## Other

drummers-lowrise

Message boards : Project Staging Area : Generalized Fermat Prime Search

 Subscribe SortOldest firstNewest firstHighest rated posts first
Author Message
John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 18918 - Posted: 5 Nov 2009 | 22:28:34 UTC

Welcome to PrimeGrid's Generalized Fermat Prime Search

The numbers F(b,n) = b^2^n+1 (with n and b integers, b greater than one) are called generalized Fermat numbers. In the special case where b=2, they are called Fermat numbers, named after Pierre de Fermat who first studied them.

This search is for primes of the form b^2^n+1. With David Underbakke's recent improvements to AthGfn64 and his development of Genefx64, the Generalized Fermat Prime Search has become very attractive.

A previous project already completed a substantial amount of work. The Generalized Fermat Prime Search was a premier project from 2001-2004 ranking second only to GIMPS in organization and size of primes found. We'll be doing some spot double checks as well as filling in the gaps that remain. Then we'll advance the limits already established.

As mentioned earlier, David Underbakke's AthGfn64 sieve program will be used to sieve. His Genefx64 program will be used to PRP test the remaining candidates and PFGW will be used to primality test all PRP's found. Once the b limits of Genefx64 have been reached, we'll switch to Genefer, developed by Yves Gallot. After the b limits of Genefer have been reached, we'll use his Genefer80. Finally, when those b limits have been reached, we'll finish the rest of the range with PFGW.

The search range is as follows:

n = 2, 4, 8, 16 for b < 1,000,000,000
n = 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144, 524288, 1048576, 2097152, 4194304, 8388608, 16777216 for n < 100,000,000

Much of the smaller n work has already been completed. The main search will start at n=32768 in PRPNet (prpnet.primegrid.com:12005). We'll test some gaps and then start testing above 3149688^32768+1.

As sieving progresses, 65536, 131072*, 262144, and 524288 will be added to PRPNet so users can have a selection of testing times from which to choose. NOTE: There's only been 1 prime found at n=262144 and none found at n=524288. Therefore, finds at those levels will be quite significant. Of course, they'll take some time to test as well. :)

*131072 is currently reserved.

Sieve depths as of 4 November are:

n = 32768; p=500P
n = 65536; p=500P
n = 131072; p=395P
n = 262144; p=530P
n = 524288; p=120P

Sieving will continue throughout the search. We hope to push all of these up to 1E within the next month or so.

A special thanks to David Underbakke for his AthGfn6v and Genefx64 and to Mark Rodenkirch for PRPNet. Also, thanks to them both for their collaboration in updating Genefx64, Genefer, and Genefer80 to prepare the programs for a distributed effort.

We'll be testing the following for Top 5000 primes:
NOTE: top b limits are listed as GeneferCUDA, Genefx64, Genefer, Genefer80

n=32768
553602<b<743788
855124<b<999236
1277444<b<1519380
1519380<b<1755378
1755378<b<1909372
2034902<b<2147196
2167350<b<2279250
2696506<b<2973894
3000336<b<3109540
3149688<b<10021136
10021136<b
Top limits: 1.87M, 1.5M, 2.1M, 65M+

n=65536
671600<b<843832
1057476<b<1266062
1540550<b<1828502
1874512<b<2162068
2187182<b<17629398
17684828<b<19502212
19502212<b
Top limits: 1.53M, 1.2M, 1.7M, 51M+

n=131072
130816<b<386892
386892<b<572186
572186<b<1176694
1176694<b<1361244
1372930<b
Top limits: 1.255M, 1M, 1.4M, 42M+

n=262144
24518<b
Top limits: 815K, 865K, 1,2M, 35M+

n=524288
2<b
Top limits: 710K, 735K, 975K, 29M+
____________

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 18937 - Posted: 6 Nov 2009 | 19:42:22 UTC

Whilst I understand that "We'll be doing some spot double checks as well as filling in the gaps that remain", it seems that the only n being tested at present is 15 (i.e. 2^15 = 32768) and on ranges previously covered.

When will we be moving to new test ranges?

____________
Warped

John M. Johnson "Novex"
Volunteer tester

Joined: 16 Aug 07
Posts: 625
ID: 10876
Credit: 1,066,951
RAC: 0

Message 18938 - Posted: 6 Nov 2009 | 19:50:50 UTC

I'm curious about that myself, but then again my friend we just started GFN a few days ago so give it a few weeks I'm sure :)
____________

John M. Johnson "Novex"

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 18940 - Posted: 6 Nov 2009 | 19:56:27 UTC

Fair enough. It's not a criticism, merely my curiosity.

Essentially, I'm interested in what the plans are for the different n and what ranges will be double-checked before we get into virgin territory.
____________
Warped

John M. Johnson "Novex"
Volunteer tester

Joined: 16 Aug 07
Posts: 625
ID: 10876
Credit: 1,066,951
RAC: 0

Message 18942 - Posted: 6 Nov 2009 | 20:04:37 UTC - in response to Message 18938.

I'm curious about that myself, but then again my friend we just started GFN a few days ago so give it a few weeks I'm sure :)

Aye, we both have eager beaver curiosity :)
____________

John M. Johnson "Novex"

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 18961 - Posted: 7 Nov 2009 | 1:31:31 UTC - in response to Message 18937.

Whilst I understand that "We'll be doing some spot double checks as well as filling in the gaps that remain", it seems that the only n being tested at present is 15 (i.e. 2^15 = 32768) and on ranges previously covered.

When will we be moving to new test ranges?

We are exploring "gaps" in the primes posted at the Prime Pages to see if they are true gaps or caused by missed primes. A quick glance shows these as possible "gaps" for n=32768:

553602<b<743788 (currently testing)
855124<b<999236
1277444<b<1519380
1519380<b<1755378
1755378<b<1909372
2034902<b<2147196
2167350<b<2279250
2696506<b<2973894
3000336<b<3109540
3149688<b<10021136
10021136<b<100M

While most of these ranges are shown as complete by the previous project, we suspect there might be a few primes in there. The only way to know for sure is to double check their entire work...which is not our intention at this time.

We started at 32768 because that's the lowest n that will qualify for the top 5000 primes and the tests themselves are relatively quick.

As sieving advances on the other n, we'll insert those into PRPNet as well. As n increases, the "gaps" are fewer so we'll enter untested territory sooner.

A prime find at n=262144 or 524288 will be quite significant! :)
____________

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 18967 - Posted: 7 Nov 2009 | 4:00:36 UTC - in response to Message 18961.

Thanks John. As usual, you're a mine of information.

As a matter of interest, who is doing the sieve work?

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 18970 - Posted: 7 Nov 2009 | 8:10:41 UTC - in response to Message 18967.

As a matter of interest, who is doing the sieve work?

Whoops, what a silly question! We're doing it over at the same place as PPSE Sieve! Oh well, at least I realised the error while out cycling.

John M. Johnson "Novex"
Volunteer tester

Joined: 16 Aug 07
Posts: 625
ID: 10876
Credit: 1,066,951
RAC: 0

Message 18988 - Posted: 7 Nov 2009 | 16:47:30 UTC - in response to Message 18970.

As a matter of interest, who is doing the sieve work?

Whoops, what a silly question! We're doing it over at the same place as PPSE Sieve! Oh well, at least I realised the error while out cycling.

Heh yeah I find going for a drive clears my thoughts if I'm stuck on a problem or project that I can't solve, it normally comes to me in the drive :) On another note, I see GFN is getting more people to join since I've been doing WU's as well thats awesome for the new project. Your crunching them as well right Warped?
____________

John M. Johnson "Novex"

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 18995 - Posted: 7 Nov 2009 | 18:28:51 UTC - in response to Message 18988.

On another note, I see GFN is getting more people to join since I've been doing WU's as well thats awesome for the new project. Your crunching them as well right Warped?

Yes John, I've been doing some at both port 12003 and port 12005. I'm back onto sieve work now though. I participated in the original project back in 2002 as well.

Good luck - there's plenty of these to crunch. :-)

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 19014 - Posted: 8 Nov 2009 | 15:17:45 UTC

We are empty on 12005.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Warped

Joined: 25 Aug 08
Posts: 288
ID: 27792
Credit: 27,584,075
RAC: 0

Message 20268 - Posted: 4 Jan 2010 | 18:02:01 UTC

I've been crunching a few if these again after being away for my annual holiday. I note that we have progressed to n=65536. However, I have noticed a number of instances where a particular core completes a task (say 673074^65536+1) only for another core to pick up that same task from scratch again. Is each number being crunched twice for validation purposes?
____________
Warped

Lennart SM5YMT
Honorary cruncher

Joined: 7 May 07
Posts: 1125
ID: 7989
Credit: 694,692,344
RAC: 0

Message 20269 - Posted: 4 Jan 2010 | 19:09:22 UTC - in response to Message 20268.

I have increased time now. Hope that will solve the problem.

Lennart

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 20270 - Posted: 4 Jan 2010 | 19:54:28 UTC - in response to Message 20269.

There are a couple of components here that could affect this.

First, expiration. The server has a hard time limit in which a work unit must be completed. This is based upon the number of decimal digits of that workunit. This is what happened in your case.

Second, double-checking. The server has a switch that enables double-checks. This double-check is used to ensure that two tests for the same number have the same result. If the results do not match, then one (or both) of the tests are invalid and another test has to be done until two tests have matching results. It is extremely unlikely (well under 1/2^64) for a test to have an error, for that error to be undetected by the software, for the test to complete, for the client to report a residue for the test (instead of reporting an error), and for that residue to match a residue from a valid test. The double-checking logic can be configured on the server side to ensure that a test is run on a different computer than other tests for that number. I presume that Lennart has the server configured to run double-checks on a different user/client combination. If not, it would be possible for you to get a double-check on the same computer that did the first test.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 20271 - Posted: 4 Jan 2010 | 21:02:53 UTC - in response to Message 20270.

I presume that Lennart has the server configured to run double-checks on a different user/client combination. If not, it would be possible for you to get a double-check on the same computer that did the first test.

Since genefer.exe roundofferr, I stopped work on port 12005 which basiccaly ment no clients working on that port.
I assume there are only a few clients working on that port -> higher chance of doing double checks on the same user/client.
(looking at http://prpnet.primegrid.com:12005/pending_tests.html - there are no pending tests at the moment)
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 20274 - Posted: 5 Jan 2010 | 1:31:35 UTC - in response to Message 20271.

Since genefer.exe roundofferr, I stopped work on port 12005 which basiccaly ment no clients working on that port.
I assume there are only a few clients working on that port -> higher chance of doing double checks on the same user/client.
(looking at http://prpnet.primegrid.com:12005/pending_tests.html - there are no pending tests at the moment)

If you have genefer80, then the client will try to test with that version of genefer. If you have PFGW, phrot, or LLR configured, it will try with those. This was added in the 2.4.7 client.

unconnected
Volunteer tester

Joined: 4 Apr 09
Posts: 94
ID: 37970
Credit: 312,957,509
RAC: 8,654

Message 20792 - Posted: 3 Feb 2010 | 7:32:57 UTC

We are empty on port 12003.

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 24225 - Posted: 9 Jun 2010 | 17:47:01 UTC

GFN Update

It's been a long time coming but it's finally here. N=262144 and 524288 have now been added to PRPNet!!! Thanks to Honza for coordinating the sieve and to the many others who helped him sieve, we now have sieve depths at unprecedented levels. :) We will resume sieving at some point in the future but for now, it's time to see if we can find a prime. The sieves are at the following depths:

• N=32768, suspended at 1001P
• N=65536, suspended at 1500P
• N=131072, suspended at 2000P
• N=262144, suspended at 2510P
• N=524288, suspended at 3070P

There has only been 1 prime found at N=262144 and none found at N=524288. Therefore, finds at these levels will be quite significant...and of course, their testing times will be quite long too. Primes are at Mega Prime level. We have yet to do any testing of N=524288.

To participate in any of the GFN searches, uncomment the following in the master_prpclient.ini file:

//server=GFN32768:0:1:prpnet.primegrid.com:12005
//server=GFN65536:0:1:prpnet.primegrid.com:12003
//server=GFN262144:0:1:prime2u.com:11002
//server=GFN524288:0:1:prime2u.com:11001

Note: N=131072 is reserved

Win64 users have the best advantage with these searches as they can utilize Genefx64 with is the fastest of the four programs (Genefer, Genefer80, and pfgw are the other three). While these searches are not limited to only Win64 users, testing times using the other programs are longer.

The upper bounds for each program are as follows:

Genefx64 Genefer Genefer80 pfgw The upper bound m = 32768, b = 1,510,000 : 2,085,000 : 63,660,000 : + The upper bound m = 65536, b = 1,240,000 : 1,695,000 : 50,980,000 : + The upper bound m = 131072, b = 1,025,000 : 1,365,000 : 41,850,000 : + The upper bound m = 262144, b = 865,000 : 1,170,000 : 34,790,000 : + The upper bound m = 524288, b = 735,000 : 975,000 : 28,820,000 : +

So as you can see, N=524288 can be tested up to b=735,000 using Genefx64 before switching over to Genefer to be tested up to b=975,000. Currently, only N=32768 can no longer utilize Genefx64 or Genefer as testing has already reached b=3,169,866.

Finally, the programs are only available as follows:

• GenefX64 - Version 2.2.0 (Windows 64 bit only)
• Genefer - Version 2.2.0 (Linux, Windows, MacIntel, MacPPC)
• Genefer80 - Version 2.2.0 (Windows only)
• pfgw - Version 3.3.4 (Linux, Windows, MacIntel)

Best of Luck in the search. Who's going to make history by finding the first N=524288 prime??? :)
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24231 - Posted: 9 Jun 2010 | 23:10:49 UTC - in response to Message 24225.

Note: N=131072 is reserved

What does this mean? Does this mean the PrimeGrid has it reserved or does someone else have it reserved?

Scott Brown
Volunteer moderator
Volunteer tester
Project scientist

Joined: 17 Oct 05
Posts: 2107
ID: 1178
Credit: 8,211,199,358
RAC: 4,821,823

Message 24233 - Posted: 9 Jun 2010 | 23:22:45 UTC - in response to Message 24231.

Note: N=131072 is reserved

What does this mean? Does this mean the PrimeGrid has it reserved or does someone else have it reserved?

I'm guessing that Puzzle Peter is probably doing this range with all those "non-networked" boxes of his. :)

____________
141941*2^4299438-1 is prime!

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24234 - Posted: 9 Jun 2010 | 23:45:10 UTC - in response to Message 24233.

Note: N=131072 is reserved

What does this mean? Does this mean the PrimeGrid has it reserved or does someone else have it reserved?

I'm guessing that Puzzle Peter is probably doing this range with all those "non-networked" boxes of his. :)

If so, then good luck to him in finding some primes.

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24240 - Posted: 10 Jun 2010 | 17:16:58 UTC

I picked up the first test for n=524288 for my Win64 machine and it came up with this:

[2010-06-10 03:14:39 GMT] Candidate: 4^524288+1 Program: genefX64.exe Residue: PRP Time: 0 seconds
[2010-06-10 03:34:06 GMT] PFGW could not prove primality for 4^524288+1. Time: 1167 seconds

I know little about these things but aren't composite PRP's rare among huge numbers such as this? So I'm really asking has my computer produced an erroneous result? It's a C2Q Q9550 at stock clock (2.83 GHz). I did interrupt the test once to change the PRPNet server configuration.

