... kind of. On your site the P-1 probability calculator talks about 'insane memory allocation' (2400 relative primes) When you hover over them, it tells you the number of relative primes required. I have plenty of memory but so far no exponent has gone above 960 relative primes. I'm sure it's been discussed, but what is Prime95s max # of relative primes right now? |
[QUOTE=flashjh;291683]I have plenty of memory but so far no exponent has gone above 960 relative primes. I'm sure it's been discussed, but what is Prime95s max # of relative primes right now?[/QUOTE]Those number come from [url=http://www.mersenneforum.org/showthread.php?t=14562&page=2]this thread[/url], especially starting around post #29. The largest number of RPs I've seen processed is 1440 but that was a rare (unique?) occurrence.
|
[QUOTE=James Heinrich;291698]Those number come from [URL="http://www.mersenneforum.org/showthread.php?t=14562&page=2"]this thread[/URL], especially starting around post #29. The largest number of RPs I've seen processed is 1440 but that was a rare (unique?) occurrence.[/QUOTE]
Cool, I guess we'll see when [URL="http://www.mersenneforum.org/showthread.php?p=291650#post291650"]this baby[/URL] gets to stage 2. |
It's been so long since my main P-1 machine found a factor that I ran a known-factor exponent through it last night just to make sue it was still working - luckily it found the factor. Guess I just have a really dry set of P-1 numbers right now.
|
M595999993
[QUOTE=James Heinrich;291659]No, there's room for you to beat me: the largest FFT in Prime95 (32M) is capped at M596,000,000 so that means [URL="http://mersenne-aries.sili.net/M595999993"][COLOR=#0066cc]M595,999,993[/COLOR][/URL] is the largest candidate. But, being the largest candidate, it's already [URL="http://v5www.mersenne.org/report_exponent/?exp_lo=595999993"][COLOR=#0066cc]had some P-1 attention[/COLOR][/URL] from at least 3 users, most recently [I]B1=4479490, B2=107507760 by "Åke Tilander" on 2011-12-03[/I].[/QUOTE]
[QUOTE=flashjh;291683]Anyway, Åke Tilander has that assignment out for LL testing as of 29 Feb 2012. I would hate to P-1 it and find a factor while it's assigned ;)[/QUOTE] Well, for me its no problem if anyone would like to try to P-1 M595999993 to an even higher level. It would be difficult though to select the boundaries in the most reasonable way in order not to redo the same work again. I think though that the P-1 I already did is more or less to the level which is considered as reasonable. The assignment gave a credit of 403.6681 Ghz-Days. I had to "little" memory (24 GB) for Brent-Suyama to kick in. The largest exponent I have done with B-S is M397626811 (E=6). M595999993 is the largest exponent prime95 can handle at the moment. I did the P-1 in order to test the capabilities of the program. Sometimes weaknesses in programs appears when you use them close to their limits. So anyone can feel free to do more P-1s on M595999993 or to trial factor it beyond ^80 or even try to do a LL (double-check). Please post here if you would like to do more work on M595999993 so that we don't do the same work. If someone wants to do a LL double-check please use in "prime.txt": ResultsFileIterations=20000000 InterimFiles=20000000 Because then we will have interim-files saved every 20000000 iteration and interim We4 residues every 20000000 interation, so we can compare the residues and will have intermediary files to go back to if the residues don't match. I have not yet decided if I will do the LL on M595999993 because it will demand such a lot of resources for such a long time but I presently consider it. If not I will release the exponent. I am worried about the error rate. |
[QUOTE=aketilander;291741]Well, for me its no problem if anyone would like to try to P-1 M595999993 to an even higher level. It would be difficult though to select the boundaries in the most reasonable way in order not to redo the same work again. I think though that the P-1 I already did is more or less to the level which is considered as reasonable. The assignment gave a credit of 403.6681 Ghz-Days. I had to "little" memory (24 GB) for Brent-Suyama to kick in. The largest exponent I have done with B-S is M397626811 (E=6).
M595999993 is the largest exponent prime95 can handle at the moment. I did the P-1 in order to test the capabilities of the program. Sometimes weaknesses in programs appears when you use them close to their limits. So anyone can feel free to do more P-1s on M595999993 or to trial factor it beyond ^80 or even try to do a LL (double-check). Please post here if you would like to do more work on M595999993 so that we don't do the same work. If someone wants to do a LL double-check please use in "prime.txt": ResultsFileIterations=20000000 InterimFiles=20000000 Because then we will have interim-files saved every 20000000 iteration and interim We4 residues every 20000000 interation, so we can compare the residues and will have intermediary files to go back to if the residues don't match. I have not yet decided if I will do the LL on M595999993 because it will demand such a lot of resources for such a long time but I presently consider it. If not I will release the exponent. I am worried about the error rate.[/QUOTE] I'd like to try the P-1, if you don't mind. If you decide to unreserve it from LL, let me know so I can grab it for P-1 from PrimeNet. PrimeNet is down for maintenance... it was TFd to 80, correct? Edit: OK, I need an opinion. If I TF to 82 first, Prime95 gives: [CODE]Optimal P-1 factoring of M595999993 using up to 117950MB of memory. Assuming no factors below 2^82 and 2 primality tests saved if a factor is found. Optimal bounds are B1=4990000, B2=130987500 Chance of finding a factor is an estimated 4.38% Using Core2 type-3 FFT length 32M, Pass1=2K, Pass2=16K, 4 threads [/CODE] If I leave it at 80: [CODE]Optimal P-1 factoring of M595999993 using up to 117950MB of memory. Assuming no factors below 2^80 and 2 primality tests saved if a factor is found. Optimal bounds are B1=5530000, B2=167282500 Chance of finding a factor is an estimated 5.5% Using Core2 type-3 FFT length 32M, Pass1=2K, Pass2=16K, 4 threads [/CODE] At this point, is it better to go with higher TF, lower P-1 probablity or the other way around? |
[QUOTE=flashjh;291748]I'd like to try the P-1, if you don't mind.[/QUOTE]You can, of course, do whatever you want, but I suggest it would be better if you tried P-1 on another exponent, since Åke already did a pretty good P-1 on 595999993. [url=http://mersenne-aries.sili.net/M595999973]M595999973[/url] already has a small factor; the smallest unfactored exponent is [url=http://mersenne-aries.sili.net/M595999939]M595,999,939[/url].
[QUOTE=flashjh;291748]At this point, is it better to go with higher TF, lower P-1 probablity or the other way around?[/QUOTE]Take TF up to at least 82 (about a week on a GPU) and do the smaller P-1. |
[QUOTE=James Heinrich;291754]You can, of course, do whatever you want, but I suggest it would be better if you tried P-1 on another exponent, since Åke already did a pretty good P-1 on 595999993. [URL="http://mersenne-aries.sili.net/M595999973"]M595999973[/URL] already has a small factor; the smallest unfactored exponent is [URL="http://mersenne-aries.sili.net/M595999939"]M595,999,939[/URL].
Take TF up to at least 82 (about a week on a GPU) and do the smaller P-1.[/QUOTE] :smile: |
Ok, looking for some help here.
