Prime95 version 25.6 - Page 9

Prime95 · 2008-01-27, 23:06

Quote:

Originally Posted by harlee

The server is reporting that the mprime program from mprime255-MacOSX.tar.gz is untrusted (v25.4 is fine).

I think I fixed this.

Quote:

Also, sometime in the future will there be a 64-bit version for MacOSX?

Maybe, I'd have to do some research. I also don't know if Agner Fog's object file conversion program handles 64-bit Mac OSX.

harlee · 2008-01-27, 23:42

Quote:

Originally Posted by Prime95

I think I fixed this.

Confirmed

James Heinrich · 2008-01-27, 23:43

Quote:

Originally Posted by gowen72

Any idea when the 64-bit windows version will be available?

I second the request... I'm still running 25.5 on my Vista64 machine (don't worry, not doing TF), since I figured the 64-bit 25.6 would be released in short order after the 32-bit version, but it's been more than a month now...

garo · 2008-01-28, 16:20

Now that ftp is working again, perhaps George could upload the Windows and Linux versions of 25.6 on the site. Thanks.

2008-01-29, 16:16

Hi All...

First off, thank you for reading this message. Hopefully you can shed some light on an issue I'm having testing my QX6850 quad core CPU on the EVGA 680i motherboard.

Is there any reason why my system would fail the Large FFT test on v25.5 after 20 minutes, while it passed at least two tests for 8+ hours on v25.6? Rounding error checking is enabled on both. Also, I have been able to pass MemTestx86 overnight as well, so there should be no memory problems.

I appreciate your help!

--Phil

chappjc · 2008-01-30, 05:20

Thanks for the Windows 64 build. I will let you know if the fight for memory is fixed with 4 worker threads. Stage 1 should be done soon.

Because of the stated SSE2 bug in 25.5, I restarted all exponents back to p-1 stage 1 (only a few days wasted). Correct to do?

Unfortunately, 25.5 just passed 4 exponents since Jan 27. Are the following results not to be trusted? Or does that just mean I may have wasted time doing the primality tests?

Code:

[Sun Jan 27 03:52:12 2008]
UID: _/_, M40025417 is not prime. Res64: 7D1675C6CBA4E67E. Wd4: BE5E775F,34531919,00000000
[Sun Jan 27 04:22:28 2008]
UID: _/_, M40025551 is not prime. Res64: 4C901AEC3D7B2B54. Wd4: 7EE6FCE3,20678651,00000000
[Sun Jan 27 08:19:27 2008]
UID: _/_, M40025459 is not prime. Res64: D414865A4C58F158. Wd4: D626DB9E,39940620,00000000
[Tue Jan 29 08:53:51 2008]
UID: _/_, M40197013 is not prime. Res64: B6012BC9717CBD38. Wd4: 8A17136E,328892,00000000

FYI: I am running a Q6600, 4GB ram w/ hardware memory remapping enabled, and Windows XP Pro x64. I only give prime95 1024MB of the 4094MB (all of which XP x64 sees).

On a side note, the torture test still shows 1750MB as maximum selectable memory to use. I am sure this has been addressed elsewhere, but I am not to concerned about it.... just FYI in case that means anything. Yes, I have properly configured my machine to use all 4GB and I use it regularly in other 64-bit scientific apps.

moo · 2008-01-30, 05:48

I am doing tf-lmh i have 7 computers running about 8 hours a day. it is working real well...

Prime95 · 2008-01-30, 13:07

Quote:

Originally Posted by chappjc

Because of the stated SSE2 bug in 25.5, I restarted all exponents back to p-1 stage 1 (only a few days wasted). Correct to do?

The SSE2 bug only affected trial factoring above 2^64. Your P-1 jobs did not need to be restarted. Your 4 LL results are unaffected as well.

chappjc · 2008-01-30, 18:58

Cool. I don't know what is meant by "trial factoring above 2^64"...

I am just testing for fun and don't know the math behind it. Anyway I still am curious to see what happens in stage 2 with four worker threads.

James Heinrich · 2008-02-01, 02:37

Quote:

Originally Posted by chappjc

Anyway I still am curious to see what happens in stage 2 with four worker threads.