Also, in the above quote from test_results.log the genefX64 run time is missing. Should it be there? The program itself reported a run time of 5h 36min.
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24241 - Posted: 10 Jun 2010 | 18:43:07 UTC - in response to Message 24240.

I picked up the first test for n=524288 for my Win64 machine and it came up with this:
[2010-06-10 03:14:39 GMT] Candidate: 4^524288+1 Program: genefX64.exe Residue: PRP Time: 0 seconds
[2010-06-10 03:34:06 GMT] PFGW could not prove primality for 4^524288+1. Time: 1167 seconds

I know little about these things but aren't composite PRP's rare among huge numbers such as this? So I'm really asking has my computer produced an erroneous result? It's a C2Q Q9550 at stock clock (2.83 GHz). I did interrupt the test once to change the PRPNet server configuration.

Also, in the above quote from test_results.log the genefX64 run time is missing. Should it be there? The program itself reported a run time of 5h 36min.

The time of 0 seconds is extremely suspicious. It seems that genefx64 might have had an error and that PRPNet was unable to detect it. Can you run it from the command line and post the output? We might need to get some assistance from David on this.

Sideshow_Larry

Joined: 10 May 09
Posts: 15
ID: 39800
Credit: 44,805,529
RAC: 0

Message 24242 - Posted: 10 Jun 2010 | 19:24:01 UTC - in response to Message 24241.

I've got work units that range from 40^524288+1 to 60^524288+1 running on a couple of Win64 machines (one Vista one Win7) and they have been crunching all morning so I'm looking at around 18 hrs to complete. Getting one of the first units might crunch in very little time IMO. It would seem that it would be easy to test that particular unit (except I'm not certain how to actually accomplish that part).

____________

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24244 - Posted: 10 Jun 2010 | 19:50:50 UTC - in response to Message 24241.

Sorry, this is all in the middle of the Lost finale here but I have it running now. 1048575 steps to go so it'll be a while. I'll post results in the morning (0400 UTC).

As I said, I interrupted the first run with ^C. It did say this

[2010-06-09 23:44:20 GMT] GFN524288: No data in file [genefer.log]. Is genefer broken?
but resumed from checkpoint.
____________

[SG]Puzzle-Peter
Volunteer tester

Joined: 14 Jun 08
Posts: 374
ID: 24128
Credit: 92,652,187
RAC: 0

Message 24245 - Posted: 10 Jun 2010 | 20:26:28 UTC - in response to Message 24233.

I'm guessing that Puzzle Peter is probably doing this range with all those "non-networked" boxes of his. :)

No, I am busy at the 27121 project. Must be somebody else.
____________
There are only 10 kinds of people - those who understand binary and those who don't

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24248 - Posted: 10 Jun 2010 | 21:44:41 UTC - in response to Message 24244.

Sorry, this is all in the middle of the Lost finale here but I have it running now. 1048575 steps to go so it'll be a while. I'll post results in the morning (0400 UTC).

As I said, I interrupted the first run with ^C. It did say this
[2010-06-09 23:44:20 GMT] GFN524288: No data in file [genefer.log]. Is genefer broken?
but resumed from checkpoint.

genefer shouldn't create that log file unless it will write to it. This is rather mysterious. Did it say where the checkpoint resumed from?

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24249 - Posted: 10 Jun 2010 | 22:15:29 UTC - in response to Message 24248.

Sorry, this is all in the middle of the Lost finale here but I have it running now. 1048575 steps to go so it'll be a while. I'll post results in the morning (0400 UTC).

As I said, I interrupted the first run with ^C. It did say this
[2010-06-09 23:44:20 GMT] GFN524288: No data in file [genefer.log]. Is genefer broken?
but resumed from checkpoint.

genefer shouldn't create that log file unless it will write to it. This is rather mysterious. Did it say where the checkpoint resumed from?

Sorry, I closed the window in which it had run and I can't remember exactly. The file genefer.ckpt was there and I do remember thinking it didn't appear to lose much time, if any.
____________

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24251 - Posted: 11 Jun 2010 | 3:59:29 UTC - in response to Message 24241.

Command line test produced the same result:

C:\man\prpclient-3.3.0alpha-windows\programs>Genefx64.exe
GenefX64 2.2.0 (x86 - 64-bit - SSE3) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Usage: GenefX64 -b run bench
GenefX64 -t run test
GenefX64 -r run residue test
GenefX64 -q test quick expression
GenefX64 <filename> test <filename>
GenefX64 use interactive mode

1. bench
2. test
3. test residue
4. normal
4
N: 524288
b: 4
4^524288+1 is a probable prime.
(315653 digits) (err = 0.0000) (time = 5:17:32) 04:03:07
File genefer.log has the last two result lines and file verif.txt the number itself.
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24255 - Posted: 11 Jun 2010 | 12:04:15 UTC - in response to Message 24251.

Command line test produced the same result:
C:\man\prpclient-3.3.0alpha-windows\programs>Genefx64.exe
GenefX64 2.2.0 (x86 - 64-bit - SSE3) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Usage: GenefX64 -b run bench
GenefX64 -t run test
GenefX64 -r run residue test
GenefX64 -q test quick expression
GenefX64 <filename> test <filename>
GenefX64 use interactive mode

1. bench
2. test
3. test residue
4. normal
4
N: 524288
b: 4
4^524288+1 is a probable prime.
(315653 digits) (err = 0.0000) (time = 5:17:32) 04:03:07
File genefer.log has the last two result lines and file verif.txt the number itself.

It is important to run with the -r option. That is used to verify that it compiled correctly. If this number isn't prime, then there could be a bug with genefx64.

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24257 - Posted: 11 Jun 2010 | 17:13:11 UTC - in response to Message 24255.

It is important to run with the -r option. That is used to verify that it compiled correctly. If this number isn't prime, then there could be a bug with genefx64.

Output of genefx64 -r:
10234^64+1 is composite. (RES=e99e17786d3bacc2) (257 digits) (err = 0.0000) (time = 0:00:00) 19:54:27 10032^128+1 is composite. (RES=1d9767ae8d652d65) (513 digits) (err = 0.0000) (time = 0:00:00) 19:54:27 584328^256+1 is composite. (RES=32798be013527165) (1477 digits) (err = 0.0029) (time = 0:00:00) 19:54:27 419000^512+1 is composite. (RES=fee1d2c0894a662f) (2879 digits) (err = 0.0020) (time = 0:00:01) 19:54:28 352220^1024+1 is composite. (RES=6c7ebc76febe0583) (5680 digits) (err = 0.0022) (time = 0:00:00) 19:54:28 366672^2048+1 is composite. (RES=fd37a864102cf81e) (11396 digits) (err = 0.0039) (time = 0:00:01) 19:54:29 285064^4096+1 is composite. (RES=f81ed37ab6df33f9) (22344 digits) (err = 0.0039) (time = 0:00:05) 19:54:34 230234^8192+1 is composite. (RES=5972a302c255e081) (43927 digits) (err = 0.0039) (time = 0:00:25) 19:54:59 151902^16384+1 is composite. (RES=0a1a98e69b17f1e9) (84895 digits) (err = 0.0024) (time = 0:01:43) 19:56:42

It does say in readme_genefers.txt that genefer and genefx64 "can only
handle a limited range of lower b values." I'll run the number with genefer80 just to try and put this to bed (hopefully).
____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 24258 - Posted: 11 Jun 2010 | 18:16:46 UTC

FYI...Lennart has run -r with genefx64, genefer, and genefer80 and all three match. Additionally, he ran pfgw without -t so that we could determine residue:

4^524288+1 is composite: RES64: [11BEF945BB1F0767] (1316.5260s+0.0078s)

Sadly, this is one of those instances in which a prp is proven composite. To be safe, we have also asked David to review this. Additionally, Lennart is running the test using genefer to see how it classifies the candidate.
____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 24291 - Posted: 13 Jun 2010 | 14:48:00 UTC - in response to Message 24258.

Sadly, this is one of those instances in which a prp is proven composite. To be safe, we have also asked David to review this. Additionally, Lennart is running the test using genefer to see how it classifies the candidate.

Lennart's testing with all three genefers produced the same result. It appears that this is a known observation about the genefers. David provided this nice explanation:

The genefer implementation of the probable test is susceptible to worst case conditions where the b value (b^n+1) is a small power of 2. You will notice there is the same problem with many bases. The following all create a false positive:

2^256+1, 4^256+1, 16^256+1, 2^512+1, 4^512+1, 16^512+1, 2^1024+1, 4^1024+1, 16^1024+1, 2^2048+1, 4^2048+1, 16^2048+1, 2^4096+1, 4^4096+1, 16^4096+1,
2^8192+1, 4^8192+1, 16^8192+1, 2^16384+1, 4^16384+1, 16^16384+1, 2^32768+1, 4^32768+1, 16^32768+1

This behavior hasn't been seen before largely because all the small b values were tested before genefer was used. We can expect this special circumstance for small b's < 1024 that are powers of 2 (2,4,8,16,32,64,128,256, and 512). Other than that particular nuance, there's nothing else to be concerned about.
____________

samuel7
Volunteer tester

Joined: 1 May 09
Posts: 89
ID: 39425
Credit: 257,416,435
RAC: 0

Message 24293 - Posted: 13 Jun 2010 | 15:56:13 UTC - in response to Message 24291.

Sadly, this is one of those instances in which a prp is proven composite. To be safe, we have also asked David to review this. Additionally, Lennart is running the test using genefer to see how it classifies the candidate.

Lennart's testing with all three genefers produced the same result. It appears that this is a known observation about the genefers. David provided this nice explanation:

The genefer implementation of the probable test is susceptible to worst case conditions where the b value (b^n+1) is a small power of 2. ...

This behavior hasn't been seen before largely because all the small b values were tested before genefer was used. We can expect this special circumstance for small b's < 1024 that are powers of 2 (2,4,8,16,32,64,128,256, and 512). Other than that particular nuance, there's nothing else to be concerned about.

Thanks very much for clearing up this (non-)issue.
____________

Joined: 20 Apr 10
Posts: 17
ID: 58970
Credit: 5,466,606
RAC: 0

Message 24327 - Posted: 15 Jun 2010 | 13:00:26 UTC

And another PRP with b being small power of 2:

[2010-06-15 03:36:01 GMT] Candidate: 256^524288+1 Program: genefX64.exe Residue: PRP Time: 0 seconds [2010-06-15 08:47:54 GMT] PFGW could not prove primality for 256^524288+1. Time: 18712 seconds

What program and with what parameters should I use to prove this as prime/composite?
____________

Jay
Volunteer tester

Joined: 28 Apr 10
Posts: 82
ID: 59636
Credit: 10,419,429
RAC: 0

Message 24348 - Posted: 16 Jun 2010 | 4:43:22 UTC - in response to Message 24327.

And another PRP with b being small power of 2:

[2010-06-15 03:36:01 GMT] Candidate: 256^524288+1 Program: genefX64.exe Residue: PRP Time: 0 seconds [2010-06-15 08:47:54 GMT] PFGW could not prove primality for 256^524288+1. Time: 18712 seconds

What program and with what parameters should I use to prove this as prime/composite?

Using llr:
1*2^4194304+1 is not prime. Proth RES64: 77F6FA403A09D72B Time : 41858.673 sec.

I have no idea if llr is the appropriate program, I just thought I would give it a try. Took a lot longer than I expected (~11.6 hours).
____________

Joined: 20 Apr 10
Posts: 17
ID: 58970
Credit: 5,466,606
RAC: 0

Message 24353 - Posted: 16 Jun 2010 | 6:49:33 UTC - in response to Message 24348.

Using llr:
1*2^4194304+1 is not prime. Proth RES64: 77F6FA403A09D72B Time : 41858.673 sec.

I have no idea if llr is the appropriate program, I just thought I would give it a try. Took a lot longer than I expected (~11.6 hours).

Lennart has already explained this to me via PM. PFGW automagically kicks in when PRP is found and in my case proved this number as composite.

Looks like we just have to dig through all these b's lower than 1024...
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 24356 - Posted: 16 Jun 2010 | 12:34:02 UTC - in response to Message 24353.

I don't see any harm in removing all candidates where b is a power of 2. These are Fermat numbers and the primality of all of them that this project will touch is already known.

David

Joined: 16 Jun 10
Posts: 8
ID: 62417
Credit: 0
RAC: 0

Message 24359 - Posted: 16 Jun 2010 | 14:37:10 UTC - in response to Message 24356.

Actually, I believe only the b values that are powers of 2 and perfect squares are fermat numbers. That would be b= 2, 4, 16, 256, and 65536.

2^262144+1 which is Fermat F18
4^262144+1 = 2^524288+1 which is Fermat F19
16^262144+1 = 4^524288+1 = 2^1048576+1 which is Fermat F20
256^262144+1 = 16^524288+1 = 2^2097152+1 which is Fermat F21
65536^262144+1 = 256^524288+1 = 2^4194304+1 which is F22
65536^524288+1 = 2^8388608+1 which is F23

There are know factors for F18 through F23. In the case of the 256^524288+1, which was F22 I can give two factors, 64658705994591851009055774868504577 which is prime, and 3853959202444067657533632211*2^24+1, discovered very recently, by D. Bessell & Woltman on March 26, 2010.

I would remove only b values 2,4,16,256, and 65536.

Seventh Serenity

Joined: 18 Oct 05
Posts: 30
ID: 1184
Credit: 105,671
RAC: 0

Message 25156 - Posted: 27 Jul 2010 | 8:42:37 UTC

Port 11001 doesn't seem to be working.

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 28841 - Posted: 6 Dec 2010 | 13:58:33 UTC

Hi ,
I convert genefer.c to CUDA.
Thank you,

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 28861 - Posted: 6 Dec 2010 | 16:58:21 UTC - in response to Message 28841.

How much memory does it need? I have a card with only 128MB.
____________
Murphy (AtP)

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 28866 - Posted: 6 Dec 2010 | 17:51:15 UTC - in response to Message 28841.

Hi ,
I convert genefer.c to CUDA.
Thank you,

It can't be used by PrimeGrid/PRPNet until it supports everything in genefer 2.2.0.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 28873 - Posted: 6 Dec 2010 | 18:53:32 UTC

I've done considerable work on GFN sieving and primality test.
I would like to try GPU version for off-line running, starting small Ns (running CUDA/Win) which are not available on PRPNet.

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 28874 - Posted: 6 Dec 2010 | 19:10:55 UTC - in response to Message 28873.

I've done considerable work on GFN sieving and primality test.
I would like to try GPU version for off-line running, starting small Ns (running CUDA/Win) which are not available on PRPNet.

You still need a way to verify that there were no errors during the test. In other words, you need to compare to known residues to know that the build is valid. Version 1.3 has no code to produce residues.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 28875 - Posted: 6 Dec 2010 | 19:29:10 UTC - in response to Message 28874.

You still need a way to verify that there were no errors during the test. In other words, you need to compare to known residues to know that the build is valid. Version 1.3 has no code to produce residues.