I have a machine that does nothing but P-1. I had UDIMM memory in it and the memory kept going bad, I was getting restarts, etc. I replaced the UDIMM memory with RDIMM ECC memory on the KGPE-D16 QVL. The system has been very stable, but I noticed it stopped finding S-2 factors. It had been so long since it found S-2 factor that I gave it [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=54699223"]M54699223[/URL] to work on since Dubslow found one: [QUOTE=Dubslow;291362]Brent-Suyama 0_o [URL="http://mersenne-aries.sili.net/exponent.php?factordetails=3546977485247966555997217"][COLOR=#0066cc]http://mersenne-aries.sili.net/exponent.php?factordetails=3546977485247966555997217[/COLOR][/URL] [COLOR=#0066cc]:mellow:[/COLOR] k = 2^4 × 3 × 1367 × 1427 × [U][B]346 268 953[/B][/U] (81.55 bits) B2 = 8,906,250; factor = 346,268,953 factor/B2 = 38.9 For every other expo on this comp, E=12, but P95 doesn't report E when a factor is found.[/QUOTE] So when I ran it, the system didn't find the factor: [CODE] [Mon Mar 12 14:48:12 2012] UID: flashjh/P1Main, M54699223 completed P-1, B1=475000, B2=9025000, E=12, We4: 878AB6EE [/CODE] My B2 was a little higher than Dubslow's, but that shouldn't matter, correct? Anyone have any ideas why the system isn't fining S-2 factors? |
I would try one that doesn't require Brent-Suyama. The extension is kinda funny when it comes to the math, and it doesn't necessarily check every potential k under a given bound, otherwise we wouldn't bother with E and just call it a higher B2. I wouldn't be surprised at all if the different bounds did in fact play a part in what E=12 can find. I would run a S2 on a factor which can in fact be found without E (e.g. [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853"]this one[/URL] or most any other on the GPU272 top factors report), and if that factor is found, then rerun this E=12 test with exactly the same bounds as I had (reported on Mersenne-aries).
|
[QUOTE=Dubslow;292805]I would try one that doesn't require Brent-Suyama. The extension is kinda funny when it comes to the math, and it doesn't necessarily check every potential k under a given bound, otherwise we wouldn't bother with E and just call it a higher B2. I wouldn't be surprised at all if the different bounds did in fact play a part in what E=12 can find. I would run a S2 on a factor which can in fact be found without E (e.g. [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853"]this one[/URL] or most any other on the GPU272 top factors report), and if that factor is found, then rerun this E=12 test with exactly the same bounds as I had (reported on Mersenne-aries).[/QUOTE]
I already started it again with your exact settings. If that doesn't work I'll run one that doesn't require BS that this system did find before. That way I'll know it worked before. I just can't imagine out of all the #s this system has run that it hasn't found another S-2 yet. |
Chance and probability are fickle things, as this three year drought can attest to. :razz:
|
[QUOTE=Dubslow;292805]I would try one that doesn't require Brent-Suyama. The extension is kinda funny when it comes to the math, and it doesn't necessarily check every potential k under a given bound, otherwise we wouldn't bother with E and just call it a higher B2. I wouldn't be surprised at all if the different bounds did in fact play a part in what E=12 can find. I would run a S2 on a factor which can in fact be found without E (e.g. [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853"]this one[/URL] or most any other on the GPU272 top factors report), and if that factor is found, then rerun this E=12 test with exactly the same bounds as I had (reported on Mersenne-aries).[/QUOTE]
BTW - Did you run [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853"][COLOR=#0066cc]this one[/COLOR][/URL]? What was your B1 B2? |
1 Attachment(s)
[QUOTE=flashjh;292812]BTW - Did you run [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853"][COLOR=#0066cc]this one[/COLOR][/URL]?[/QUOTE]
Yes. In this case the bounds shouldn't matter (no BS) but I'll upload the results file, check again. @James: Your site doesn't recognize the E value when a factor is found, see attached. Edit: That factor was credited as TF. [url]http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853[/url] Edit2: While I'm on the topic, when people submit old TF results, instead of putting in a new date (like for P-1 or LL results) it keeps the old "unknown" result, as well as the uploaded result. For example, on the above exponent, you can see Bdot's factoring runs are all duplicated. Edit3: It also occurs to me that it would be nice to see if a P-1 run required the BS extension to find the factor, i.e. having some sort of flag on the page that says "BS extension used (E=*)". This will only be useful for any later factors found, but it would still be nice. flash, the bounds are P-1 found a factor in stage #2, B1=475000, B2=9500000, E=6. UID: Dubslow/Gravemind, M49197853 has a factor: 3907182061566328992465817 |
[QUOTE=Dubslow;292813]Your site doesn't recognize the E value when a factor is found, see attached.[/QUOTE]Yes, I saw you try that.
What version of Prime95 are you using that does that? Or did you just add that manually? |
[QUOTE=James Heinrich;292814]Yes, I saw you try that.
What version of Prime95 are you using that does that? Or did you just add that manually?[/QUOTE] George added the E value in 27.4. (I've also edited the above post.) |
[QUOTE=Dubslow;292813]Yes. In this case the bounds shouldn't matter (no BS) but I'll upload the results file, check again.
@James: Your site doesn't recognize the E value when a factor is found, see attached. Edit: That factor was credited as TF. [url]http://mersenne-aries.sili.net/exponent.php?exponentdetails=49197853[/url] Edit2: While I'm on the topic, when people submit old TF results, instead of putting in a new date (like for P-1 or LL results) it keeps the old "unknown" result, as well as the uploaded result. For example, on the above exponent, you can see Bdot's factoring runs are all duplicated. Edit3: It also occurs to me that it would be nice to see if a P-1 run required the BS extension to find the factor, i.e. having some sort of flag on the page that says "BS extension used (E=*)". This will only be useful for any later factors found, but it would still be nice. flash, the bounds are P-1 found a factor in stage #2, B1=475000, B2=9500000, E=6. UID: Dubslow/Gravemind, M49197853 has a factor: 3907182061566328992465817[/QUOTE] Thanks |
[QUOTE=flashjh;292804]I replaced the UDIMM memory with RDIMM ECC memory on the KGPE-D16 QVL.[/QUOTE]
Have you run a torture test for 24 hours? |
[QUOTE=Dubslow;292813]@James:
Edit: Edit2: Edit3:[/QUOTE]Hmm, seems there's a few things that have become broken I need to look into. :ermm: [QUOTE=Dubslow;292815]George added the E value in 27.4. (I've also edited the above post.)[/QUOTE]Didn't now .4 was out. Now that's added to Prime95, I've also added it to my site. I need to sanitize the existing data per your above-noted comments, but future result submissions should accept the E=__. |
[QUOTE=Prime95;292817]Have you run a torture test for 24 hours?[/QUOTE]
No :blush:, only memtest86+. I was having so many restarts due to bad memory that once the memory passed I didn't think I needed another full torture test... once the current P-1 completes, I'll run one. Edit: BTW - on the UDIMM memory, the system wouldn't even pass a basic blend test. I did run a few custom 32-core blend tests and everything passed on RDIMM. |
[QUOTE=flashjh;292808]I already started it again with your exact settings. If that doesn't work I'll run one that doesn't require BS that this system did find before. That way I'll know it worked before. I just can't imagine out of all the #s this system has run that it hasn't found another S-2 yet.[/QUOTE]
It didn't find BS factor with Dubslow's settings... [QUOTE=Prime95;292817]Have you run a torture test for 24 hours?[/QUOTE] Test is running now... |
Toture test worked fine.
[URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=52038281"]M52038281[/URL] found the factor (again) in S-2. I just finished benchmarking 26.6 and 27.4. I think I've just had a bad streak of exponents. I still wish I knew why it didn't find the same factor as Dubslow in BS? |
[QUOTE=flashjh;293041] I still wish I knew why it didn't find the same factor as Dubslow in BS?[/QUOTE]
I dunno, you'd need the math guys there. Like I said before, the BS extension is much more random in how it works, though admittedly I didn't think it was random enough that running the same bounds wouldn't find the same factor. |
[QUOTE=flashjh;293041]I still wish I knew why it didn't find the same factor as Dubslow in BS?[/QUOTE]
BS also depends on the number of relative primes. |
[QUOTE=Prime95;293044]BS also depends on the number of relative primes.[/QUOTE]
In that case, flash, run it again (same bounds) with your memory set as "up to 10,000 MB". (I can only guess at what my RP count was, but just about everything else has either been 432 or 480 or else somewhere in 400-500.) |
Last run for M54699223
[QUOTE=Prime95;293044]BS also depends on the number of relative primes.[/QUOTE]
On my system, for that exponent it would be 480 unless I set the memory lower, but I'm not sure 10000 will yield less than 480, I'll know tomorrow morning ;) Could having the max actually make the system miss a factor? Here's the line I'm running now (with [Worker #1] set to memory=10000 in local.txt): [CODE]pfactor=N/A,1,2,54699223,-1,74,2[/CODE] It yielded the same info Dublsow turned in to James' site: [CODE][Mar 14 22:54] Worker starting [Mar 14 22:54] Setting affinity to run worker on logical CPU #1 [Mar 14 22:54] Optimal P-1 factoring of M54699223 using up to 10000MB of memory. [Mar 14 22:54] Assuming no factors below 2^74 and 2 primality tests saved if a factor is found. [Mar 14 22:54] Optimal bounds are B1=475000, B2=8906250 [Mar 14 22:54] Chance of finding a factor is an estimated 3.12% [Mar 14 22:54] Setting affinity to run helper thread 1 on logical CPU #2 [Mar 14 22:54] Using AMD K10 type-3 FFT length 2880K, Pass1=640, Pass2=4608, 4 threads [Mar 14 22:54] Setting affinity to run helper thread 2 on logical CPU #3 [Mar 14 22:54] Setting affinity to run helper thread 3 on logical CPU #4[/CODE] [QUOTE=Dubslow;293046]In that case, flash, run it again (same bounds) with your memory set as "up to 10,000 MB". (I can only guess at what my RP count was, but just about everything else has either been 432 or 480 or else somewhere in 400-500.)[/QUOTE] It's running again. If this doesn't work, I'll just have to presume BS is more than just settings until I have a better understanding of the math and the implementation algorithms. Edit: The only reason I'm spending time on this, is I don't want to be missing factors on my system, and, if something isn't working correctly, it's a waste of time for me to run P-1s on this system. Thanks for the help. |
It's also an interesting perspective on the BS extension, since it does appear that regular Stage 2 works. Even if BS doesn't work, P-1 should still be fine. (I can't say for sure, but I don't believe the factor chance includes the BS possibility.)
|
[QUOTE=Dubslow;293056]It's also an interesting perspective on the BS extension, since it does appear that regular Stage 2 works. Even if BS doesn't work, P-1 should still be fine. (I can't say for sure, but I don't believe the factor chance includes the BS possibility.)[/QUOTE]
Last time I ran this I used [CODE] Pminus1=1,2,54699223,-1,475000,8906250 [/CODE] So, this time the system should be doing exactly what yours did. I don't know if you're using Intel or AMD? Maybe the CPU and FFT Type has something to do with it also? (Obviosuly I have no clue how the algorithm works). |
No, CPU type and FFT have nothing to do with it, but like I said, regular Stage 2 works fine, so your CPU is fine for P-1. (It's an i7-2600K, if you're still curious.)
|
Update: Here's what's happening:
[CODE][Mar 15 05:31] M54699223 stage 1 complete. 1370506 transforms. Time: 23814.032 sec. [Mar 15 05:31] Starting stage 1 GCD - please be patient. [Mar 15 05:33] Stage 1 GCD complete. Time: 168.350 sec. [Mar 15 05:33] Available memory is 10000MB. [Mar 15 05:33] Using 9123MB of memory. Processing 384 relative primes (0 of 384 already processed). [/CODE] I'll post the result later... |
[QUOTE=flashjh;293091]Update: Here's what's happening:
[CODE][Mar 15 05:31] M54699223 stage 1 complete. 1370506 transforms. Time: 23814.032 sec. [Mar 15 05:31] Starting stage 1 GCD - please be patient. [Mar 15 05:33] Stage 1 GCD complete. Time: 168.350 sec. [Mar 15 05:33] Available memory is 10000MB. [Mar 15 05:33] Using 9123MB of memory. Processing 384 relative primes (0 of 384 already processed). [/CODE] I'll post the result later...[/QUOTE] Ok, it found the BS factor: [CODE] [Thu Mar 15 13:20:17 2012] P-1 found a factor in stage #2, B1=475000, B2=8906250, E=12. UID: flashjh/P1Main, M54699223 has a factor: 3546977485247966555997217 [/CODE] Now I have to ask the P-1 experts why? Setting the max memory to 10000 causes relative primes to drop from 480 to 384. With 480 it didn't find the BS factor but with 384 it did. Is this system missing a lot more factors because of this? Thanks for the guidance... |
I think it's like this: There are a certain number of k's for which one factor is between B1 and B2, while the other factors are <B1. Imagine a list:
K1 K2 K3 K4 K5 K6 etc... Lets assume that this is a list of Ks with a factor >B2, in order. Now, bear in mind that this is a very very rough analogy, but I [i]believe[/i] it's something like with 480 RPs, you are able to check every second K on that list with E=12. But with RPs <480, you can check if every third K is a factor. You can see then that with different RPs you check different K values, and it happens that this particular factor can be found with those given bounds and the right number of RPs. This is also why I said that BS is "random" with respect to what factors with K > B2* it can possibly find. Your computer is working just fine. *Of course here I mean (one factor of K)>B2, but shorthand FTW! Edit: Let me re-emphasize that this is a very stretched analogy based on a rough idea of the math. Anyone who's studied it is probably able to be more precise/accurate. |
[QUOTE=Dubslow;293145]I think it's like this: There are a certain number of k's for which one factor is between B1 and B2, while the other factors are <B1. Imagine a list:
K1 K2 K3 K4 K5 K6 etc... Lets assume that this is a list of Ks with a factor >B2, in order. Now, bear in mind that this is a very very rough analogy, but I [I]believe[/I] it's something like with 480 RPs, you are able to check every second K on that list with E=12. But with RPs <480, you can check if every third K is a factor. You can see then that with different RPs you check different K values, and it happens that this particular factor can be found with those given bounds and the right number of RPs. This is also why I said that BS is "random" with respect to what factors with K > B2* it can possibly find. Your computer is working just fine. *Of course here I mean (one factor of K)>B2, but shorthand FTW! Edit: Let me re-emphasize that this is a very stretched analogy based on a rough idea of the math. Anyone who's studied it is probably able to be more precise/accurate.[/QUOTE] That makes sense. I'm running it one more time with no memory setting to see what happens (I wouldn't have done this, but I accidentally left the pminus1 line in and it was already at 63% S-1, so I changed it to [CODE]pfactor=N/A,1,2,54699223,-1,74,2[/CODE] and I'll let it finish. After that, I'll just let the system run away with normal settings (unless someone has a better suggestion :smile:) |
[QUOTE=flashjh;292804]Ok, looking for some help here.