Prime95 v25.6 (32-bit) seems to have issues with RAM allocation with 4 threads running. Thread #2 is off doing TF by itself and not bothering anyone, but threads 1,3,4 are fighting for RAM. And thread #1 loses badly:

Code:

<worker #1>
[Work thread Jan 31 20:35] Other worker threads are using lots of memory now.
[Work thread Jan 31 20:35] Looking for work that uses less memory.
[Work thread Jan 31 20:35] Using FFT length 14K
[Work thread Jan 31 20:35] ECM on M285161: curve #1 with s=5350586733974292, B1=50000, B2=5000000
[Work thread Jan 31 20:35] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Work thread Jan 31 20:35] Other worker threads are using lots of memory now.
[Work thread Jan 31 20:35] Looking for work that uses less memory.
[Work thread Jan 31 20:35] Resuming trial factoring of M56002549 to 2^67
[Work thread Jan 31 20:35] Trial factoring of M56002549 to 2^67 is 54.80% complete.
[Work thread Jan 31 20:39] Restarting thread with new memory settings.
[Work thread Jan 31 20:39] Using FFT length 64K
[Work thread Jan 31 20:39] ECM on M1278787: curve #1 with s=1998625661849865, B1=50000, B2=5000000
[Work thread Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Work thread Jan 31 20:39] Using 795MB of memory in stage 2.
[Work thread Jan 31 20:39] Out of memory!
[Work thread Jan 31 20:39] Out of memory!
[Work thread Jan 31 20:39] Out of memory!
[Work thread Jan 31 20:39] Out of memory!
[Work thread Jan 31 20:39] Work thread stopped.

Code:

<worker #2>
[Jan 31 20:33] Starting trial factoring of M56002663 to 2^64
[Jan 31 20:33] Trial factoring of M56002663 to 2^64 is 0.00% complete.
[Jan 31 20:42] Trial factoring of M56002663 to 2^64 is 59.68% complete.  Time: 528.609 sec.
[Jan 31 20:48] M56002663 no factor from 2^63 to 2^64, Wd1: 198874AE

Code:

<worker #3>
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 16K
[Jan 31 20:39] ECM on M335131: curve #1 with s=6209015767097669, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M934579: curve #1 with s=6239734167449929, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M934793: curve #1 with s=1258245562754721, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M934811: curve #1 with s=6003982973766342, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M935393: curve #1 with s=6296775646874421, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M935443: curve #1 with s=7660319945386812, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 5K
[Jan 31 20:39] ECM on M103841: curve #1 with s=5647712756922425, B1=50000, B2=5000000
[Jan 31 20:39] Restarting thread with new memory settings.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M932941: curve #1 with s=8445098530078204, B1=50000, B2=5000000
[Jan 31 20:39] Using 1010MB of memory in stage 2.
[Jan 31 20:39] Stage 2 init complete. 18650 transforms, 2 modular inverses. Time: 28.790 sec.
[Jan 31 20:40] M932941 curve 1 stage 2 at prime 4502203 [89.94%].
[Jan 31 20:40] Stage 2 complete. 75799 transforms, 1 modular inverses. Time: 22.014 sec.
[Jan 31 20:40] Stage 2 GCD complete. Time: 0.615 sec.

Code:

<worker #4>
[Jan 31 20:35] Stage 1 complete. 1286025 transforms, 1 modular inverses. Time: 36.250 sec.
[Jan 31 20:35] Using 600MB of memory in stage 2.
[Jan 31 20:35] Stage 2 init complete. 18313 transforms, 1 modular inverses. Time: 8.245 sec.
[Jan 31 20:36] M933001 curve 3 stage 2 at prime 1596211 [31.23%]. Time: 80.930 sec.
[Jan 31 20:38] M933001 curve 3 stage 2 at prime 3264047 [64.93%]. Time: 81.130 sec.
[Jan 31 20:39] Stage 2 complete. 592886 transforms, 2 modular inverses. Time: 76.642 sec.
[Jan 31 20:39] Stage 2 GCD complete. Time: 0.617 sec.
[Jan 31 20:39] M933001 completed 3 ECM curves, B1=50000, B2=5000000, Wd1: 186B2A8C
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M933019: curve #1 with s=6299273903278204, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Other worker threads are using lots of memory now.
[Jan 31 20:39] Looking for work that uses less memory.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M933059: curve #1 with s=4568838324112108, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Using 600MB of memory in stage 2.
[Jan 31 20:39] Restarting thread with new memory settings.
[Jan 31 20:39] Restarting thread with new memory settings.
[Jan 31 20:39] Using FFT length 48K
[Jan 31 20:39] ECM on M933019: curve #1 with s=6299273903278204, B1=50000, B2=5000000
[Jan 31 20:39] Stage 1 complete. 41 transforms, 1 modular inverses. Time: 0.005 sec.
[Jan 31 20:39] Using 600MB of memory in stage 2.
[Jan 31 20:39] Stage 2 init complete. 18313 transforms, 1 modular inverses. Time: 28.552 sec.
[Jan 31 20:39] M933019 curve 1 stage 2 at prime 53147 [0.06%].