Yes, still have .log files for N=32...2048 (up to b=100M) and N=4096 up to b=11M7
11781472^4096+1 is a probable composite. (RES=3bae6312ebca69ea) (28964 digits) (err = 0.0057) (time = 0:00:24)
I've also done about 16k tests for N=262144, b<100k
99224^262144+1 is a probable composite. (RES=9772a7694d7df19f) (1309834 digits) (err = 0.0059) (time = 10:41:11)

I guess a good base for testing original CPU Genefer 220 vs CUDA version.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 28900 - Posted: 6 Dec 2010 | 22:44:48 UTC - in response to Message 28861.

Hi ,

How much memory does it need? I have a card with only 128MB.

Need double precision support card,GTX260 is low end.

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 28922 - Posted: 7 Dec 2010 | 11:56:41 UTC - in response to Message 28875.

Hi ,

I guess a good base for testing original CPU Genefer 220 vs CUDA version.

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 28992 - Posted: 8 Dec 2010 | 7:39:58 UTC - in response to Message 28875.

Hi ,

99224^262144+1 is a probable composite. (RES=9772a7694d7df19f) (1309834 digits) (err = 0.0059) (time = 10:41:11)

I check on GTX460.

cudagenefer.0.6\$ time ./CUDAGenefer input
GeneFer 1.3 (CUDA) Copyright (C) 2001-2002, Yves GALLOT
A program for finding large probable generalized Fermat primes.

GeneFer -t run test
GeneFer <filename> test <filename>
GeneFer use interactive mode

Start test of file 'input'.
99224^262144+1 is a probable composite (err = 6.84e-03). (RES=f19f)

real 87m14.045s
user 33m6.040s
sys 54m7.760s

you can compare "f19f".

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 29280 - Posted: 14 Dec 2010 | 14:38:06 UTC - in response to Message 28992.

Hi ,
Support genefer 2.2.0.
http://mersenneforum.org/showpost.php?p=241779&postcount=29

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29771 - Posted: 21 Dec 2010 | 21:41:09 UTC

Checked Windows version (compiled by Ken using 2.3 SDK, perhaps slow for GTX 580).
Heavy CPU load for 8192, 16384 and 32768 then it come better.

I would like to try version compiled using 3.2 SDK if provided.

Generalized Fermat Number Bench
2009574^8192+1 Time: 312 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 331 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 373 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 541 us/mul. Err: 1.88e-001 395325 digits
874718^131072+1 Time: 900 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 1.33 ms/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 2.38 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 4.56 ms/mul. Err: 1.64e-001 5946413 digits
380742^2097152+1 Time: 9.17 ms/mul. Err: 3.63e-001 11703432 digits
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 29785 - Posted: 22 Dec 2010 | 1:00:38 UTC

Hi ,Honza
Nice Report.
Thank you,

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 29795 - Posted: 22 Dec 2010 | 6:19:18 UTC

Ken has built Windows version of Shoichiro's CUDAGenefer. It can be downloaded here: CUDAGenefer-0.95 (Linux64 by Shoichiro). This one was updated to 0.95 and built with CUDA 3.2. :) If you need the cufft.dll, it can be downloaded here.

We hope to have a Linux version soon. EDIT: Now available for 64 bit.

For now, this can be tested on the following GFN ports:

//server=GFN65536:0:1:prpnet.primegrid.com:12003
//server=GFN262144:0:1:prime2u.com:11002
//server=GFN524288:0:1:prime2u.com:11001

GFN32768 has already reached the max b limit of CUDAgenefer. Do not attempt this port as the PRPNet client will switch to pfgw to test.

To run CUDAGenefer, just update the genefer section in the master_prpclient.ini file before updating all the clients.

// This is the name of the genefer executables used for GFN searches. Up
// to three different Genefer programs can be specified. The client will
// attempt a test with genefx64 first. If a round off error occurs, it will
// try genefer. If a round of occurs in genefer, it will try genefer80. If
// genefer80 fails, then the number cannot be tested. The order they are
// specified here is not important.
Win32 & 64 users should have the following:
geneferexe=CUDAGenefer.exe
geneferexe=genefer.exe
geneferexe=genefer80.exe

Linux64 users should have the following:
geneferexe=./CUDAGenefer
geneferexe=./genefer

MacIntel32 users should have the following:
geneferexe=./CUDAGeneferMacIntel32
geneferexe=./genefer

MacIntel64 users should have the following:
geneferexe=./CUDAGeneferMacIntel64
geneferexe=./genefer

For Win64 users, GenefX64 will not run while using CUDAGenefer. Mark is working on a solution for the next PRPNet release which will allow GenefX64 to be included again.

p.s. If all goes well, CUDAGenefer will be added to the official packages (well, to Linux and Windows anyway). Also, we will hopefully have a response to this post in early to mid January. ;)
____________

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29797 - Posted: 22 Dec 2010 | 7:40:29 UTC

0.95 is much better, doesn't stretch CPU core to the limit with lowe exponents, only half of the core for ^8192. Also x3 faster for high n.

CUDAGen is somewhere between Genefer or Genefer80 in term of b limits, right?
It may be worth to complite 64-bit version for CUDA (closer to GeneFX64).

I'll try with PRPNet later today.

Generalized Fermat Number Bench
2009574^8192+1 Time: 265 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 271 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 276 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 332 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 397 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 579 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 931 us/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 1.59 ms/mul. Err: 1.48e-001 5946413 digits
380742^2097152+1 Time: 3.03 ms/mul. Err: 3.63e-001 11703432 digits
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

pschoefer
Volunteer developer
Volunteer tester

Joined: 20 Sep 05
Posts: 660
ID: 845
Credit: 2,125,591,761
RAC: 210,112

Message 29801 - Posted: 22 Dec 2010 | 9:25:55 UTC - in response to Message 29795.

Where can I get the required dlls without installing the complete SDK?
____________

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29803 - Posted: 22 Dec 2010 | 10:16:46 UTC

I've prepared and uploaded it to http://www.uloz.to/7057823/cufft-rar
(it's valid to 2011-01-31)

Just unRAR and put it where CUDAGenefer is and you should be ready to go.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29811 - Posted: 22 Dec 2010 | 11:57:55 UTC

Used only geneferexe=CUDAGenefer.exe and skipped other exe
(this would need client adjustement to incorporate another app for genefer)

Looks like Top b limits for CUDAGenefer are lower than stated in this thread:
maxErr exceeded for 103352^262144+1, 0.5000 > 0.4500

Will try with 524288...
EDIT: It started on port prime2u.com:11001
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29815 - Posted: 22 Dec 2010 | 13:39:51 UTC

Server accepted result on prime2u.com:11001 using CUDAGenefer/Windows.
(took ~75 minutes)
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 29825 - Posted: 22 Dec 2010 | 17:16:53 UTC

DLL mentioned earlier shoult be at http://pgllr.mine.nu/PRPNet/cufft.rar

CUDAGenefer it quite disk intensive (>2GB of writting for ^262144). I suspect checkpointing is the cause. Should be less frequent for GPU version.

103352^262144+1 is a probable composite. (RES=342b3eab10905202)
(1314474 digits) (err = 0.0044) (time = 0:44:34) 18:28:59
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

pschoefer
Volunteer developer
Volunteer tester

Joined: 20 Sep 05
Posts: 660
ID: 845
Credit: 2,125,591,761
RAC: 210,112

Message 29830 - Posted: 22 Dec 2010 | 18:38:05 UTC

A few more benchmark results:

GTX460 @ 800/1600/2000
2009574^8192+1 Time: 342 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 364 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 394 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 519 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 663 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 933 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.7 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 3.21 ms/mul. Err: 1.48e-001 5946413 digits
380742^2097152+1 Time: 6.6 ms/mul. Err: 3.63e-001 11703432 digits

GTX470 @ 750/1500/1676
2009574^8192+1 Time: 303 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 310 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 337 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 417 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 560 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 701 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.15 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 2.04 ms/mul. Err: 1.48e-001 5946413 digits
380742^2097152+1 Time: 3.87 ms/mul. Err: 3.63e-001 11703432 digits
____________

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 29839 - Posted: 22 Dec 2010 | 20:04:18 UTC - in response to Message 29825.

On my GT460/64bitLinux

103352^262144+1 is a probable composite. (RES=342b3eab10905202)
(1314474 digits) (err = 0.0044) (time = 1:46:58) 09:55:44

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 29841 - Posted: 22 Dec 2010 | 20:29:25 UTC - in response to Message 29825.

CUDAGenefer it quite disk intensive (>2GB of writting for ^262144). I suspect checkpointing is the cause. Should be less frequent for GPU version.

Reduce (1/16) next version.

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 29992 - Posted: 26 Dec 2010 | 5:32:33 UTC

MacIntel 32 and 64 CUDA executables (compute capability 1.3) from the 0.95 code have been built by Iain. We are looking for a tester to confirm they are working properly. You can download CUDAGeneferMacIntel.0.95 here: http://pgllr.mine.nu/software/genefer/

Please run it with the -r option, -l option, and then the -b option. And please post your results. The most important is the -r option which should produce matching residues with the following:

230234^8192+1 is composite. (RES=5972a302c255e081)
(43927 digits) (err = 0.0024) (time = 0:00:47) 10:38:42
151902^16384+1 is composite. (RES=0a1a98e69b17f1e9)
(84895 digits) (err = 0.0024) (time = 0:01:31) 10:40:13
177444^32768+1 is a probable composite. (RES=f69cc3e2334a43a8)
(172002 digits) (err = 0.0034) (time = 0:03:19) 10:43:32
157476^65536+1 is a probable composite. (RES=9f64b3f0d545615c)
(340605 digits) (err = 0.0042) (time = 0:08:03) 10:51:35
52186^131072+1 is composite. (RES=1b196d6c0e4d778f)
(618340 digits) (err = 0.0042) (time = 0:19:14) 11:10:49
70000^262144+1 is a probable composite. (RES=fa15b4b858fd2ff0)
(1270114 digits) (err = 0.0042) (time = 0:48:57) 11:59:46

-l option should give you something similar to this:

Generalized Fermat Number b Limits
The upper bound m = 8192, b = 2645000, Err = 0.2812
Starting b = 3340000, Err b = 2650000, Err = 0.3125, 5 Err b = 0
The upper bound m = 16384, b = 2300000, Err = 0.2969
Starting b = 2720000, Err b = 2305000, Err = 0.3047, 5 Err b = 0
The upper bound m = 32768, b = 1870000, Err = 0.2969
Starting b = 2200000, Err b = 1875000, Err = 0.3125, 5 Err b = 0
The upper bound m = 65536, b = 1530000, Err = 0.2969
Starting b = 1790000, Err b = 1535000, Err = 0.3125, 5 Err b = 0
The upper bound m = 131072, b = 1255000, Err = 0.2969
Starting b = 1450000, Err b = 1260000, Err = 0.3125, 5 Err b = 0
The upper bound m = 262144, b = 1000000, Err = 0.2891
Starting b = 1180000, Err b = 1005000, Err = 0.3125, 5 Err b = 0
The upper bound m = 524288, b = 815000, Err = 0.2969
Starting b = 960000, Err b = 820000, Err = 0.3125, 5 Err b = 0
The upper bound m = 1048576, b = 710000, Err = 0.2891
Starting b = 780000, Err b = 715000, Err = 0.3281, 5 Err b = 0
The upper bound m = 2097152, b = 575000, Err = 0.2969
Starting b = 630000, Err b = 580000, Err = 0.3125, 5 Err b = 0

And finally the -b should give you something like this:

2009574^8192+1 Time: 314 us/mul. Err: 3.82e-01 51636 digits
1632282^16384+1 Time: 321 us/mul. Err: 2.53e-01 101791 digits
1325824^32768+1 Time: 346 us/mul. Err: 2.19e-01 200622 digits
1076904^65536+1 Time: 425 us/mul. Err: 1.72e-01 395325 digits
874718^131072+1 Time: 548 us/mul. Err: 3.47e-01 778813 digits
710492^262144+1 Time: 674 us/mul. Err: 4.21e-01 1533952 digits
577098^524288+1 Time: 1.08 ms/mul. Err: 2.01e-01 3020555 digits
468750^1048576+1 Time: 1.91 ms/mul. Err: 1.64e-01 5946413 digits
380742^2097152+1 Time: 3.65 ms/mul. Err: 3.63e-01 11703432 digits

____________

mdettweiler

Joined: 25 Apr 07
Posts: 271
ID: 7489
Credit: 7,380,245
RAC: 118

Message 29993 - Posted: 26 Dec 2010 | 5:45:43 UTC - in response to Message 29992.

Quick question...why do some of the tests say "probable composite" and others just "composite"? The Fermat SPRP test as performed by PFGW (and until recently, LLR) gives a definite negative result (it's only for a positive result that there is some uncertainty). Is genefer using an entirely different kind of PRP test (one that actually can produce "probable composite" results)?

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 29994 - Posted: 26 Dec 2010 | 8:29:41 UTC - in response to Message 29993.

Quick question...why do some of the tests say "probable composite" and others just "composite"? The Fermat SPRP test as performed by PFGW (and until recently, LLR) gives a definite negative result (it's only for a positive result that there is some uncertainty). Is genefer using an entirely different kind of PRP test (one that actually can produce "probable composite" results)?

"probable" is only depend b value(b^n+1).
Goodness of Yves's hart, Probable.

mdettweiler

Joined: 25 Apr 07
Posts: 271
ID: 7489
Credit: 7,380,245
RAC: 118

Message 30004 - Posted: 26 Dec 2010 | 17:15:59 UTC - in response to Message 29994.

Quick question...why do some of the tests say "probable composite" and others just "composite"? The Fermat SPRP test as performed by PFGW (and until recently, LLR) gives a definite negative result (it's only for a positive result that there is some uncertainty). Is genefer using an entirely different kind of PRP test (one that actually can produce "probable composite" results)?

"probable" is only depend b value(b^n+1).
Goodness of Yves's hart, Probable.

Hmm...are you saying that for some values of b it can give only a probable negative result, whereas for others it can give a definite result? (I guess what I'm wondering is, what's the use of the test if it can't give you a definite answer either way? If that is the case, then none of the "probable composite" tests really tell you anything for sure.)

David

Joined: 16 Jun 10
Posts: 8
ID: 62417
Credit: 0
RAC: 0

Message 30009 - Posted: 26 Dec 2010 | 21:13:42 UTC

In regard to probable prime and probable composite, on June 17th, 2001 Yves Gallot posted a concern about the accuracy of the FFT transform used in many proving programs, include prime95 used in the Mersenne search.

Based upon his analysis, there was an extremely small possibility that many "proven primes" were actually not totally proven. That is why two different proving programs as used for the final proof of many significant primes.

For a GFN to be called a probable prime, it is actually a prime at the following probability

1 - 1/(GFN number)^2

To be a probable composite, it is a composite at the same probability

1 - 1/(GFN Number)^2

For a 500,000 digit number, the chances of being wrong are so small

1 - 1/(500000 digit number)^2 that I do not worry about being wrong.

In one sense, when genefer says a number is a probable prime, it is proven as the same level as many other programs. Yves was very particular about a proof, so left the final step to a much slower balanced FFT representation.