I have a machine that does nothing but P-1. I had UDIMM memory in it and the memory kept going bad, I was getting restarts, etc. I replaced the UDIMM memory with RDIMM ECC memory on the KGPE-D16 QVL. The system has been very stable, but I noticed it stopped finding S-2 factors. It had been so long since it found S-2 factor that I gave it [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=54699223"]M54699223[/URL] to work on since Dubslow found one: So when I ran it, the system didn't find the factor: [CODE] [Mon Mar 12 14:48:12 2012] UID: flashjh/P1Main, M54699223 completed P-1, B1=475000, B2=9025000, E=12, We4: 878AB6EE [/CODE] My B2 was a little higher than Dubslow's, but that shouldn't matter, correct? Anyone have any ideas why the system isn't fining S-2 factors?[/QUOTE] Same results as before: [CODE] [Fri Mar 16 03:48:54 2012] UID: flashjh/P1Main, M54699223 completed P-1, B1=475000, B2=9025000, E=12, We4: 878AB6EE [/CODE] Relative primes (and max memory) is the difference. Dubslow, you seem to find a lot of BS factors, I wonder if less memory is actually better for P-1? |
Just two, I think. I'm quite sure it's just statistical noise.
I do have my own problem though. Since switching to 27.4, all my P-1s have gone from E=12 to E=6 (except for like 2 or 3 out of 15+ total). I didn't change anything, and they all use up to 10,000 MB of RAM like I told flash. Any ideas? Because I have found those 2 BS factors, this does have me a bit more worried than it would otherwise. (George already said he did not modify the bounds algorithm.) |
[QUOTE=Dubslow;293215] (George already said he did not modify the bounds algorithm.)[/QUOTE]
AVX FFTs consume a little more memory than SSE2 FFTs. |
27.3 and 27.2 had E=12.
|
[QUOTE=Dubslow;293224]27.3 and 27.2 had E=12.[/QUOTE]
Curious. 27.4 uses a little less memory than 27.3. |
I went forever without a S-2 factor and then it finds a S-1 and a S-2 in the same morning :smile:
[URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=52606901"]M52606901[/URL] [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=52591241"]M52591241[/URL] [QUOTE=Dubslow;293215]Just two, I think. I'm quite sure it's just statistical noise. I do have my own problem though. Since switching to 27.4, all my P-1s have gone from E=12 to E=6 (except for like 2 or 3 out of 15+ total). I didn't change anything, and they all use up to 10,000 MB of RAM like I told flash. Any ideas? Because I have found those 2 BS factors, this does have me a bit more worried than it would otherwise. (George already said he did not modify the bounds algorithm.)[/QUOTE] Did you figure out the issue? |
[QUOTE=flashjh;293584]
Did you figure out the issue?[/QUOTE] Nope, it's randomly gone back to E=12, and for my most recent result (as of this post, the file is a symlink), it didn't print E at all. [url]http://dubslow.tk/gimps/results.txt[/url] |
It would be preferable if you guys wouldn't shorten Brent-Suyama to BS.
(B.Sc. to BS is OK, that's pretty much fair in my book.) |
[QUOTE=Batalov;293589]It would be preferable if you guys wouldn't shorten Brent-Suyama to BS.
(B.Sc. to BS is OK, that's pretty much fair in my book.)[/QUOTE] No problem :smile: [BREAK] Unbelievable... two S-2 in as many days:[CODE][Mar 20 05:31] P-1 found a factor in stage #2, B1=550000, B2=12650000, E=12. [Mar 20 05:31] [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=52605569"]M52605569[/URL] has a factor: 1499018992232168845493327[/CODE] |
[QUOTE=Dubslow;293588]Nope, it's randomly gone back to E=12, and for my most recent result (as of this post, the file is a symlink), it didn't print E at all.
[url]http://dubslow.tk/gimps/results.txt[/url][/QUOTE] Huh, it had one random E=6 overnight. Nice going flash! Now I just need to find some... |
[QUOTE=Dubslow;293588]Nope, it's randomly gone back to E=12, and for my most recent result (as of this post, the file is a symlink), it didn't print E at all.
[url]http://dubslow.tk/gimps/results.txt[/url][/QUOTE] [QUOTE=Dubslow;293625]Huh, it had one random E=6 overnight.[/QUOTE] I take it back, it appears to be correlated with exponent. I'm not sure what the actual boundary is, but 49M has so far been a good separator. Below that, I get E=6, above that, I get E=12, as well as a 45M that reported no E at all. Weird. I'll go through my [URL="http://dubslow.tk/gimps/old-results.txt"]old results[/URL] and see if there are any other correlations; preliminarily, I found M45520697 with E=12 (edit: and M45063679, M46104649), however all the rest under 49M appear to have E=6; I also found M52479659 without any E reported (edit: M52144357 has no E, nor do M52220219, M52567117). It seems that this 49M rule mostly applies, but not always. Thoughts? (Request: Prime95 report version when recording results? As far as these files are concerned, I typically upgraded within a day of a new version coming out, excepting 27.1.) |
If my P-1 code include an extra 2 and 3 in the E found using a given bound, will that throw off the result?
|
No, it's called the Brent-Suyama extension to the P-1 algorithm.