In short, 3 threads seem to simultaneously try to allocate 600+1010+795 = 2405MB (of 1792MB configured as available). I guess there must be a few cycles between a thread deciding how much memory is already allocated (and therefore how much it can claim) and when it actually allocates that memory for itself, such that two poorly timed threads can simultaneously decide they can use the same available memory (and one of them fails and dies). In this instance, thread #1 just stopped itself. I came home from work today to find Prime95 completely crashed (probably due to another instance of a similar issue?)

James Heinrich · 2008-02-01, 02:47

I think a number of issues (threads fighting for memory, threads idle due to lack of memory while other no-RAM work is waiting, etc) could be solved by making the worker threads less autonomous and making the main thread more of a controlling supervisor -- it decides which threads get how much RAM, and which workers should work on which work (rather than having the worktodo hardcoded to a specific worker thread). Yes, I do realize this is probably not a trivial change, but I think it's probably an important issue to consider as CPUs scale from 1 to 2 to 4 to ... cores.

2008-01-30, 05:20	#94
chappjc Jul 2007 10100₂ Posts	Thanks for the Windows 64 build. I will let you know if the fight for memory is fixed with 4 worker threads. Stage 1 should be done soon. Because of the stated SSE2 bug in 25.5, I restarted all exponents back to p-1 stage 1 (only a few days wasted). Correct to do? Unfortunately, 25.5 just passed 4 exponents since Jan 27. Are the following results not to be trusted? Or does that just mean I may have wasted time doing the primality tests? Code: [Sun Jan 27 03:52:12 2008] UID: _/_, M40025417 is not prime. Res64: 7D1675C6CBA4E67E. Wd4: BE5E775F,34531919,00000000 [Sun Jan 27 04:22:28 2008] UID: _/_, M40025551 is not prime. Res64: 4C901AEC3D7B2B54. Wd4: 7EE6FCE3,20678651,00000000 [Sun Jan 27 08:19:27 2008] UID: _/_, M40025459 is not prime. Res64: D414865A4C58F158. Wd4: D626DB9E,39940620,00000000 [Tue Jan 29 08:53:51 2008] UID: _/_, M40197013 is not prime. Res64: B6012BC9717CBD38. Wd4: 8A17136E,328892,00000000 FYI: I am running a Q6600, 4GB ram w/ hardware memory remapping enabled, and Windows XP Pro x64. I only give prime95 1024MB of the 4094MB (all of which XP x64 sees). On a side note, the torture test still shows 1750MB as maximum selectable memory to use. I am sure this has been addressed elsewhere, but I am not to concerned about it.... just FYI in case that means anything. Yes, I have properly configured my machine to use all 4GB and I use it regularly in other 64-bit scientific apps. Last fiddled with by chappjc on 2008-01-30 at 05:27

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Prime95 version 27.3	Prime95	Software	148	2012-03-18 19:24
Prime95 version 26.3	Prime95	Software	76	2010-12-11 00:11
Prime95 version 25.5	Prime95	PrimeNet	369	2008-02-26 05:21
Prime95 version 25.4	Prime95	PrimeNet	143	2007-09-24 21:01
When the next prime95 version ?	pacionet	Software	74	2006-12-07 20:30

2008-01-28, 16:20	#92
garo Aug 2002 Termonfeckin, IE 2²×691 Posts	Now that ftp is working again, perhaps George could upload the Windows and Linux versions of 25.6 on the site. Thanks.

2008-01-29, 16:16	#93
Yankee516 2²·3·521 Posts	Hi All... First off, thank you for reading this message. Hopefully you can shed some light on an issue I'm having testing my QX6850 quad core CPU on the EVGA 680i motherboard. Is there any reason why my system would fail the Large FFT test on v25.5 after 20 minutes, while it passed at least two tests for 8+ hours on v25.6? Rounding error checking is enabled on both. Also, I have been able to pass MemTestx86 overnight as well, so there should be no memory problems. I appreciate your help! --Phil

2008-01-30, 05:48	#95
moo Jul 2004 Nowhere 809 Posts	I am doing tf-lmh i have 7 computers running about 8 hours a day. it is working real well...

2008-01-30, 18:58	#97
chappjc Jul 2007 2²×5 Posts	Cool. I don't know what is meant by "trial factoring above 2^64"... I am just testing for fun and don't know the math behind it. Anyway I still am curious to see what happens in stage 2 with four worker threads.

2008-02-01, 02:47	#99
James Heinrich "James Heinrich" May 2004 ex-Northern Ontario 11×311 Posts	I think a number of issues (threads fighting for memory, threads idle due to lack of memory while other no-RAM work is waiting, etc) could be solved by making the worker threads less autonomous and making the main thread more of a controlling supervisor -- it decides which threads get how much RAM, and which workers should work on which work (rather than having the worktodo hardcoded to a specific worker thread). Yes, I do realize this is probably not a trivial change, but I think it's probably an important issue to consider as CPUs scale from 1 to 2 to 4 to ... cores.