-------------------

(Yves Gallot posting June 17, 2001)

I wrote a new transform to compute quickly x^2 mod k.2^n+1 which can use any
base W=2^m as internal base representation. Then I tried to find the optimal
value that can be used for primality testing. Today, my conclusion is that
many numbers in the database of prime numbers (even some proven by Proth)
are just some probable primes. Please, read this and tell me what you think

First of all, we should define what we call a prime number. The mathematical
definition is not sufficient because we use some computers, some programs or
even some hand computations for the proofs and maths don't indicate what to
do with possible failures. The usage is to use an algorithm which is
mathematically exact and would give the correct answer "N is prime" or "N is
not prime" if the implementation was perfect. To check the correctness of
the implementation, usually we double check the result with an independent
computation (which can use the same hardware and software if the operations
are independent). Then let consider that
(1) N is prime if
- the algorithm used to check primality was proven to give the correct
- and the computation is double-checked with some different operations.

Today, the programs, used for the search for large primes, are based on two
algorithms: a theorem (such as Lucas' theorem or Proth's theorem) and a
transform. Some theorems exist to find some "probable prime" numbers but
they cannot be used because they are not in accordance with (1).

What about the transform? Today fast programs use some floating point FFT.
Let W the base representation, n the length of the transform and i the
number of bits in the mantissa. It is proven that if
(2) W^2 * n * log_2 n < 2^i
then the result is exact (you can find this formula in "Numerical recipes in
C: the art of scientific computing", Cambridge University Press. Some other
formulas exist with a rigorous proof on the maximum error but all of them
generate some closed bounds). The exact bound is unknown and (2) is
pessimistic: the result of the transform is often correct for some W and n
such that (2) is not verified. We can use this remark to write some fast
programs for the search for large probable primes, considering that the
output of the program is "N is a prp" or "I don't know". But if we should
prove the primality (or the converse) of a number, we should use a
transform and a base representation such that (2) (or a similar proven
formula) is true. Otherwise, the algorithm used for the "proof" has no
mathematical proof and because of (1), we cannot write that N is prime.

For example, let a GFN b^32768+1. Then n=16384 and we have i=53.
70906^2 * 16384 * 14 < 2^53 then we can use a simple transform to prove the
primality of 70906^32768+1. But 1041870^2 * 16384 * 14 > 2^53 then if we use
a simple transform for the proof and double check the result with the same
transform, we "just" prove that the probability that 1041870^32768+1 is
prime is 1/(1041870^32768+1)^2. The primality is not mathematically proven.
If we want to complete the proof, we should use a general purpose algorithm
with a 16-bit internal representation.

Let N=k.2^n+1. Today, the length of the transform used by Proth is the
smallest power of 2 larger than n/16 and W=65536. Then the proof is correct
is n < 1,048,576 otherwise the number is just a prp. An internal base
representation smaller than 2^16 should be used for the proof.

According to the base representation used by prime95, can we say that
"2^6972593-1 is prime" or that "the probability that 2^6972593-1 is prime is
1-1/(2^6972593-1)^2"?

I am myself astonished by this result, but I don't see any error in my
reasoning. I think that we (the author of programs) made a terrible
confusion between the search for large prime numbers (maybe I should write
large prp) and the primality proof of these numbers.

Yves

mdettweiler

Joined: 25 Apr 07
Posts: 271
ID: 7489
Credit: 7,380,245
RAC: 118

Message 30010 - Posted: 26 Dec 2010 | 21:32:09 UTC - in response to Message 30009.

Hmm...I'm still a little confused. I understand how probable primes work, and that a PRP is not known for sure to be prime (but nonetheless is quite likely to be) until a deterministic test, such as an N-1/N+1 test as implemented by PFGW or LLR, is performed. Such a PRP test (as I understand it) can only produce two results, PRP or composite (the latter being a definite answer)--assuming, that is, that the computer performing the test is stable and did not make any errors. Am I missing something here? Because if the PRP test works as I understand it, then there should be no such thing as a probable composite (unless you mean it in the sense that it is not known if the computer made an error in the calculation).

David

Joined: 16 Jun 10
Posts: 8
ID: 62417
Credit: 0
RAC: 0

Message 30013 - Posted: 26 Dec 2010 | 22:08:18 UTC

There are 2 parts, the mathematics which you seem to understand, and the FFT transform, the shortcut math used in the software. The transform was the item in question. There is an extremely small chance the transform has an error. Read it again about the transform. I am on cell phone now so I cannot re-enter the error formula. Yves was basically challenging the rigor of the proof for the FFT transform used in many primality proving programs including his own. The risk was so infinitesimally small no one ever addressed it.

David

Joined: 16 Jun 10
Posts: 8
ID: 62417
Credit: 0
RAC: 0

Message 30014 - Posted: 26 Dec 2010 | 22:12:49 UTC

Short version. If probable composite it is mathematically composite

mdettweiler

Joined: 25 Apr 07
Posts: 271
ID: 7489
Credit: 7,380,245
RAC: 118

Message 30015 - Posted: 26 Dec 2010 | 23:09:13 UTC - in response to Message 30013.

Ah, that makes sense now. So if I'm understanding you correctly, based on Yves's conclusions, PFGW and LLR (and any other similar programs using FFTs) also produce what could technically be considered "probable composites" at times, but just aren't designed to properly analyze and report that uncertainty?

Gary Craig
Volunteer tester

Joined: 30 Dec 09
Posts: 3213
ID: 52890
Credit: 1,005,618,748
RAC: 3

Message 30048 - Posted: 28 Dec 2010 | 5:23:19 UTC

John, Iain...

Happy to give this a try. Sadly, no luck with the Mac version, either 32 or 64 bit. Library issues.

Here's my computer, according to the BOINC client:
Processor: 2 GenuineIntel Intel(R) Core(TM)2 Duo CPU E8335 @ 2.93GHz [x86 Family 6 Model 23 Stepping 10]
Processor features: FPU VME DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV PAT PSE36 CLFSH DS ACPI MMX FXSR SSE SSE2 SS HTT TM SSE3 MON DSCPL VMX EST TM2 SSSE3 CX16 TPR PDCM SSE4.1
OS: Mac OS X 10.6.5 (Darwin 10.5.0)
Memory: 4.00 GB physical, 559.03 GB virtual
Disk: 595.85 GB total, 558.79 GB free
Local time is UTC -8 hours
NVIDIA GPU 0: GeForce GT 120 (driver version unknown, CUDA version 3020, compute capability 1.1, 256MB, 80 GFLOPS peak)

I run BOINC PPS Sieve GPU tasks with no problem. According to the CUDA system prefs panel, my drivers are up to date (CUDA driver 3.2.17, GPU driver 1.6.24.17 (256.00.15f04)). For this test, I shut BOINC down completely. Here's a session log:

gary% pwd /Users/gary/CUDAGeneferMacIntel.0.95 gary% setenv DYLD_LIBRARY_PATH "/library/Application Support/BOINC Data/projects/www.primegrid.com":/usr/local/cuda/lib gary% ls -l total 288 -rwxr-xr-x@ 1 gary gary 72228 Dec 23 04:58 CUDAGeneferMacIntel32* -rwxr-xr-x@ 1 gary gary 72752 Dec 23 04:59 CUDAGeneferMacIntel64* gary% ./CUDAGeneferMacIntel32 -r GenefX64 2.2.0 (CUDA3.2/MacOS x86) Copyright (C) 2001-2003, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright (C) 2010, Shoichiro Yamada A program for finding large probable generalized Fermat primes. CUDAGenefer32.cu(175) : cudaSafeCall() Runtime API error : invalid argument. gary% g 230234^8192+1... gary% gary% gary% ./CUDAGeneferMacIntel64 -r dyld: Library not loaded: @rpath/libcufft.dylib Referenced from: /Users/gary/CUDAGeneferMacIntel.0.95/./CUDAGeneferMacIntel64 Reason: no suitable image found. Did find: /usr/local/cuda/lib/libcufft.dylib: mach-o, but wrong architecture Trace/BPT trap gary%

I'm not sure where the weird "g 230234"... text came from, echoed on the command line, when I tried to run the xxx32 version. I didn't type it. Anyway, when I tried to run the 64 bit version, it wasn't happy with the version of libcufft that it found. I was unable to find any other version of that library on my machine (else I would have changed DYLD_LIBRARY_PATH to include it first), and a quick google search didn't show anything promising.

If I reverse the order of the two DYLD_LIBRARY_PATH directories such that /usr/local/cuda/lib is first, the output of the 32 bit version is a little different:

gary% ./CUDAGeneferMacIntel32 -r GenefX64 2.2.0 (CUDA3.2/MacOS x86) Copyright (C) 2001-2003, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke Copyright (C) 2010, Shoichiro Yamada A program for finding large probable generalized Fermat primes. CUDAGenefer32.cu(193) : cufftSafeCall() CUFFT error. gary% g 230234^8192+1... gary%

The attempt at running the 64 bit version was the same as the first time (bad library).

I'm ready and willing to try attempt #2. Let me know if I've done anything obviously wrong.

--Gary

Gary Craig
Volunteer tester

Joined: 30 Dec 09
Posts: 3213
ID: 52890
Credit: 1,005,618,748
RAC: 3

Message 30049 - Posted: 28 Dec 2010 | 5:54:40 UTC - in response to Message 30048.

Ugh. I just re-read earlier posts in this thread, and #28900 says that one needs a double precision GPU to run this. If that's still the case, then sadly, I'm "down for the count" on this thread (compute capability = 1.1).

--Gary

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30050 - Posted: 28 Dec 2010 | 6:02:02 UTC

I had almost exactly the same problem, only I tried just the 64-bit version.

OSX 10.6.5 (Darwin 10.5.0)
NViDIA GeForce GT 330M 256MB

dyld: Library not loaded: @rpath/libcufft.dylib Referenced from: /Users/jimmy/Desktop/pgPSA/CUDAGeneferMacIntel.0.95/./CUDAGeneferMacIntel64 Reason: no suitable image found. Did find: /usr/local/cuda/lib/libcufft.dylib: mach-o, but wrong architecture

newb question: I remember there being a reason but don't know what it was - why is the path /./CUDAGenef...; I've never been good at compiling things from source. Terminals and command lines always seem to get the better of me. See username.

Note I have yet to try for real on my 32-bit XP machine. It's a 3.0GHz Pentium 4 but also runs an overclocked GTX 460 with 1GB of RAM (as much as the system...that can't be good can it?)

Did try to compile it on XP before but I think it failed because of something I did. Not quite sure what but I messed up my network settings and that affected everything else somehow. There's a loose capacitor somewhere, I just know it.

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30051 - Posted: 28 Dec 2010 | 6:06:24 UTC

Ach of course we need DP. I think that's the first thing that msft told us. Oopsies, we both forgot. Forgot that Apple doesn't work with NViDIA like good business partners. >_>

So now I've gotta try again on the GTX 460 machine of mine. Tomorrow. Currently a bit tired and to be honest I've been reading about number theory all day. Also the GPU is busy ;)

So what Macs do have DP cards? And don't say Mac Pro unless we're also porting to OpenCL with APP. Might not be a bad idea (and is the logical next step) but leveraging CUDA is the first thing.

Anyone got a Mac Pro with a Quadro FX...? I can't honestly think of a system that natively runs both Mac OS X and an NViDIA DP or Fermi card.
____________

Iain Bethune
Honorary cruncher

Joined: 28 Jan 09
Posts: 1588
ID: 34775
Credit: 194,056,043
RAC: 0

Message 30101 - Posted: 29 Dec 2010 | 20:51:31 UTC - in response to Message 30051.

Just checked the current options available through the Apple store, and you are correct - there is no DP-capable card available - the MacBook Pros have Geforce 320M or GT 330M, and the Mac Pros and iMacs all have ATI only.

In any case, for future needs, I've recompiled the code to run on any machine running OS 10.5 or later, CUDA 3.2+

Cheers

- Iain

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30108 - Posted: 30 Dec 2010 | 3:09:13 UTC - in response to Message 30101.

Gary, NullCoding, Iain
Thank you Lots of work.

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30128 - Posted: 30 Dec 2010 | 19:27:08 UTC

January's PRPNet Challenge Project

Taking the lead from rogue's November post about the anniversary of Pierre de Fermat's death, we decided to make January a focus on the PSA's Generalized Fermat Prime Search. The project is offered through PrimeGrid's PRPNet. NOTE: PrimeGrid's BOINC Challenge series doesn't start until 3 February 2011.

The original Generalized Fermat Prime Search by Yves Gallot was very active from 2001-2003. Now a decade later from its debut, it's about time to revitalize the search. Thanks in large part to David Underbakke, Mark Rodenkirch, and Shoichiro Yamada, the GFN prime search is ready to make a comeback.

The month goal will be to find PrimeGrid's first GFN prime. We have 5 deeply sieved files and have yet to find a single GFN prime. The advanced goal will be to find a GFN prime for each of the following N's: 32768, 65536, 262144, 524288.

N=65536; 20 known primes
N=131072 8 known primes
N=262144 1 known prime
N=524288 0 known primes

With the breakthrough of porting Genefer 2.2.0 to the GPU (CUDA), the search has become much more attractive. Testing times for the higher N's have been drastically reduced. NOTE: GPU app requires double precision support.

There will be a PRPNet Challenge in the middle of the month...sometime around the 12th for a few days.

The GFN Prime Search is open to all platforms.

• CUDAGenefer - used on Windows and Linux64, CUDA 3.2+
• CUDAGeneferMacIntel - used on MacIntel running OS 10.5 or later, CUDA 3.2+
• GenefX64 - used on Windows 64
• Genefer - used on Linux 32&64, Windows 32&64, MacIntel, MacPPC
• Genefer80 - used on Windows 32&64

However, the CUDAGenefer and GenefX64 applications offer the most efficient testing.

More Details to come.
____________

Iain Bethune
Honorary cruncher

Joined: 28 Jan 09
Posts: 1588
ID: 34775
Credit: 194,056,043
RAC: 0

Message 30148 - Posted: 31 Dec 2010 | 7:52:26 UTC - in response to Message 30128.

If someone could point me to the source for genefx64 and genefer80 I can probably manage mac ports in time for the challenge.

Cheers
-Iain

Steve_Martin@SETI.USA

Joined: 15 Sep 05
Posts: 160
ID: 776
Credit: 58,532,603
RAC: 0

Message 30150 - Posted: 31 Dec 2010 | 8:03:47 UTC - in response to Message 30148.

If someone could point me to the source for genefx64 and genefer80 I can probably manage mac ports in time for the challenge.

Cheers
-Iain

There is a url in the Read_me file, try that

Steve_Martin@SETI.USA

Joined: 15 Sep 05
Posts: 160
ID: 776
Credit: 58,532,603
RAC: 0

Message 30183 - Posted: 31 Dec 2010 | 20:24:25 UTC - in response to Message 30150.

If someone could point me to the source for genefx64 and genefer80 I can probably manage mac ports in time for the challenge.

Cheers
-Iain

There is a url in the Read_me file, try that

also try the source folder in
http://pgllr.mine.nu/PRPNet/

Iain Bethune
Honorary cruncher

Joined: 28 Jan 09
Posts: 1588
ID: 34775
Credit: 194,056,043
RAC: 0

Message 30199 - Posted: 1 Jan 2011 | 7:20:44 UTC - in response to Message 30183.