The more memory you have available, the higher E you get -- Prime95 is able to select (I believe) E=2,3,4,6,8,12. Most of us doing serious amounts of P-1 get E=6 or 12. (3 GiB gives me E=6, and 10 GiB gives me E=12, for standard LL-size exponents). What it does is allows certain possible factor candidates above B2 to be tested, but not all of them -- that would just be a higher B2. [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=325517"]This[/URL], [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=7175153"]this[/URL] and [URL="http://mersenne-aries.sili.net/exponent.php?factordetails=3546977485247966555997217"]this[/URL] are examples of a factor found with the Brent-Suyama extension -- note how the largest factor of k is in all cases (significantly) larger than B2. |
P-1 factoring with cuda, anybody? :P
This is my first tentative, it can not handle numbers higher then 64 bits in an efficient manner, and there are lots and lots of things I don't know, about both cuda and the math... but I am learning... Here is what you can get with a B1 up to 10k, and a B2 about 100 times larger, and if no factor is found after that, I still do a couple of squares (what is shown as stage 2b) so, first is the exponent (below 400), second is the factor found. Note that the factors may be composite (for such a small exponents the bounds are too large). [CODE] 2 :: No factor. Try increasing the bound. 3 :: No factor. Try increasing the bound. 5 :: No factor. Try increasing the bound. 7 :: No factor. Try increasing the bound. 11 :: 23 is a factor in stage 2a. 13 :: No factor. Try increasing the bound. 17 :: No factor. Try increasing the bound. 19 :: No factor. Try increasing the bound. 23 :: 47 is a factor in stage 1. 29 :: 1103 is a factor in stage 1. 31 :: No factor. Try increasing the bound. 37 :: 223 is a factor in stage 2a. 41 :: 13367 is a factor in stage 1. 43 :: 2099863 is a factor in stage 1. 47 :: 2351 is a factor in stage 1. 53 :: 69431 is a factor in stage 1. 59 :: 179951 is a factor in stage 2a. 61 :: No factor. Try increasing the bound. 67 :: 761838257287 is a factor in stage 2a. 71 :: 212885833 is a factor in stage 1. 73 :: 439 is a factor in stage 1. 79 :: 2687 is a factor in stage 1. 83 :: 167 is a factor in stage 2a. 89 :: No factor. Try increasing the bound. 97 :: 11447 is a factor in stage 2a. 101 :: No factor. Try increasing the bound. 103 :: 2550183799 is a factor in stage 1. 107 :: No factor. Try increasing the bound. 109 :: 745988807 is a factor in stage 2a. 113 :: 78939089 is a factor in stage 1. 127 :: No factor. Try increasing the bound. 131 :: 263 is a factor in stage 1. 137 :: No factor. Try increasing the bound. 139 :: 5625767248687 is a factor in stage 2a. 149 :: No factor. Try increasing the bound. 151 :: 165799 is a factor in stage 1. 157 :: 852133201 is a factor in stage 2b. 163 :: 150287 is a factor in stage 1. 167 :: 2349023 is a factor in stage 2a. 173 :: 1505447 is a factor in stage 1. 179 :: 359 is a factor in stage 2a. 181 :: 43441 is a factor in stage 2a. 191 :: 383 is a factor in stage 1. 193 :: 13821503 is a factor in stage 1. 197 :: 7487 is a factor in stage 1. 199 :: No factor. Try increasing the bound. 211 :: 15193 is a factor in stage 2b. 223 :: 196687 is a factor in stage 1. 227 :: No factor. Try increasing the bound. 229 :: 20492753 is a factor in stage 1. 233 :: 189714193 is a factor in stage 2a. 239 :: 84487457 is a factor in stage 1. 241 :: 22000409 is a factor in stage 2a. 251 :: 503 is a factor in stage 2a. 257 :: No factor. Try increasing the bound. 263 :: 23671 is a factor in stage 2a. 269 :: 13822297 is a factor in stage 2a. 271 :: 15242475217 is a factor in stage 2b. 277 :: 1121297 is a factor in stage 2b. 281 :: 80929 is a factor in stage 2b. 283 :: 9623 is a factor in stage 1. 293 :: No factor. Try increasing the bound. 307 :: 14608903 is a factor in stage 1. 311 :: 5344847 is a factor in stage 2a. 313 :: 10960009 is a factor in stage 2a. 317 :: 9511 is a factor in stage 1. 331 :: No factor. Try increasing the bound. 337 :: 18199 is a factor in stage 1. 347 :: No factor. Try increasing the bound. 349 :: No factor. Try increasing the bound. 353 :: 931921 is a factor in stage 2b. 359 :: 719 is a factor in stage 2a. 367 :: 12479 is a factor in stage 2a. 373 :: 25569151 is a factor in stage 1. 379 :: 180818808679 is a factor in stage 1. 383 :: 1440847 is a factor in stage 2a. 389 :: 269160341965838569 is a factor in stage 2a. 397 :: 202471 is a factor in stage 1. [/CODE] |
*jump on LaurV*
Give it to me! |
:max:
|
So when does the OpenCL version come out! :wink:
|
[QUOTE=Batalov;293589]It would be preferable if you guys wouldn't shorten Brent-Suyama to BS.
(B.Sc. to BS is OK, that's pretty much fair in my book.)[/QUOTE] BS here (and quite probably elsewhere) has only one meaning: :poop: |
1 Attachment(s)
[QUOTE=davieddy;294037]BS here (and quite probably elsewhere) has only one meaning:
:poop:[/QUOTE] [ATTACH]7824[/ATTACH] |
I just had an interesting P-1 find, [URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8598691[/URL], which I am guessing is a B-S find, since the bounds used were 100,000 and 1,825,000, and it's listed as needing a minimum 12,796,033 B2.
|
[QUOTE=bcp19;295448]I just had an interesting P-1 find, [URL]http://mersenne-aries.sili.net/M8598691[/URL], which I am guessing is a B-S find, since the bounds used were 100,000 and 1,825,000, and it's listed as needing a minimum 12,796,033 B2.[/QUOTE]Certainly B-S.
If you submit your results to the site it would be graphed more clearly (the graph currently only shows the previous P-1 bounds). |
[QUOTE=James Heinrich;295453]Certainly B-S.
If you submit your results to the site it would be graphed more clearly (the graph currently only shows the previous P-1 bounds).[/QUOTE] I just submitted the whole results file, since it had never been sent to your site before. |
[QUOTE=Prime95;152314]I'll be watching over the next few months how many people sign up for P-1 factoring. We have tons of people doing LL and TF, we need more people signing up for P-1.
In the new scheme of things, if P-1 doesn't get done then when an exponent is assigned to an LL tester he must do the P-1, last 2 levels of TF, and then the LL test. If not enough people sign up for P-1, I'll have to tweak "do what makes the most sense" to assign some P-1 factoring. We have plenty of time to work this out as the leading edge of exponents that need P-1 factoring and are holding up TF is above 48 million. It will take a while before the leading edge of LL gets that high.[/QUOTE] I have some comment, but I don't intend to get bard again. |
Found another B-S factor. [URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8521171[/URL] (results file not submitted yet)
edit: Also have 13 bad prior P-1's on 8.5M so far. |
[QUOTE=bcp19;298377]Found another B-S factor. [URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8521171[/URL] (results file not submitted yet)[/QUOTE]
What B1/B2 did you use? |
[QUOTE=flashjh;298379]What B1/B2 did you use?[/QUOTE]According to [URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8521171[/URL], B1=45000/B2=686250 (Look on the "Actual" line at mid-page and under the "P-1 results:" close below that.)
|
[QUOTE=flashjh;298379]What B1/B2 did you use?[/QUOTE]
100000/1825000 E=6 [QUOTE=cheesehead;298384]According to [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=8521171"][COLOR=#000080]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8521171[/COLOR][/URL], B1=45000/B2=686250 (Look on the "Actual" line at mid-page and under the "P-1 results:" close under that.)[/QUOTE] That was the original P-1, since a factor was found, it does not show the new bounds. |
[QUOTE=bcp19;298385]100000/1825000 E=6
That was the original P-1, since a factor was found, it does not show the new bounds.[/QUOTE] Okay. I just figured that you'd submitted the result between post #1224 and now. Also, under "date found" it says "2 May 2012". Is [URL="http://mersenne-aries.sili.net/48292035869045144656328161"]48292035869045144656328161[/URL] the factor you found? I'm puzzled about how the numbers on the page are consistent. |
[QUOTE=cheesehead;298387]Okay. I just figured that you'd submitted the result between post #1224 and now.