Cheers got the source now.

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30204 - Posted: 1 Jan 2011 | 13:34:47 UTC - in response to Message 30128.

• CUDAGenefer - used on Windows and Linux64, CUDA 3.2+
• CUDAGeneferMacIntel - used on MacIntel running OS 10.5 or later, CUDA 3.2+

Please add " GPU support requires double precision GPU processors. " or something good quote.

rroonnaalldd
Volunteer developer
Volunteer tester

Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0

Message 30233 - Posted: 2 Jan 2011 | 16:12:27 UTC

A new version of "cudagenefer" by Ken is ready for take off.
____________
Best wishes. Knowledge is power. by jjwhalen

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30237 - Posted: 2 Jan 2011 | 17:42:49 UTC - in response to Message 30233.

A new version of "cudagenefer" by Ken is ready for take off.

Is there any changelog?

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30238 - Posted: 2 Jan 2011 | 17:45:04 UTC - in response to Message 30237.

I just changed the initial text printed to work with prpnet. No other changes.
____________

Wabbit98

Joined: 5 Dec 05
Posts: 46
ID: 1662
Credit: 6,756,776
RAC: 337

Message 30239 - Posted: 2 Jan 2011 | 17:53:11 UTC

To John or anyone else in charge,

Just curious where the server and user stats are kept for ports 11002 another GFN port. Looking at this page http://www.primegrid.com/forum_thread.php?id=1215 it does not show that port or the other GFN port.

Ni
____________

Steve_Martin@SETI.USA

Joined: 15 Sep 05
Posts: 160
ID: 776
Credit: 58,532,603
RAC: 0

Message 30240 - Posted: 2 Jan 2011 | 18:20:23 UTC - in response to Message 30239.

To John or anyone else in charge,

Just curious where the server and user stats are kept for ports 11002 another GFN port. Looking at this page http://www.primegrid.com/forum_thread.php?id=1215 it does not show that port or the other GFN port.

Ni

General Stats: http://vcn94.homelinux.org/PRPNet/
GFN262144 port 11002 User Stats:http://vcn94.homelinux.org/PRPNet/alluser_stats.php?proj=GFN262144

port:11002
server stats:http://prime2u.com:11002/server_stats.html
user stats:http://prime2u.com:11002/user_stats.html
pending tests:http://prime2u.com:11002/pending_tests.html

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30241 - Posted: 2 Jan 2011 | 18:23:24 UTC

Note that the newest release of CUDAGenefer only works with PRPNet 4.1.0 clients and 4.1.0 servers. The difference between this release and the previous release is that the previous called itself "genefx64" in code (as seen when using the -v option). The new one calls itself "genefercuda".

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30242 - Posted: 2 Jan 2011 | 19:30:25 UTC - in response to Message 30241.

Note that the newest release of CUDAGenefer only works with PRPNet 4.1.0 clients and 4.1.0 servers. The difference between this release and the previous release is that the previous called itself "genefx64" in code (as seen when using the -v option). The new one calls itself "genefercuda".

Please take note of what rogue says above. If you are upgrading to the new GeneferCUDA, you won't be able to use it with the existing PRPNet servers. You must continue to use the previous version. We are currently testing a newer PRPClient & server which will accept the updated GeneferCUDA.
____________

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30254 - Posted: 3 Jan 2011 | 0:07:08 UTC

Since CUDAGenefer has become GeneferCUDA, I changed the name of my build file too. It's now at GeneferCUDA.zip.
____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30271 - Posted: 3 Jan 2011 | 14:25:19 UTC

PRPNet Update for GeneferCUDA

PRPNet 4.1.0 has been officially released and now includes proper support for GeneferCUDA. Other additions that are of interest to the upcoming GFN Challenge are Team stats.

We'll do our best to update client packages and servers as soon as possible.
____________

TroubledBunny
Volunteer tester

Joined: 4 Jun 07
Posts: 607
ID: 8931
Credit: 469,237,688
RAC: 13,339

Message 30302 - Posted: 4 Jan 2011 | 14:20:58 UTC

Could someone confirm which port(s) the challenge is on please?

server=GFN2:0:2:prpnet.primegrid.com:12003
server=GFN:0:2:prpnet.primegrid.com:12005

For now, this can be tested on the following GFN ports:

//server=GFN65536:0:1:prpnet.primegrid.com:12003
//server=GFN262144:0:1:prime2u.com:11002
//server=GFN524288:0:1:prime2u.com:11001

Thanks,
Pete
____________

35 x 2^3587843+1 is prime!

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30306 - Posted: 4 Jan 2011 | 15:39:09 UTC

Not to confirm but...

It would not make much sense running GPU on lower exponent. For example 65635 is about 4x faster on GTX480 comparing to CPU ->quad-core CPU does about the same job as high-end GPU.

I was looking for fast-running through low exponents on GPU (spent a lot of time sieving and PRP testing on low Ns on GFN) but GPUs are good for high exponents.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30310 - Posted: 4 Jan 2011 | 17:53:57 UTC

John, I think we can insert those b<100k for N=262144 which I haven't tested off-line.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

toms83

Joined: 17 Dec 09
Posts: 238
ID: 51975
Credit: 433,796,019
RAC: 0

Message 30311 - Posted: 4 Jan 2011 | 18:06:58 UTC

____________
Polish National Team

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30322 - Posted: 4 Jan 2011 | 19:40:34 UTC - in response to Message 30311.

Not sure if it's ready for public. If I understood it well, server side needs to be also updated in order to support genefer_CUDA.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Carlos Pinho

Joined: 11 Aug 05
Posts: 87
ID: 208
Credit: 105,796
RAC: 0

Message 30324 - Posted: 4 Jan 2011 | 19:55:36 UTC
Michael Millerick
Volunteer tester

Joined: 4 Feb 09
Posts: 714
ID: 35074
Credit: 177,131,550
RAC: 30,522

Message 30327 - Posted: 4 Jan 2011 | 20:59:00 UTC - in response to Message 30324.

It may just be me, but the contents of that package are completely different from what usually comes in the PRPnet packages. I think I will just wait until the links here get updated. I am in no hurry to update to the new version...at least not until I finally manage to get my PPS Sieve badge to silver.
____________

KWSN Raw Data

Joined: 8 Jan 10
Posts: 5
ID: 53314
Credit: 668,937,172
RAC: 0

Message 30344 - Posted: 4 Jan 2011 | 22:41:40 UTC

I may have missed this but how do you use multi GPUs

Thanks

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30347 - Posted: 4 Jan 2011 | 23:14:47 UTC - in response to Message 30344.

I may have missed this but how do you use multi GPUs

Thanks

That is a question for both genefercuda, but also PRPNet. I can't answer until someone can tell me if any CUDA app can specify the specific GPU to run on (akin to processor affinity). If so, then PRPNet would need to wait until genefercuda supports a command line option to specify affinity. The change to the PRPNet client is minimal as it already supports that feature for LLR.

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30348 - Posted: 4 Jan 2011 | 23:25:47 UTC - in response to Message 30347.

My PSieve-CUDA apps can specify the GPU to run on. So it is possible from that standpoint. If "--device" followed by a number is the mechanism used to specify the GPU, that makes it easier to port to BOINC later on.
____________

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30351 - Posted: 5 Jan 2011 | 0:09:57 UTC

Now It is always DeviceID=0;

cudaSetDevice(0);

TPSieve souce code:

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30352 - Posted: 5 Jan 2011 | 0:26:34 UTC - in response to Message 30351.

My proposal :

GeneferCUDA [-d N]
-d N set GPUID=N(default 0)

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30356 - Posted: 5 Jan 2011 | 1:20:00 UTC

PRPNet with GeneferCUDA support

We are now ready to start testing. A test server of 4.1.0 has been set up and open to Windows 32/64 as well as Linux 64 clients. Now we just need testers. Please PM me if you are able to help (prior PRPNet experience is recommended). ;)

The current work is preferably for GPU users and is set up for GFN prime search N=262144.

The new PRPNet has the following update:

Version 4.1.0: December 2010 - New Release prpclient: Added teamid= option to the prpclient.ini file. This allows the client to do tests on behalf of a team. Added support for "genefer_cuda" version of genefer. prpserver: Added support for "genefer_cuda" version of genefer. Added support for the team reported by a client. A team ID is captured for each test, thus it is possible for clients (and users) to have some tests for one team and other tests for another team. Added team_stats.html and team_primes.html. Added "Team" column to user_stats.html, user_primes.html, and pending_tests.html. Fixed an issue with inserting GFNs as the value for column c was not set.

Note: Based on current thread discussion, we'll probably have new builds soon to handle multiple GPU's.
____________

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30367 - Posted: 5 Jan 2011 | 5:56:56 UTC - in response to Message 30356.

Support device number.
http://rapidshare.com/files/440865904/GeneferCUDA.0.97.tar.gz

\$ ./GeneferCUDA
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Usage: GeneferCUDA [-d N] -b run bench
GeneferCUDA [-d N] -t run test
GeneferCUDA [-d N] -r run residue test
GeneferCUDA [-d N] -q test quick expression
GeneferCUDA [-d N] <filename> test <filename>
-d N set device number=N(default 0)
GeneferCUDA use interactive mode

\$ ./GeneferCUDA -d 1 -b
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

device_number >= device_count ... exiting
\$ ./GeneferCUDA -d 0 -b
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Generalized Fermat Number Bench
2009574^8192+1 Time: 428 us/mul. Err: 3.82e-01 51636 digits
1632282^16384+1 Time: 452 us/mul. Err: 2.53e-01 101791 digits
1325824^32768+1 Time: 490 us/mul. Err: 2.19e-01 200622 digits
1076904^65536+1 Time: 638 us/mul. Err: 1.72e-01 395325 digits
874718^131072+1 Time: 814 us/mul. Err: 3.47e-01 778813 digits
710492^262144+1 Time: 1.15 ms/mul. Err: 4.21e-01 1533952 digits
577098^524288+1 Time: 2.1 ms/mul. Err: 2.01e-01 3020555 digits
468750^1048576+1 Time: 3.96 ms/mul. Err: 1.64e-01 5946413 digits
380742^2097152+1 Time: 8.12 ms/mul. Err: 3.63e-01 11703432 digits

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30369 - Posted: 5 Jan 2011 | 6:16:09 UTC - in response to Message 30367.

Support device number.
http://rapidshare.com/files/440865904/GeneferCUDA.0.97.tar.gz

Windows build updated.

____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30412 - Posted: 5 Jan 2011 | 17:50:38 UTC

I've posted PRPNet 4.1.1 over at my website, http://home.roadrunner.com/~mrodenkirch/prpnet_4.1.1.zip. Since I can't run genefercuda, I am unable to test. If someone can build and test, please let me know if it works as desired.

I also need someone to test against one of the other genefer versions to make sure that I didn't break it for them.

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30423 - Posted: 5 Jan 2011 | 20:31:01 UTC - in response to Message 30412.

On Linux64:

\$ ./prpclient
[2011-01-05 14:18:43 EST] PRPNet Client application v4.1.1 started
[2011-01-05 14:18:44 EST] server=GFN262144TEST: Getting work from server prime2u.com at port 13000
[2011-01-05 14:18:46 EST] server=GFN262144TEST: PRPNet server is version 4.1.1
Send a message to the client in this file such as an
explanation of the project or remind them the current release.
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Start test of file 'work_server=GFN262144TEST.in' - 14:18:46
4990^262144+1 is composite. (RES=16e5fb9041b8f6b0)
(969435 digits) (err = 0.0000) (time = 1:03:24) 15:22:10
[2011-01-05 15:22:10 EST] server=GFN262144TEST: 1*4990^262144+1 is not prime. Residue 16e5fb9041b8f6b0
[2011-01-05 15:22:10 EST] Total Time: 1:03:27 Total Tests: 1 Total PRPs Found: 0
[2011-01-05 15:22:12 EST] server=GFN262144TEST: Returning work to server prime2u.com at port 13000
[2011-01-05 15:22:12 EST] server=GFN262144TEST: INFO: Test for 4990^262144+1 was accepted
[2011-01-05 15:22:13 EST] server=GFN262144TEST: INFO: All 1 test results were accepted
[2011-01-05 15:22:14 EST] server=GFN262144TEST: Getting work from server prime2u.com at port 13000
[2011-01-05 15:22:16 EST] server=GFN262144TEST: PRPNet server is version 4.1.1
Send a message to the client in this file such as an
explanation of the project or remind them the current release.
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)

affinity=1

\$ ./prpclient
[2011-01-05 15:23:18 EST] PRPNet Client application v4.1.1 started
[2011-01-05 15:23:21 EST] server=GFN262144TEST: Getting work from server prime2u.com at port 13000
[2011-01-05 15:23:24 EST] server=GFN262144TEST: PRPNet server is version 4.1.1
Send a message to the client in this file such as an
explanation of the project or remind them the current release.
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

device_number >= device_count ... exiting
[2011-01-05 15:23:27 EST] server=GFN262144TESTwork_server=GFN262144TEST.in: Could not open file [genefer.log] for reading. Assuming user stopped with ^C
What should the client do with current workunits?
1 = Report completed and abort the rest
2 = Report completed tests to server
3 = Nothing
Choose option:

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30426 - Posted: 5 Jan 2011 | 21:01:20 UTC - in response to Message 30423.

If you only have one GPU, do not specify any value for the affinity.

Note that I just made a quick change to the source because if affinity is in the ini file, but not specified, it would set the device to 0, which might trigger an error in genefer. I don't think it would be an issue, but one never knows...

Iain Bethune
Honorary cruncher

Joined: 28 Jan 09
Posts: 1588
ID: 34775
Credit: 194,056,043
RAC: 0

Message 30428 - Posted: 5 Jan 2011 | 21:35:43 UTC - in response to Message 30426.

Just found out the Quadro 4000 supports the Mac Pro, as an aftermarket option, so it is possible to run this code after all :) At least if, someone has forked out for the card!

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30431 - Posted: 5 Jan 2011 | 21:54:06 UTC - in response to Message 30412.

To update to 4.1.1, all that is needed is the prpclient. You can download them here:

These builds should only be used by those participating in the testing. Once testing is over, full packages will be available for everyone. :)

____________

KWSN Raw Data

Joined: 8 Jan 10
Posts: 5
ID: 53314
Credit: 668,937,172
RAC: 0

Message 30434 - Posted: 5 Jan 2011 | 22:11:06 UTC

I am testing on a box that has 2 gtx 460

they both fire up for a few minutes but then gpu 2 fails gpu 1 continues on no issue

here is the text

Microsoft Windows [Version 6.1.7600]
[2006-01-28 01:41:12 CST] PRPNet Client application v4.1.1 started
[2006-01-28 01:41:12 CST] User name 53314 at email address is GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in' - 01:41:12
Testing 108760^262144+1... 4313088 steps to go
maxErr exceeded for 108760^262144+1, 0.5000 > 0.4500
Genefer80 2.2.0 (x86 - 32-bit - 80-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Cannot open '-d'
[2006-01-28 01:43:45 CST] GFN262144: Could not open file [genefer.log] for readi
ng. Assuming user stopped with ^C
[2006-01-28 01:43:45 CST] Accepted force quit. Waiting to close sockets before
exiting
[2006-01-28 01:43:45 CST] Total Time: 0:02:34 Total Tests: 0 Total PRPs Found
: 0
[2006-01-28 01:43:45 CST] Client shutdown complete

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30439 - Posted: 5 Jan 2011 | 22:38:08 UTC - in response to Message 30434.