Also, under "date found" it says "2 May 2012". Is [URL="http://mersenne-aries.sili.net/48292035869045144656328161"]48292035869045144656328161[/URL] the factor you found? I'm puzzled about how the numbers on the page are consistent.[/QUOTE] Yes. James' site picked up from Primenet that I submitted that factor and credited it on the 2nd, the day it was submitted. Since I have not yet submitted the results file to his site though, it does not show the updated P-1 that I performed. If you look at [URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=8598691[/URL] for example, you will see that I have submitted that results file. I usually like to have a sizeable file when I upload results, plus procrastination hapens. |
[QUOTE=bcp19;298460]I usually like to have a sizeable file when I upload results, plus procrastination hapens.[/QUOTE]Which is fine, you can upload as big or small results file as you like to my site. However, if you're going to comment on P-1 bounds and refer the data on my site, it would be helpful to the rest of us to see what you're talking about if you submit at least that single result to mersenne-aries.sili.net first so we can see the pretty graphs. :smile:
|
[QUOTE=bcp19;298377]Found another B-S factor[/QUOTE]For those who are fascinated by P-1 factors found thanks to the Brent-Suyama extension, I now have a list of known ones here:
[url]http://mersenne-aries.sili.net/brent-suyama.php[/url] Note: there are certainly far more than the 84 currently listed, but they'll appear slowly over the next few days as I crawl my database and fill in the k factorizations (and thereby the minimum required B1/B2 to compare with the actual B1/B2 used). |
[QUOTE=James Heinrich;298639]For those who are fascinated by P-1 factors found thanks to the Brent-Suyama extension, I now have a list of known ones here:
[url]http://mersenne-aries.sili.net/brent-suyama.php[/url] Note: there are certainly far more than the 84 currently listed, but they'll appear slowly over the next few days as I crawl my database and fill in the k factorizations (and thereby the minimum required B1/B2 to compare with the actual B1/B2 used).[/QUOTE] Nice, I have two in the list and I didn't even know. |
[QUOTE=James Heinrich;298639]For those who are fascinated by P-1 factors found thanks to the Brent-Suyama extension, I now have a list of known ones here:
[URL]http://mersenne-aries.sili.net/brent-suyama.php[/URL] Note: there are certainly far more than the 84 currently listed, but they'll appear slowly over the next few days as I crawl my database and fill in the k factorizations (and thereby the minimum required B1/B2 to compare with the actual B1/B2 used).[/QUOTE] That is a nice job! OTOH, I would doubt that some 64-66 bits factors (with astronomical k's!!) in that list were found by P-1 at all. They were reported as big composites by some workers, tricking PrimeNet to believe they are P-1 (see Axon thread here around iirc), but they were (some new, some previously known) clearly TF factors. OTOTOH :razz:, if you are still here around, please have a look to my comment in the kepler thread related to some gtx690 being a dual chip card. |
[QUOTE=LaurV;298647]That is a nice job!
OTOH, I would doubt that some 64-66 bits factors (with astronomical k's!!) in that list were found by P-1 at all. They were reported as big composites by some workers, tricking PrimeNet to believe they are P-1 (see Axon thread here around iirc), but they were (some new, some previously known) clearly TF factors. OTOTOH :razz:, if you are still here around, please have a look to my comment in the kepler thread related to some gtx690 being a dual chip card.[/QUOTE] Factors can only appear on the list if a results file/line was uploaded to his website saying it was P-1 and giving the specific bounds used. Otherwise he couldn't do the min/actual calculation he does (and otherwise he wouldn't know for sure it was P-1, as you pointed out). |
[QUOTE=James Heinrich;298639]For those who are fascinated by P-1 factors found thanks to the Brent-Suyama extension, I now have a list of known ones here:
[URL]http://mersenne-aries.sili.net/brent-suyama.php[/URL] Note: there are certainly far more than the 84 currently listed, but they'll appear slowly over the next few days as I crawl my database and fill in the k factorizations (and thereby the minimum required B1/B2 to compare with the actual B1/B2 used).[/QUOTE][URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=7854383[/URL] is an interesting case. Suggestion for extra work for you to do when bored: On [URL]http://mersenne-aries.sili.net/brent-suyama.php[/URL] change the "Minimum Required" heading and columns to four columns: minimum required B1/B2 with Brent-Suyama, minimum required B1/B2 without Brent-Suyama. Cases like 7854383 show why minimum required B1 could be different between with and without Brent-Suyama, unless I'm mistaken. Please enlighten me if I'm mistaken. Even snazzier would be to show the extra columns only when there's a B1 difference, as with 7854383. But you may not be that bored for a long time, so ... |
[QUOTE=Dubslow;298650]Factors can only appear on the list if a results file/line was uploaded to his website saying it was P-1 and giving the specific bounds used. Otherwise he couldn't do the min/actual calculation he does (and otherwise he wouldn't know for sure it was P-1, as you pointed out).[/QUOTE]
I didn't know that, sorry. I misinterpreted his affirmation about "crawling" as "crawling the PrimeNet database". There, factors found with P-1 appear as "F-PM1", and the others found by TF appears simple as "F". But reporting a big composite factor as P-1, results in marking both smaller factors as P-1. You can in fact TF one exponent, and if you are lucky to find 2 factors, then you report their product and get a big jump in P-1 credit. |
[QUOTE=LaurV;298653]I didn't know that, sorry. I misinterpreted his affirmation about "crawling" as "crawling the PrimeNet database". There, factors found with P-1 appear as "F-PM1", and the others found by TF appears simple as "F". But reporting a big composite factor as P-1, results in marking both smaller factors as P-1. You can in fact TF one exponent, and if you are lucky to find 2 factors, then you report their product and get a big jump in P-1 credit.[/QUOTE]
True, but PrimeNet doesn't record the method used in the public tables (like [URL="http://mersenne.org/report_factors/"]Factor Report[/URL]) (much less the actual P-1 bounds); the type determination is only for credit purposes. However, since very few users actually upload their results to Mersenne-aries, it might be worth it to James to crawl through the factors and look for those with factors more than a few bits above standard TF, and perhaps put those in a separate table marked "possible B-S factors". [QUOTE=cheesehead;298652][URL]http://mersenne-aries.sili.net/exponent.php?exponentdetails=7854383[/URL] is an interesting case. Suggestion for extra work for you to do when bored: On [URL]http://mersenne-aries.sili.net/brent-suyama.php[/URL] change the "Minimum Required" heading and columns to four columns: minimum required B1/B2 with Brent-Suyama, minimum required B1/B2 without Brent-Suyama. Cases like 7854383 show why minimum required B1 could be different between with and without Brent-Suyama, unless I'm mistaken. Please enlighten me if I'm mistaken. Even snazzier would be to show the extra columns only when there's a B1 difference, as with 7854383. But you may not be that bored for a long time, so ...[/QUOTE] The problem is that Brent-Suyama depends heavily on what bounds are used and the "relative prime" count that Prime95 prints. It's more or less a random crap shoot above B2, as far as which particular FCs are tested, depending on B2 (and even then on the RP count like I just mentioned). See, for example [URL="http://mersenne-aries.sili.net/exponent.php?exponentdetails=54699223"]M54699223[/URL] -- flash had seen a long dearth of S2 factors, and so [URL="http://www.mersenneforum.org/showthread.php?p=292804#post292804"]tested his machine[/URL] against this factor -- with a higher B2, he actually missed the B-S factor. [I](Edit: That entire page of that second link, and the first post of the following page is about B-S and P-1. It's worth reading the page in its entirety -- among other discussions, there's a slightly more extended analogy.)[/I] The point is there isn't a "minimum required with B-S" because it's a crap shoot. (If it weren't a crap shoot, that'd be the same as just testing to the higher B2.) |