I am testing on a box that has 2 gtx 460

they both fire up for a few minutes but then gpu 2 fails gpu 1 continues on no issue

\$ ./GeneferCUDA.exe -d 1 -q 108760^262144+1

KWSN Raw Data

Joined: 8 Jan 10
Posts: 5
ID: 53314
Credit: 668,937,172
RAC: 0

Message 30441 - Posted: 5 Jan 2011 | 22:46:50 UTC - in response to Message 30439.

Expression must be in b^m+1 form

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30442 - Posted: 5 Jan 2011 | 23:01:54 UTC - in response to Message 30441.

Expression must be in b^m+1 form

\$ GeneferCUDA.exe -d 1 work_GFN262144.in
Thank you Lots of work,

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30447 - Posted: 5 Jan 2011 | 23:28:48 UTC - in response to Message 30434.

I am testing on a box that has 2 gtx 460

they both fire up for a few minutes but then gpu 2 fails gpu 1 continues on no issue

here is the text

Microsoft Windows [Version 6.1.7600]
[2006-01-28 01:41:12 CST] PRPNet Client application v4.1.1 started
[2006-01-28 01:41:12 CST] User name 53314 at email address is GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in' - 01:41:12
Testing 108760^262144+1... 4313088 steps to go
maxErr exceeded for 108760^262144+1, 0.5000 > 0.4500
Genefer80 2.2.0 (x86 - 32-bit - 80-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Cannot open '-d'
[2006-01-28 01:43:45 CST] GFN262144: Could not open file [genefer.log] for readi
ng. Assuming user stopped with ^C
[2006-01-28 01:43:45 CST] Accepted force quit. Waiting to close sockets before
exiting
[2006-01-28 01:43:45 CST] Total Time: 0:02:34 Total Tests: 0 Total PRPs Found
: 0
[2006-01-28 01:43:45 CST] Client shutdown complete

I'll fix that tomorrow. Continue using the 4.1.0 client until then.

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30499 - Posted: 6 Jan 2011 | 15:47:21 UTC - in response to Message 30447.

I uploaded a new 4.1.1. I split affinity in the prpclient.ini into cpuaffinity and gpuaffinity and I modified it to not use the -d option with the other versions of genefer.

If someone could give it a spin, I would appreciate it.

KWSN Raw Data

Joined: 8 Jan 10
Posts: 5
ID: 53314
Credit: 668,937,172
RAC: 0

Message 30500 - Posted: 6 Jan 2011 | 16:04:25 UTC

I just tryed it and it is not executing on device 1 or gpu 2

Myabe I did not configure it correctly I put the following ni my ini

// This sets the CPU affinity for LLR on multi-CPU machines. It defaults to
// -1, which means that LLR can run on an CPU.
cpuaffinity=

// This sets the GPU affinity for CUDA apps on multi-GPU machines. It defaults to
// -1, which means that the CUDA app can run on an GPU.
gpuaffinity=1

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30501 - Posted: 6 Jan 2011 | 16:27:32 UTC

I checked on port GFN32768:100:1:prpnet.primegrid.com:12005

It errored out on GeneferCuda, GeneFX64 and continued using Genefer80 as expected.
CPU affinity still does work.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30504 - Posted: 6 Jan 2011 | 19:04:46 UTC

Hi, I've gone from developer to user. I'm trying to run GeneferCUDA 0.97 on Linux64. I upgraded my Cuda toolkit from 2.3 to 3.2, and I linked libcufft.so.3 to the directory where I'm trying to run GeneferCUDA, which seemed to help. But when I try to run a test (option 2) on device 0 (or nothing), I get:

GeneferCUDA.cu(198) : cudaSafeCall() Runtime API error : unspecified driver error.

My driver is version 256.53, and I really don't want to upgrade because 260.17.* seems quite buggy to me. What is the minimum driver version needed to run GeneferCUDA?

Thanks!
____________

KWSN Raw Data

Joined: 8 Jan 10
Posts: 5
ID: 53314
Credit: 668,937,172
RAC: 0

Message 30505 - Posted: 6 Jan 2011 | 19:18:57 UTC

Is there a link to the GeneferCUDA source

Thanks

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30508 - Posted: 6 Jan 2011 | 19:26:08 UTC - in response to Message 30505.

Is there a link to the GeneferCUDA source

Thanks

Support device number.
http://rapidshare.com/files/440865904/GeneferCUDA.0.97.tar.gz

Windows build updated.

Both of these links have the source included.
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30511 - Posted: 6 Jan 2011 | 20:23:02 UTC - in response to Message 30500.

I just tryed it and it is not executing on device 1 or gpu 2

Myabe I did not configure it correctly I put the following ni my ini

// This sets the CPU affinity for LLR on multi-CPU machines. It defaults to
// -1, which means that LLR can run on an CPU.
cpuaffinity=

// This sets the GPU affinity for CUDA apps on multi-GPU machines. It defaults to
// -1, which means that the CUDA app can run on an GPU.
gpuaffinity=1

This is now fixed. I have updated the zip file.

This version also fixes a bug in computing the decimal length of primorials.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30521 - Posted: 6 Jan 2011 | 21:50:14 UTC - in response to Message 30501.

I checked on port GFN32768:100:1:prpnet.primegrid.com:12005

It errored out on GeneferCuda, GeneFX64 and continued using Genefer80 as expected.
CPU affinity still does work.

What I wanted to state - CPU affinity still does NOT work.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30522 - Posted: 6 Jan 2011 | 22:21:53 UTC

Tried to run with GeneferCUDA.exe (on Win XP) and got alert "..failed to start... cudart32_32-16.dll was not found. ..."
Where can I get this?
____________
Murphy (AtP)

Ken_g6
Volunteer developer

Joined: 4 Jul 06
Posts: 900
ID: 3110
Credit: 176,356,682
RAC: 117,754

Message 30523 - Posted: 6 Jan 2011 | 22:27:34 UTC - in response to Message 30522.

Dlls and .so files should be in the CUDA Toolkit.
____________

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30524 - Posted: 6 Jan 2011 | 22:29:07 UTC - in response to Message 30522.

Tried to run with GeneferCUDA.exe (on Win XP) and got alert "..failed to start... cudart32_32-16.dll was not found. ..."
Where can I get this?

It's a part of CUDA Toolkit, IIRC.
I've extracted DLL needed to run GeneferCUDA - http://pgllr.mine.nu/PRPNet/cudart.rar
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30525 - Posted: 6 Jan 2011 | 22:36:36 UTC - in response to Message 30521.

I checked on port GFN32768:100:1:prpnet.primegrid.com:12005

It errored out on GeneferCuda, GeneFX64 and continued using Genefer80 as expected.
CPU affinity still does work.

What I wanted to state - CPU affinity still does NOT work.

Are you talking about cpuaffinity= or gpuaffinity=?

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30528 - Posted: 6 Jan 2011 | 23:24:23 UTC - in response to Message 30525.

Are you talking about cpuaffinity= or gpuaffinity=?

CPU affinity.

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30530 - Posted: 6 Jan 2011 | 23:51:54 UTC - in response to Message 30528.

Are you talking about cpuaffinity= or gpuaffinity=?

CPU affinity.

CPU affinity is for LLR only. GPU affinity is for genefercuda only. None of the other versions of genefer support the feature.

If I'm not understanding the problem, then I need to understand the specific problem you are encountering and I need to know that it is the latest one that I posted today.

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30541 - Posted: 7 Jan 2011 | 7:17:13 UTC

PRPClient 4.1.1 packages have been updated. Currently only available for Linux64 and Windows.

• prpclient v4.1.1 beta & documentation
• master_prpclient.ini

____________

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30542 - Posted: 7 Jan 2011 | 9:21:08 UTC - in response to Message 30530.

CPU affinity is for LLR only. GPU affinity is for genefercuda only. None of the other versions of genefer support the feature.

If I'm not understanding the problem, then I need to understand the specific problem you are encountering and I need to know that it is the latest one that I posted today.

Thanks for explanation.
It was CPU affinity for Genefer in latest PRPNet. If it's not supported, I understand why I can't see it.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Joined: 6 Dec 10
Posts: 50
ID: 75840
Credit: 3,315,642
RAC: 0

Message 30543 - Posted: 7 Jan 2011 | 9:48:20 UTC - in response to Message 30541.

PRPClient 4.1.1 packages have been updated. Currently only available for Linux64 and Windows.

Linux64:
[2011-01-07 04:43:36 EST] GFN262144TEST: Getting work from server prime2u.com at port 13000
[2011-01-07 04:43:38 EST] GFN262144TEST: INFO: No available candidates are left on this server.
Send a message to the client in this file such as an
explanation of the project or remind them the current release.
[2011-01-07 04:43:39 EST] GFN262144TEST: Getting work from server prime2u.com at port 13000
[2011-01-07 04:43:40 EST] GFN262144TEST: INFO: No available candidates are left on this server.
Send a message to the client in this file such as an
explanation of the project or remind them the current release.

TroubledBunny
Volunteer tester

Joined: 4 Jun 07
Posts: 607
ID: 8931
Credit: 469,237,688
RAC: 13,339

Message 30545 - Posted: 7 Jan 2011 | 10:34:47 UTC

W7 64bit GTX 570 stock speeds

[2011-01-07 09:39:09 GST] PRPNet Client application v4.1.1 started [2011-01-07 09:39:09 GST] User name TroubledBunny at email address is XXX@sky.com [2011-01-07 09:39:09 GST] GFN262144TEST: Getting work from server prime2u.com at port 13000 [2011-01-07 09:39:11 GST] GFN262144TEST: PRPNet server is version 4.1.1 Send a message to the client in this file such as an explanation of the project or remind them the current release. GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0 Copyright (C) 2001-2003, Yves Gallot (v1.3) Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0) Copyright (C) 2010, Shoichiro Yamada (CUDA) A program for finding large probable generalized Fermat primes. Start test of file 'work_GFN262144TEST.in' - 09:39:11 5324^262144+1 is composite. (RES=ad3013c6d9ad6d27) (976811 digits) (err = 0.0000) (time = 0:36:34) 10:15:45 [2011-01-07 10:15:45 GST] GFN262144TEST: 1*5324^262144+1 is not prime. Residue ad3013c6d9ad6d27 [2011-01-07 10:15:45 GST] Total Time: 0:36:42 Total Tests: 1 Total PRPs Found: 0 [2011-01-07 10:15:46 GST] GFN262144TEST: Returning work to server prime2u.com at port 13000 [2011-01-07 10:15:46 GST] GFN262144TEST: INFO: Test for 5324^262144+1 was accepted [2011-01-07 10:15:46 GST] GFN262144TEST: INFO: All 1 test results were accepted

____________

35 x 2^3587843+1 is prime!

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30546 - Posted: 7 Jan 2011 | 12:09:27 UTC

GeneferCuda, GeneFX64 and skipped to Genefer80 (missed Genefer).
Are b limits hard-coded into client? I canâ€™t remember right now...

[2011-01-07 12:39:43 SE(Å”] PRPNet Client application v4.1.1 started [2011-01-07 12:39:43 SE(Å”] User name Honza at email address is xxx [2011-01-07 12:39:43 SE(Å”] GFN32768: Getting work from server prpnet.primegrid.com at port 12005 [2011-01-07 12:39:45 SE(Å”] GFN32768: PRPNet server is version 3.3.1 PRPNet Server version 3.1.5 GFN N=32768 1 GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0 Copyright (C) 2001-2003, Yves Gallot (v1.3) Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0) Copyright (C) 2010, Shoichiro Yamada (CUDA) A program for finding large probable generalized Fermat primes. Start test of file 'work_GFN32768.in' - 12:39:45 Testing 3284106^32768+1... 708608 steps to go maxErr exceeded for 3284106^32768+1, 0.5000 > 0.4500 GenefX64 2.2.0 (x86 - 64-bit - SSE3) Copyright (C) 2001-2003, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke A program for finding large probable generalized Fermat primes. Start test of file 'work_GFN32768.in' - 12:39:49 Testing 3284106^32768+1... 708608 steps to go maxErr exceeded for 3284106^32768+1, 0.5000 > 0.4500 Genefer80 2.2.0 (x86 - 32-bit - 80-b X87) Copyright (C) 2001-2003, Yves Gallot Copyright 2009, Mark Rodenkirch, David Underbakke A program for finding large probable generalized Fermat primes. Start test of file 'work_GFN32768.in'. 3284106^32768+1 is a probable composite. (RES=865f757526a27776) (213530 digits) (err = 0.0015) (time = 0:27:27) 13:07:18 [2011-01-07 13:07:18 SE(Å”] GFN32768: 1*3284106^32768+1 is not prime. Residue 865f757526a27776 [2011-01-07 13:07:18 SE(Å”] Total Time: 0:27:35 Total Tests: 1 Total PRPs Found: 0 [2011-01-07 13:07:18 SE(Å”] GFN32768: Returning work to server prpnet.primegrid.com at port 12005 [2011-01-07 13:07:19 SE(Å”] GFN32768: INFO: Test for 3284106^32768+1 was accepted [2011-01-07 13:07:19 SE(Å”] GFN32768: INFO: All 1 test results were accepted

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30547 - Posted: 7 Jan 2011 | 13:29:04 UTC - in response to Message 30542.

CPU affinity is for LLR only. GPU affinity is for genefercuda only. None of the other versions of genefer support the feature.

If I'm not understanding the problem, then I need to understand the specific problem you are encountering and I need to know that it is the latest one that I posted today.

Thanks for explanation.
It was CPU affinity for Genefer in latest PRPNet. If it's not supported, I understand why I can't see it.

In 4.1.0 and in my first attempt at 4.1.1, affinity= was used for both. I split that into cpuaffinity= and gpuaffinity=, so you would have needed to update your ini to add those entries and set them.

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30548 - Posted: 7 Jan 2011 | 13:31:41 UTC - in response to Message 30546.

GeneferCuda, GeneFX64 and skipped to Genefer80 (missed Genefer).
Are b limits hard-coded into client? I canÂ§t remember right now...

Hmmm. I don't see anything amiss in the code. How is your ini file configured? I presume that genefer exists in the folder that the client is running in. What is the output when you try to run genefer?

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30560 - Posted: 7 Jan 2011 | 15:20:57 UTC

This is a part from prpclient.ini

geneferexe=GeneferCUDA.exe geneferexe=genefX64.exe geneferexe=genefer.exe geneferexe=genefer80.exe cpuaffinity=0 gpuaffinity=-1

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30567 - Posted: 7 Jan 2011 | 16:58:27 UTC - in response to Message 30560.

This is a part from prpclient.ini

geneferexe=GeneferCUDA.exe geneferexe=genefX64.exe geneferexe=genefer.exe geneferexe=genefer80.exe cpuaffinity=0 gpuaffinity=-1

That looks fine, but what is the output when you try to execute genefer.exe?

Is this on Windows or *nix? Are you able to debug?

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30570 - Posted: 7 Jan 2011 | 17:09:37 UTC - in response to Message 30567.