[QUOTE=Dubslow;298655]True, but PrimeNet doesn't record the method used in the public tables (like [URL="http://mersenne.org/report_factors/"]Factor Report[/URL])[/QUOTE]
That's why the "crawling" part: crawling through the "exponent status" pages and extracting exponents having "F-PM1" keyword :D. Of course, as I said, I completely forgot the fact that James keeps his own DB with history (contrary to PrimeNet where the history is gone when a factor is found). |
[QUOTE=LaurV;298659]I completely forgot the fact that James keeps his own DB with history (contrary to PrimeNet where the history is gone when a factor is found).[/QUOTE]
I actually have a question about this. Is the history completely erased from the database or is it that it is not any longer displayed? You can for instance find LL-results for a specific exponent in the database even if a factor is found, but if you use [URL="http://www.mersenne.org/report_exponent/"]report_exponent[/URL] you only get information about the factor. That is: How much of old results are still kept in the database? Is something at all erased? Is it possible to retrieve all information in database for a specific exponent if you would like to even in cases when a factor has been found later? |
[QUOTE=aketilander;298664]
That is: How much of old results are still kept in the database? Is something at all erased? Is it possible to retrieve all information in database for a specific exponent if you would like to even in cases when a factor has been found later?[/QUOTE] I suspect not, though of course I have no way to prove it. Database size is a major concern for PrimeNet -- and any factor found means space saved, with the current (assumed but not proven) policies in place. |
The history is not erased, just hidden and impossible to access anymore for commoners (like us). The proof is the fact that when you access[URL="http://www.mersenne.org/report_LL/?exp_lo=1&exp_hi=1000000000&exp_date=&user_only=1&user_id=YOURUSERID&txt=1&dispdate=1&B1=Get+LL+data"] your own LL[/URL]** history, beside of "verified", "unverified" and "bad" results, there will be also a "Test results where factor was found later" section (I have a couple of them in my list). This means I did a LL, it was verified (or not), but later a factor was found. The DB remembers exactly the exponent's history (that is, who was the guy who did the initial LL, double check LL, maybe what type of factoring was used to find the factor, etc).
**in the link you have to replace yourUserId, with the real one, and must be logged on edit: which is case sensitive :D found it in the hard way |
[QUOTE=Dubslow;298655]However, since very few users actually upload their results to Mersenne-aries, it might be worth it to James to crawl through the factors and look for those with factors more than a few bits above standard TF, and perhaps put those in a separate table marked "possible B-S factors".[/QUOTE]One day, possibly, eventually, I'll crawl through my list of known factors that have no known method of discovery and try and guess which was used. But that's not something I'm planning on doing anytime soon.
|
[QUOTE=LaurV;298647]OTOH, I would doubt that some 64-66 bits factors (with astronomical k's!!) in that list were found by P-1 at all.[/QUOTE]Do you have a specific example I could look at?
As has been observed elsewhere, PrimeNet isn't very good at (a) knowing which manual results are TF vs P-1 (often assigning P-1 credit to TF results), and (b) reporting what bounds were used when a TF factor is found (which, of course, would also eliminate problem (a)). [QUOTE=Dubslow;298650]Factors can only appear on the list if a results file/line was uploaded to his website saying it was P-1 and giving the specific bounds used. Otherwise he couldn't do the min/actual calculation he does (and otherwise he wouldn't know for sure it was P-1, as you pointed out).[/QUOTE]Quite right. I have many spidered-from-PrimeNet "PM1" factors in my database, but since PrimeNet doesn't tell what bounds were used when a PM1 factor was found that's not much use in this case. I can calculate what bounds would be required to find via normal P-1, but without knowing what bounds were used there's no way to know if Brent-Suyama was invoked. [QUOTE=cheesehead;298652][URL]http://mersenne-aries.sili.net/M7854383[/URL] is an interesting case.[/quote]It is. It's the "very rare" case [i]axn[/i] talked about when he [url=http://www.mersenneforum.org/showpost.php?p=288180&postcount=1114]outlined the algorithm for finding B1/B2[/url]:[quote]If largest has a power > 1 (very rare), set B1=B2=large^power. Stop. (only stage 1 can find this one. stage 2 not needed)[/quote] [QUOTE=cheesehead;298652]change the "Minimum Required" heading and columns to four columns: minimum required B1/B2 with Brent-Suyama, minimum required B1/B2 without Brent-Suyama.[/quote]The "Minimum Requied" B1/B2 is for regular P-1 with no Brent-Suyama extension, and is easily calculated. The minimum required bounds for Brent-Suyama to find the factor are nebulous at best, and don't actually translate into minimum bounds per se. For example: [url=http://mersenne-aries.sili.net/exponent.php?exponentdetails=54699223]M54,699,223[/url] was found by [i]Dubslow[/i] with B1=475,000/B2=8,906,250. But [i]flashjh[/i] was unable to replicate the discovery with the slightly-higher bounds of B1=475,000/B2=9,025,000. It seems this is because a different number of relative primes was selected, and that affects how the Brent-Suyama extension works. So describing a "minimum B1/B2" is valid for regular P-1 (these or larger bounds will always find said factor), but not valid for Brent-Suyama extension. |
[QUOTE=James Heinrich;298639]For those who are fascinated by P-1 factors found thanks to the Brent-Suyama extension, I now have a list of known ones here:
[url]http://mersenne-aries.sili.net/brent-suyama.php[/url][/QUOTE]And, vaguely related, there's now a list of the "smoothest" P-1 factors: [url]http://mersenne-aries.sili.net/smooth.php[/url] Smoothest is defined in this context as the smallest ratio of minimum-required B2 / factor-bits. For example: [url=http://mersenne-aries.sili.net/M31001941]M31,001,941[/url] has factor: 16861341412139695521233727186051728863 k = 3 × 19 × 37 × 43 × 113 × 421 × 2749 × 5639 × 10331 × 12239 × 32159 B2= 32,159 factor = 123.67 bits Note: as with the Brent-Suyama tool, I'm still populating my database with the needed info on k factorization and thereby B1/B2, so the data will get better over the next few weeks. |
I know that someone has probably asked this question dozens of times over the life of the project, but I don't seem to find the exact thing I'm wondering. How unusual, or not, is it to find a factor in a P-1 test in Stage 1 just after the Stage 1 GCD finishes so it doesn't need to do the stage 2 primes? Since I have had my new laptop, I've found a factors at this point.
|
James, if you need data, I can go through old results.txt and extract my P-1 factors along with B1 / B2 and the ram allocated.