That looks fine, but what is the output when you try to execute genefer.exe?

Is this on Windows or *nix? Are you able to debug?

Running Windows 7 x64.
Genefer benchmark and tests are fine.

But this one gives error, wrong syntax?
Genefer -q 3284106^32768+1

Expression must be in b^m+1 form

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30578 - Posted: 7 Jan 2011 | 19:18:08 UTC - in response to Message 30570.

That looks fine, but what is the output when you try to execute genefer.exe?

Is this on Windows or *nix? Are you able to debug?

Running Windows 7 x64.
Genefer benchmark and tests are fine.

But this one gives error, wrong syntax?
Genefer -q 3284106^32768+1

Expression must be in b^m+1 form

Try
Genefer -q 3284106^^32768+1
or
Genefer -q "3284106^32768+1"

It shouldn't matter though because I found and fixed the problem. Please d/l it again.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30610 - Posted: 8 Jan 2011 | 8:51:21 UTC - in response to Message 30578.

Try
Genefer -q 3284106^^32768+1
or
Genefer -q "3284106^32768+1"

It shouldn't matter though because I found and fixed the problem. Please d/l it again.

This syntax works fine, thanks.

PRPnet client now checks with all Genefer apps.

EDIT: Upon aborting client, I choose Report and Abort the rest. It reported fine but afterwards display error "file not found" 5 times.

[2011-01-08 09:48:45 SE(Å”] GFN32768: Getting work from server prpnet.primegrid.com at port 12005
[2011-01-08 09:48:46 SE(Å”] GFN32768: PRPNet server is version 3.3.1
PRPNet Server version 3.1.5
GFN N=32768 1
GeneferCUDA 0.97 (CUDA3.2) based on Genefer v2.2.0
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, Mark Rodenkirch, David Underbakke (v2.2.0)
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN32768.in' - 09:48:46
Testing 3284138^32768+1... 708608 steps to go
maxErr exceeded for 3284138^32768+1, 0.5000 > 0.4500
GenefX64 2.2.0 (x86 - 64-bit - SSE3) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN32768.in' - 09:48:50
Testing 3284138^32768+1... 708608 steps to go
maxErr exceeded for 3284138^32768+1, 0.5000 > 0.4500
Genefer 2.2.0 (x86 - 32-bit - 64-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN32768.in'.
Testing 3284138^32768+1... 708608 steps to go
maxErr exceeded for 3284138^32768+1, 0.5000 > 0.4700
Genefer80 2.2.0 (x86 - 32-bit - 80-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN32768.in'.
Testing 3284138^32768+1... 700416 steps to go
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30621 - Posted: 8 Jan 2011 | 14:35:32 UTC - in response to Message 30610.

EDIT: Upon aborting client, I choose Report and Abort the rest. It reported fine but afterwards display error "file not found" 5 times.

This is okay. It is just the client trying to clean up temp files from the various apps. If the OS doesn't find the temp files, then it gives the message.

Honza
Volunteer moderator
Volunteer tester
Project scientist

Joined: 15 Aug 05
Posts: 1878
ID: 352
Credit: 3,000,663,968
RAC: 1,253,232

Message 30666 - Posted: 9 Jan 2011 | 9:20:48 UTC

Sieve depths as of 29 August 2010 are:

n = 32768; p=1000P
n = 65536; p=1500P
n = 131072; p=2000P
n = 262144; p=2510P
n = 524288; p=3070P

I haven't done any more sieving since then, ranges are as described above. Do we need to go deeper?
What about lower Ns primality testing? I have done all for N<=2048 up to 100M, N=4096 is up to 6M5.

____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30671 - Posted: 9 Jan 2011 | 11:53:37 UTC

I've just downloaded the 64 bit Linux PRPclient and started a few tests on my GTX 460 with the 256.53 drivers (CUDA 3.1) since the 260.19.29 drivers are up to 50% slower than the 256.53 ones when running CWPSieve (CUDA).

Successful tests:

230234^8192+1 is composite. (RES=5972a302c255e081)
(43927 digits) (err = 0.0024) (time = 0:00:58) 13:00:09
151902^16384+1 is composite. (RES=0a1a98e69b17f1e9)
(84895 digits) (err = 0.0017) (time = 0:02:01) 13:03:58

The program seems to work despite the drivers are cuda 3.1 ones. I'll probably build a CUDA 3.1 version of GeneferCUDA binary (Linux 64 bit) just to be on the safe side ;)
____________

Ian Smith

Joined: 11 Nov 10
Posts: 382
ID: 72983
Credit: 5,413,535
RAC: 0

Message 30672 - Posted: 9 Jan 2011 | 12:02:41 UTC

It seems whenever I terminate the client by pressing ctrl c it stop as normal but when I re start the client it starts the checking process from the beginning. It does not seem to re start the test from when I terminated it. Is this correct
PS in the file there is a .sav file but the prpclient say starting from the .in file

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30677 - Posted: 9 Jan 2011 | 13:55:15 UTC - in response to Message 30672.

It seems whenever I terminate the client by pressing ctrl c it stop as normal but when I re start the client it starts the checking process from the beginning. It does not seem to re start the test from when I terminated it. Is this correct
PS in the file there is a .sav file but the prpclient say starting from the .in file

Genefer creates a checkpoint file at certain iterations. I presume genfercuda is doing the same, but I haven't looked at the code. When it restarts, it should start from the checkpoint. That had been tested with the CPU versions of genefer. I cannot speak for genefercuda.

Ian Smith

Joined: 11 Nov 10
Posts: 382
ID: 72983
Credit: 5,413,535
RAC: 0

Message 30679 - Posted: 9 Jan 2011 | 14:06:15 UTC - in response to Message 30677.

It seems whenever I terminate the client by pressing ctrl c it stop as normal but when I re start the client it starts the checking process from the beginning. It does not seem to re start the test from when I terminated it. Is this correct
PS in the file there is a .sav file but the prpclient say starting from the .in file

Genefer creates a checkpoint file at certain iterations. I presume genfercuda is doing the same, but I haven't looked at the code. When it restarts, it should start from the checkpoint. That had been tested with the CPU versions of genefer. I cannot speak for genefercuda.

Hmm. You are correct the prog creats a genefer.ckpt file but when I restart the prpclient this file disappears and the test just starts fromthe beginning

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30684 - Posted: 9 Jan 2011 | 15:57:38 UTC

Checkpointing seems to work here, but I've seen messages like this after pressing CTRL+C

2011-01-09 16:34:38 CET] GFN262144TEST: No data in file [genefer.log]. Is genefer broken?

in the prpclient.log (debuglevel=2).
____________

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30703 - Posted: 9 Jan 2011 | 22:45:15 UTC - in response to Message 30679.

It seems whenever I terminate the client by pressing ctrl c it stop as normal but when I re start the client it starts the checking process from the beginning. It does not seem to re start the test from when I terminated it. Is this correct
PS in the file there is a .sav file but the prpclient say starting from the .in file

Genefer creates a checkpoint file at certain iterations. I presume genfercuda is doing the same, but I haven't looked at the code. When it restarts, it should start from the checkpoint. That had been tested with the CPU versions of genefer. I cannot speak for genefercuda.

Hmm. You are correct the prog creats a genefer.ckpt file but when I restart the prpclient this file disappears and the test just starts fromthe beginning

I have never successfully resumed from either of the two largest GFN (262144, 524288) checkpoints -- you apparently must let them run to completion in one shot!
____________
Murphy (AtP)

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30711 - Posted: 10 Jan 2011 | 0:06:45 UTC - in response to Message 30703.

Which version of genefer is not restarting from the checkpoint?

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30726 - Posted: 10 Jan 2011 | 4:50:12 UTC - in response to Message 30711.

Which version of genefer is not restarting from the checkpoint?

Genefer 2.2.0 (x86 - 32-bit - 64-b X87) Copyright 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in'.
Testing 102850^262144+1... 4358144 steps to go

== (wait for .ckpt file to update) control-C == then start:

Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in'.
Testing 102850^262144+1... 4364744 steps to go
____________
Murphy (AtP)

Wabbit98

Joined: 5 Dec 05
Posts: 46
ID: 1662
Credit: 6,756,776
RAC: 337

Message 30733 - Posted: 10 Jan 2011 | 7:33:06 UTC

Just a quick question. I have been working on a GFN number from the GFN524288 and my poor machine is slowly finishing it. Since it is such a big number that I would have quite a bit of time to finish it. But looking at the pending tests for that server.

I see someone else has it 6720^524288+1 is assigned to user: pschoefer.

Just to let everyone know this will be the only one I do for this server, and also it should be done by my calculation sometime late tomorrow night or real early the next morning. Hopefully if I turn it in I will still be able to get credit.

Ni
____________

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30736 - Posted: 10 Jan 2011 | 13:37:04 UTC - in response to Message 30726.

Which version of genefer is not restarting from the checkpoint?

Genefer 2.2.0 (x86 - 32-bit - 64-b X87) Copyright 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in'.
Testing 102850^262144+1... 4358144 steps to go

== (wait for .ckpt file to update) control-C == then start:

Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144.in'.
Testing 102850^262144+1... 4364744 steps to go

After the ^C, does the .ckpt file still exist? If so, then I would have to defer this to David. He build genefer for x86. He should be able to debug it.

Ian Smith

Joined: 11 Nov 10
Posts: 382
ID: 72983
Credit: 5,413,535
RAC: 0

Message 30741 - Posted: 10 Jan 2011 | 16:17:21 UTC - in response to Message 30736.

yes for me the .ckpt file still exists but when you restart the client it disappears and then it just starts from the beginning again. Just as described by mfbabb

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30743 - Posted: 10 Jan 2011 | 16:50:29 UTC

Ditto ...
Note that it is starting back up using the ".in" file, not the ".ckpt" file as expected.
???
The ".ckpt" file appears to be a binary file.
____________
Murphy (AtP)

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30765 - Posted: 10 Jan 2011 | 20:59:28 UTC

I updated the client (v4.1.1), so if someone could please d/l and try it out, I would appreciate it. It should no longer delete the checkpoint, but I haven't tested it.

What could happen before is that if there were multiple versions of genefer configured and the first one could not do the test due to a round-off error, it would delete the checkpoint, even if that checkpoint was written by a different version of genefer.

I will work with Shoichiro and David to modify genefer to write genefer.log under certain error conditions so that the PRPNet client can react accordingly. The worst thing that can happen now is that the client can shut down if the log file is not created. As long as you keep don't run into roundoff errors, you shouldn't be experiencing any problems.

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30770 - Posted: 10 Jan 2011 | 23:25:49 UTC - in response to Message 30765.

I updated the client (v4.1.1), so if someone could please d/l and try it out, I would appreciate it. It should no longer delete the checkpoint, but I haven't tested it.

What could happen before is that if there were multiple versions of genefer configured and the first one could not do the test due to a round-off error, it would delete the checkpoint, even if that checkpoint was written by a different version of genefer.

I will work with Shoichiro and David to modify genefer to write genefer.log under certain error conditions so that the PRPNet client can react accordingly. The worst thing that can happen now is that the client can shut down if the log file is not created. As long as you keep don't run into roundoff errors, you shouldn't be experiencing any problems.

Is this what is updated:
http://uwin.mine.nu/PRPNet_files/prpclient-4.1.1cbeta-windows.7z

If so it did not work:
[2011-01-10 17:39:42 CST] PRPNet Client application v4.1.1 started
[2011-01-10 17:39:42 CST] User name mfbabb2 at email address is mfbabb@flash.net

Genefer 2.2.0 (x86 - 32-bit - 64-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144TEST.in'.
Testing 5624^262144+1... 3260416 steps to go ^C[2011-01-10 17:42:06 CST] GFN262
144TEST: Could not open file [genefer.log] for reading. Assuming user stopped w
ith ^C
[2011-01-10 17:42:06 CST] Accepted force quit. Waiting to close sockets before
exiting
Genefer80 2.2.0 (x86 - 32-bit - 80-b X87) Copyright (C) 2001-2003, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
A program for finding large probable generalized Fermat primes.

Start test of file 'work_GFN262144TEST.in'.
Testing 5624^262144+1... 3265626 steps to go

Control-C causes program to immediately start over from the beginning using Genefer80 instead of quitting.
____________
Murphy (AtP)

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30774 - Posted: 11 Jan 2011 | 0:31:57 UTC - in response to Message 30770.

Is this what is updated: http://uwin.mine.nu/PRPNet_files/prpclient-4.1.1cbeta-windows.7z.

No. I believe that John will put a 4.1.1dbeta out there for you.

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30775 - Posted: 11 Jan 2011 | 0:47:08 UTC - in response to Message 30774.

Is this what is updated: http://uwin.mine.nu/PRPNet_files/prpclient-4.1.1cbeta-windows.7z.

No. I believe that John will put a 4.1.1dbeta out there for you.

Client only files for 4.1.1d can be found here (MacIntel and Windows, Linux64 to come). Just replace the prpclient in your folders.

http://uwin.mine.nu/PRPNet_files/

EDIT: btw, it's confirmed working on MacIntel
____________

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30777 - Posted: 11 Jan 2011 | 1:49:15 UTC

Replaced prpclient.exe in folders.

Still goes to genefer80.exe after control-C, and it starts from beginning.
Doing a second control-C stops genefer80.exe (did not give it time to checkpoint).
Starting again now resumes from a checkpoint.

Win Vista 32-bit.
First control-C should quit, not cause fall-back to next "geneferX" in sequence.
____________
Murphy (AtP)

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30778 - Posted: 11 Jan 2011 | 3:15:38 UTC - in response to Message 30777.

Replaced prpclient.exe in folders.

Still goes to genefer80.exe after control-C, and it starts from beginning.
Doing a second control-C stops genefer80.exe (did not give it time to checkpoint).
Starting again now resumes from a checkpoint.

Win Vista 32-bit.
First control-C should quit, not cause fall-back to next "geneferX" in sequence.

That seems to be a problem. If there is no genefer.log after hitting ^C, then the client should shut down.

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 30784 - Posted: 11 Jan 2011 | 14:10:07 UTC - in response to Message 30778.

That seems to be a problem. If there is no genefer.log after hitting ^C, then the client should shut down.

I posted 4.1.2 here, http://home.roadrunner.com/~mrodenkirch/prpnet_4.1.2.zip. That should take care of this issue and the other outstanding issues (although I haven't tested it on this particular issue).

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30786 - Posted: 11 Jan 2011 | 16:43:04 UTC - in response to Message 30524.

Tried to run with GeneferCUDA.exe (on Win XP) and got alert "..failed to start... cudart32_32-16.dll was not found. ..."
Where can I get this?

It's a part of CUDA Toolkit, IIRC.
I've extracted DLL needed to run GeneferCUDA - http://pgllr.mine.nu/PRPNet/cudart.rar

I got the same error but using the dll you posted doesn't fix it...entirely.

Where exactly am I supposed to put it?

-

Edit: You have to manually rename 'em.

Uh ok, so now it recognizes that I have both cuff32_32_16.dll and cudart32_32_16.dll, but I get this:

"The procedure entry point cudaFuncSetCacheConfig could not be located in the dynamic link library cudart32_32_16.dll."

What exactly does that mean? As far as troubleshooting goes, I've hit a wall here.

Edit 2: Actually I don't think I have the toolkit installed. That's odd. Explains why I haven't been able to do much testing (or anything outside of BOINC) though.

Let's see if that solves things. For some reason I don't know if it will. But if it's working for everyone else...

btw, I'm using 32-bit WinXP and have set up the PRP master.ini exactly as it should be to run GeneferCUDA, which I downloaded from Ken's site.

Is that correct, or is there another version? I'm using the 4.11.1d client, if that makes any difference.
____________

Sideshow_Larry

Joined: 10 May 09
Posts: 15
ID: 39800
Credit: 44,805,529
RAC: 0

Message 30790 - Posted: 11 Jan 2011 | 18:37:08 UTC

Can someone confirm that my GTX260 216 core is not a double precision capable card?
I've managed to get my GTX460SE to work but keep getting the is genefer broken error message on the 260 (Win7-64).
____________

Scott Brown
Volunteer moderator
Volunteer tester
Project scientist

Joined: 17 Oct 05
Posts: 2107
ID: 1178
Credit: 8,211,199,358
RAC: 4,821,823

Message 30791 - Posted: 11 Jan 2011 | 18:47:27 UTC - in response to Message 30790.

Can someone confirm that my GTX260 216 core is not a double precision capable card?
I've managed to get my GTX460SE to work but keep getting the is genefer broken error message on the 260 (Win7-64).

The GTX 260 is a double precision capable card.

____________
141941*2^4299438-1 is prime!

Sideshow_Larry

Joined: 10 May 09
Posts: 15
ID: 39800
Credit: 44,805,529
RAC: 0

Message 30800 - Posted: 11 Jan 2011 | 21:07:38 UTC - in response to Message 30791.

Updated to driver 260.99 seems to be working now. I'm at client 4.1.1c.

____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30803 - Posted: 11 Jan 2011 | 21:15:03 UTC

PRPClient Update

PRPNet v4.1.2beta has been released for testing which addresses most of the issues experienced the past few days. You can download PrimeGrid's full PRPClient packages or just client only files. They are available here:

http://uwin.mine.nu/PRPNet_files/

Thanks for all the help in troubleshooting.
____________

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30805 - Posted: 11 Jan 2011 | 21:18:58 UTC

Ok. Pardon. What?

Why is this working for other people?

Me:

WinXP 32-bit SP3
1GB system RAM
Zotac GeForce GTX 460 @ 900/1100/1800 (1024MB)

Client I tried was 4.1.1d. GeneferCUDA was 0.97 I think.

It's hard for me to cobble together all these files from across the interwebs. If they could all be in one place with the latest versions clearly marked, that'd be great.

Also it says it can't find cudart32_32_16.dll, so I renamed the included cudart.dll and it seemed to work. Or at least, it didn't give me an error. It did, however, launch two client windows and close them almost immidiately.

Someone please either a) tell me what I did wrong or b) give me a brief walkthrough on how to get GeneferCUDA running correctly over here.

Thanks...!

____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30808 - Posted: 11 Jan 2011 | 21:46:32 UTC - in response to Message 30805.

It's hard for me to cobble together all these files from across the interwebs. If they could all be in one place with the latest versions clearly marked, that'd be great.

Everything that's needed is in the PRPClient packages.

Program versions included

prpclient - Version 4.1.2beta
llr - Version 3.8.4 (Linux, MacIntel, Windows)
pfgw - Version 3.4.4 (Linux, Windows, MacIntel)
phrot - Version 0.72 (MacIntel, MacPPC, PS3)
Genefer - Version 2.2.0 (Linux, Windows, MacIntel, MacPPC)
Genefer80 - Version 2.2.0 (Windows only)
GenefX64 - Version 2.2.0 (Windows 64 bit only)
GeneferCUDA - Version 0.97 (Windows and Linux only)
cudart.dll & cufft.dll (Windows only)

We have yet to pinpoint the "cudart32_32_16.dll" issue.

[EDIT] Windows package now includes cudart32_32_16.dll and cufft32_32_16.dll from the CUDA Toolkit 3.2 XP32 installation.
____________

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 30817 - Posted: 11 Jan 2011 | 23:35:36 UTC - in response to Message 30808.

Maybe something to do with source machine being 64-bit and target machine being 32-bit OS ... ?
____________
Murphy (AtP)

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30818 - Posted: 12 Jan 2011 | 0:15:43 UTC

Glad it turns out I wasn't doing something horribly wrong with my setup. It was pretty vanilla PRPnet with GeneferCUDA dropped in from Lennart's site.

I suppose I had downloaded the wrong files from some of the wrong places, or something like that. I'll hold off on trying to fix this myself until a stable client is released and GeneferCUDA is fully supported. Then it's on to PSA for me. ;)
____________

RODRIMPEX Robert ÅšlÄ™zak

Joined: 17 May 10
Posts: 11
ID: 60680
Credit: 7,302,709
RAC: 0

Message 30820 - Posted: 12 Jan 2011 | 0:24:22 UTC

Hi, I've got that problem:

Start test of file 'work_GFN32768.in' - 01:05:59
GeneferCUDA.cu(216) : cufftSafeCall() CUFFT error.

What is a problem in that function?

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30821 - Posted: 12 Jan 2011 | 0:27:22 UTC - in response to Message 30820.

Hi, I've got that problem:

Start test of file 'work_GFN32768.in' - 01:05:59
GeneferCUDA.cu(216) : cufftSafeCall() CUFFT error.

What is a problem in that function?

Your NVIDIA graphics card (9500GT) is not DP capable.
____________

John
Honorary cruncher

Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0

Message 30823 - Posted: 12 Jan 2011 | 0:39:52 UTC

Has anyone tested v4.1.2 package which includes cudart32_32_16.dll and cufft32_32_16.dll from the CUDA Toolkit 3.2 XP32 installation?
____________

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30825 - Posted: 12 Jan 2011 | 3:47:11 UTC

Ok I've downloaded that - will test in approximately 5 minutes as I'm waiting on a DNETC thing to get out of my GPU and leave it free for testing this.

In my problem above (or somewhere, I guess I posted it here) should note it does benchmark fine. Let's see if this client fixes it!

EDIT:

It works! Well...sorta.

You have to copy and paste cudart32_32_16.dll and cufft32_32_16.dll into prpclient-1 and prpclient-2 (or however many folders you have).

Now for the odd stuff...It just uses my CPU unless I go and comment out genefer.exe and genefer80.exe.

Ok so that's done. Now Task Manager shows 2 instances of GeneferCUDA (um I only have one GPU!) But luckily, GPU-Z shows 99% load at full clocks (oc'd a ton)

I'm seeing big numbers on the screen. Is this good? :D

Testing 5744^262144+1... 3264512 steps to go maxErr exceeded for 5744^262144+1, 0.5000 > 0.4500 PRP: 1*5744^262144+1 5000/3273612

It would appear as though I am running two "tasks" at once on my GTX 460. Not that I'm concerned, but what kinda timeframe am I looking at, exactly?

And thanks a lot for whatever you did that made it work, John et al ;)
____________

Joined: 1 Oct 10
Posts: 256
ID: 69151
Credit: 106,985,418
RAC: 13,748

Message 30826 - Posted: 12 Jan 2011 | 4:42:27 UTC

Oof. 202500/3273612. I'm just gonna checkpoint it and run BOINC stuff overnight. I'm oddly rather tired, but I've the day off tomorrow to enjoy the snow whilst helping y'all finish testing this. Well...super-expediently, given the challenge starts in a half hour by my clock here.

Bottom line: it is working for me and hopefully everyone else who had similar problems. Thanks! I'll let you know results sometime tomorrow. Chances are someone else will have by then, but just for now:

Pentium 4 3.0GHz / HT
1GB system RAM
WinXP 32-bit SP3
NViDIA GeForce GTX 460 @ 910/1100/1800

Looks as though it will take a little over an hour for this test. Checkpointing now and will get back to you later.

You asked if anyone's tested it, so far I have and I'm sure others have too. ;)

Bear in mind that a) it tries to run two at once on a dual install, and probably 4 on a quad, etc...even if you have just one GPU. Looks like one task just kinda hangs there while the other powers through. b) you must copy the two .dlls to the prpclient-1 and prpclient-2 etc folders in order for the app to work.

Nice.
____________

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30827 - Posted: 12 Jan 2011 | 4:45:46 UTC - in response to Message 30826.

Bear in mind that a) it tries to run two at once on a dual install, and probably 4 on a quad, etc...even if you have just one GPU. Looks like one task just kinda hangs there while the other powers through. b) you must copy the two .dlls to the prpclient-1 and prpclient-2 etc folders in order for the app to work.

Nice.

My solution: Two PRPnet directories: 1 installation ("single") for the GPU and one ("quad") for the CPU ;)
____________

bdodson*

Joined: 13 Nov 09
Posts: 104
ID: 50079
Credit: 170,522,809
RAC: 0

Message 30865 - Posted: 12 Jan 2011 | 17:39:40 UTC - in response to Message 30827.

Bear in mind that a) it tries to run two at once on a dual install, and probably 4 on a quad, etc...even if you have just one GPU. Looks like one task just kinda hangs there while the other powers through. b) you must copy the two .dlls to the prpclient-1 and prpclient-2 etc folders in order for the app to work.

Nice.

My solution: Two PRPnet directories: 1 installation ("single") for the GPU and one ("quad") for the CPU ;)

How's the cuda3.2 -vs- 3.1 issue from your earlier post? I'm considering
switching the tesla 2s from proth_sieving, but unlikely to be able to load
drivers without sysadmin help (I don't own the machine/beast). Two cards.

-Bruce

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30866 - Posted: 12 Jan 2011 | 17:42:59 UTC - in response to Message 30865.

Bear in mind that a) it tries to run two at once on a dual install, and probably 4 on a quad, etc...even if you have just one GPU. Looks like one task just kinda hangs there while the other powers through. b) you must copy the two .dlls to the prpclient-1 and prpclient-2 etc folders in order for the app to work.

Nice.

My solution: Two PRPnet directories: 1 installation ("single") for the GPU and one ("quad") for the CPU ;)

How's the cuda3.2 -vs- 3.1 issue from your earlier post? I'm considering
switching the tesla 2s from proth_sieving, but unlikely to be able to load
drivers without sysadmin help (I don't own the machine/beast). Two cards.

-Bruce

The Genefer CUDA app reports as GenferCUDA 0.97 (CUDA3.2) but also worked with driver version 256.53 and the CUDA 3.1 SDK installed. I tested this on the test port GFN262144TEST. Currently I've installed driver version 260.19.29 and the CUDA 3.2 SDK. Everything is running fine.

BTW: Since the app seems to do DP processing it would be interesting to see the performance of some Teslas compared to their consumer counterparts.

BTW2: A test run of llrCUDA on them would be interesting too ;)
____________

Rick Reynolds

Joined: 13 Aug 10
Posts: 7777
ID: 65483
Credit: 5,140,435,769
RAC: 583,657

Message 30871 - Posted: 12 Jan 2011 | 19:24:05 UTC

Question: Running PRPNet on a Windows Server 2003 (32 bit) that has never been part of Boinc or Primegrid.

Since Prpnet is a manual run, can I run it on a server that I can't get Boinc to run on and still have it work and have the credit applied to Primegrid? My thought was to install the apps but fake the client id to one of the pc's I do have on Boinc.
____________
@AggieThePew

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 30874 - Posted: 12 Jan 2011 | 19:35:04 UTC - in response to Message 30871.

Question: Running PRPNet on a Windows Server 2003 (32 bit) that has never been part of Boinc or Primegrid.

Since Prpnet is a manual run, can I run it on a server that I can't get Boinc to run on and still have it work and have the credit applied to Primegrid? My thought was to install the apps but fake the client id to one of the pc's I do have on Boinc.

PSA credits are added to the BOINC credits but not to a special PC that is listed in your computer list. As long as the admins can connect the PRPnet user with a BOINC user everything (email) should be fine.
____________

Rick Reynolds

Joined: 13 Aug 10
Posts: 7777
ID: 65483
Credit: 5,140,435,769
RAC: 583,657

Message 30875 - Posted: 12 Jan 2011 | 19:38:02 UTC - in response to Message 30874.

Question: Running PRPNet on a Windows Server 2003 (32 bit) that has never been part of Boinc or Primegrid.

Since Prpnet is a manual run, can I run it on a server that I can't get Boinc to run on and still have it work and have the credit applied to Primegrid? My thought was to install the apps but fake the client id to one of the pc's I do have on Boinc.

PSA credits are added to the BOINC credits but not to a special PC that is listed in your computer list. As long as the admins can connect the PRPnet user with a BOINC user everything (email) should be fine.

thanks !!! will give it a go since I can't seem to get Boinc to run on the server even though I know others have found a way to work around the issue.

Rick
____________
@AggieThePew

rogue
Volunteer developer

Joined: 8 Sep 07
Posts: 1195
ID: 12001
Credit: 18,565,548
RAC: 0

Message 31002 - Posted: 14 Jan 2011 | 21:25:25 UTC

I found a bug where the client reports the wrong version of genefer to the server when a test is completed. I'll send out a patch tonight or tomorrow.

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0

Message 31048 - Posted: 16 Jan 2011 | 0:11:43 UTC

Did this WU get lost:

[2011-01-14 21:11:19 CST] User name mfbabb2 at email address is xxxxxx@xxxxx.xxx [2011-01-15 12:25:45 CST] GFN524288: 1*8440^524288+1 is not prime. Residue 51c42e47ceecab32 [2011-01-15 12:25:45 CST] Total Time: 15:14:27 Total Tests: 1 Total PRPs Found: 0 [2011-01-15 12:25:50 CST] GFN524288: Returning work to server prime2u.com at port 11001 [2011-01-15 12:25:50 CST] 324: nothing was received on socket after 10 seconds [2011-01-15 12:25:50 CST] 324: nothing was received on socket after 2 seconds [2011-01-15 12:25:52 CST] prime2u.com:11001 connect to socket failed with error 10061

____________
Murphy (AtP)

Microcruncher*
Volunteer tester

Joined: 28 Jun 09
Posts: 391
ID: 42625
Credit: 45,226,534
RAC: 0

Message 31049 - Posted: 16 Jan 2011 | 0:19:25 UTC - in response to Message 31048.

Did this WU get lost:

[2011-01-14 21:11:19 CST] User name mfbabb2 at email address is xxxxxx@xxxxx.xxx [2011-01-15 12:25:45 CST] GFN524288: 1*8440^524288+1 is not prime. Residue 51c42e47ceecab32 [2011-01-15 12:25:45 CST] Total Time: 15:14:27 Total Tests: 1 Total PRPs Found: 0 [2011-01-15 12:25:50 CST] GFN524288: Returning work to server prime2u.com at port 11001 [2011-01-15 12:25:50 CST] 324: nothing was received on socket after 10 seconds [2011-01-15 12:25:50 CST] 324: nothing was received on socket after 2 seconds [2011-01-15 12:25:52 CST] prime2u.com:11001 connect to socket failed with error 10061

If you look into the work_GFN524288.save file and don't find an entry for it there: Yes. It still appears here:

http://prime2u.com:11001/pending_tests.html
____________

mfbabb2
Volunteer tester

Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 11,541,356
RAC: 0