Jwb52z, I just found a stage 1 factor ... only the second one in about 12 years of running prime ... [CODE]P-1 found a factor in stage #1, B1=605000 M57077831 has a factor: 1712714925748426197673681[/CODE] |
That's exactly why I was asking. I don't think I've found more than 5 during the entire run of the project that I've participated in, I stopped for a while once, since it started. Maybe it's just more frequent feeling because of computer speeds being increased now compared to the past.
|
[QUOTE=Jwb52z;298836]I know that someone has probably asked this question dozens of times over the life of the project, but I don't seem to find the exact thing I'm wondering. How unusual, or not, is it to find a factor in a P-1 test in Stage 1 just after the Stage 1 GCD finishes so it doesn't need to do the stage 2 primes? Since I have had my new laptop, I've found a factors at this point.[/QUOTE]
I'd say... with default bounds by Prime95, it's probably around 40-60 S1-S2, or maybe a bit lower. It shouldn't be any less than 25-75. [QUOTE=PageFault;298837]James, if you need data, I can go through old results.txt and extract my P-1 factors along with B1 / B2 and the ram allocated. Jwb52z, I just found a stage 1 factor ... only the second one in about 12 years of running prime ... [CODE]P-1 found a factor in stage #1, B1=605000 M57077831 has a factor: 1712714925748426197673681[/CODE][/QUOTE] Why not just [URL="http://mersenne-aries.sili.net/index.php?submitresults=1"]upload/submit your results[/URL] to his website? That's what the rest of us do. Edit: Note to James, the link you provided for smooth talks about regular-smoothness, but B1 and B2 are actually power-smooth bounds, as axn described earlier. Probably best solution is that someone edits the wiki page to include both descriptions. (Or you could link to Wikipedia's [URL="http://en.wikipedia.org/wiki/Smooth_number"]article[/URL], which includes the powersmooth definition.) |
[QUOTE=Jwb52z;298836]I know that someone has probably asked this question dozens of times over the life of the project, but I don't seem to find the exact thing I'm wondering. How unusual, or not, is it to find a factor in a P-1 test in Stage 1 just after the Stage 1 GCD finishes so it doesn't need to do the stage 2 primes? Since I have had my new laptop, I've found a factors at this point.[/QUOTE]Overall, it's not that uncommon. From a recent copy of my data, I have record of 3058 P-1 factors, of which 691 (23%) were stage 1 and 2367 were stage 2 (77%). If you care to have that broken down by range:[code]Range Stage1 Stage2
All 691 2367 0M-10M 295 1529 10M-20M 179 453 20M-30M 13 85 30M-40M 18 25 40M-50M 39 97 50M-60M 138 173 60M+ 9 5[/code]Personally, I've found 21 factors in stage1 on my main machine in the last 5 months, and 79 on my other machine in the last 3 years. |
[QUOTE=PageFault;298837]James, if you need data, I can go through old results.txt and extract my P-1 factors along with B1 / B2 and the ram allocated.[/QUOTE]If you have old results.txt, please submit them all to [url]http://mersenne-aries.sili.net[/url] if you haven't already. Any data I need will be automatically parsed from that.
|
Ok James, will do ...
You find lots of factors. I have a spreadsheet somewhere and my batting average for P-1 is below average, around .03 or so. It is in recent months that I put a core on P-1 (so far 2/34 successes), the bulk were done in the course of first time testing. I added some ram last week but it doesn't seem to change the bounds. It does run faster, cutting stage 2 time from ~30 to 23 hours. What is noticeable is the # of relative primes per batch - this increased to 72 from 40. Maybe my box is too old - Pentium D 3.4 GHz with 1920 MB allocated to prime95. |
[QUOTE=PageFault;298843]I added some ram last week but it doesn't seem to change the bounds... # of relative primes per batch - this increased to 72 from 40.[/QUOTE]Your box is on the older side, but can still provide some useful contribution. Moreover, your success rate doesn't depend a whole lot on the speed of your box, although obviously you won't push through nearly as many exponents.
I tend to find a lot of factors relative to my processing power because my older box (which is still an i7-920 with 12GB) is running through [url=http://mersenne-aries.sili.net/p1small.php]previously-badly-done P-1[/url] assignments. For exponents in that range it's no problem to run a full 480 relative primes at once with a modest amount of RAM, and my i7-920 can push through about 3 exponents an hour [I]per core[/I]. My main box can also run 480/480 RPs at once, but that takes about 12GB for a "current" exponent in the 55M range. |
James, I looked at your link, P-1 poorly or not done. This is interesting ... some questions:
I clicked on "get exponents" but nothing happens, no matter what I selected. Maybe because I turned off alot of crap in IE ... If I were to get exponents, how to submit results - manual or primenet? Do I get credited? Thanks, PF |
If you fill in a starting exponent you'll get some data. Or just click one of the ranges from the "worst-done ranges" list. For example, [url=http://mersenne-aries.sili.net/p1small.php?prob=3&min=8300000&max=8400000&onlystage1=0&ignorenop1=0&showassigned=0]M8,300,000[/url]. You'll see the list of badly-done P-1, either with no stage-2, or ill-chosen bounds, either way with an effective probability of <3% (compared to standard ~5-6%). Copy-past as many assignments as you like from the left column (e.g. "Pfactor=1,2,8317429,-1,64,2") into your worktodo.txt (after stopping Prime95), restart Prime95 and let it get assignment IDs for all the work you just added. You'll certainly get PrimeNet credit for any work you do that way (Prime95 will still automatically report results when complete, you're just assigning them manually).
Note: working in this range you're not saving anyone any L-L or DC effort. But you will eventually save someone some ECM effort with any factors found. And finding factors is fun! :smile: |
Thank you for the information James Heinrich. I do appreciate your time spent locating it for me.
|
[QUOTE=James Heinrich;298842]If you have old results.txt, please submit them all to [url]http://mersenne-aries.sili.net[/url] if you haven't already. Any data I need will be automatically parsed from that.[/QUOTE]
Do you also need results from mfaktc and/or cudaLucas? Luigi |
[QUOTE=ET_;298897]Do you also need results from mfaktc and/or cudaLucas?
Luigi[/QUOTE] He doesn't [i]need[/i] them per se, but his site is capable of processing all results and tracks them; I upload everything. He has some nice features besides the collective P-1 tables. |
[QUOTE=Dubslow;298900]He doesn't [i]need[/i] them per se, but his site is capable of processing all results and tracks them; I upload everything. He has some nice features besides the collective P-1 tables.[/QUOTE]
I just noticed :smile: I will upload more results as I get home. Too bad I don't backup mfaktc results files after submission. Luigi |
[QUOTE=ET_;298897]Do you also need results from mfaktc and/or cudaLucas?[/QUOTE]I grab snippets of data from PrimeNet every hour, but unfortunately I can't get all the data that way (the reports are truncated), and not all the data I need is reported there either, so any and all results you want to submit (Prime95, mprime, mfaktc, mfakto, CudaLucas, factor5) is all very welcome.
|
James,
I fired up a small batch of the M83xxxxx exponents with poor P-1. They are moving quite fast and I'm going to help you out in properly testing these exponents. I copied all the tests from your web page (M8317429 - M8392961) and I'll put this into a spreadsheet and send you a copy so we don't cross paths. I have a full workload right now and I want to reduce my already assigned work so I'll start for serious in a few weeks or so. Cheers, PF |
[QUOTE=PageFault;299083]I'll put this into a spreadsheet and send you a copy so we don't cross paths.[/QUOTE]No need, just let Prime95 register the assignments on PrimeNet and we won't cross paths (nor will anyone else who happens to want to work in that range).
|
| All times are UTC. The time now is 10:25. |
Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.