| Author |
Message |
|---|
 Crun-chi Volunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
                         
|
|
I run 4 instances of PPRNet client and one more for GENEFERCuda since I have Geforce 560TI.
But whenever I stop client with CTRL+c it says something like "is genefer broken" and when I restart client it cannot find checkpoint for start.
I use latest 4.37 PRPNet package under windows 7 sp1 x64.
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
 
|
|
The message is OK, but the missing checkpoint is a different issue. It might be that you are stopping it before it writes the checkpoint for the first time, or it could be a different issue. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
                         
|
|
Crun-Che, is the problem only with restarting GeneferCUDA? Does it work correctly if it's not interrupted? Or does it not work at all?
You're running GFN262144, right? Your 560 TI should be able to complete one of those in less than an hour and a half. That's how long it takes my 460 to do one of those. If you let it run for 90 minutes, what happens? Does it complete the WU? Does it pass the initialization phase and star counting down? It would look something like this:
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_GFN524288.in' - 14:21:41
Resuming 88042^524288+1 from a checkpoint
Testing 88042^524288+1... 1921024 steps to go
The steps to go number should be changing approximately once every four seconds, or maybe a bit faster.
Or is the problem that it runs fine if you let it complete, but it just can't be restarted without starting at the beginning?
If it's running fine if you don't interrupt it, it's probably some kind of installation/folder privilege problem that's preventing it from writing the checkpoint file. If it's not able to complete a WU, maybe it's that memory organization problem Scott Brown identified with the 550 TI. Perhaps it affects the 560 TI also.
____________
My lucky number is 75898524288+1 |
|
|
LumiukkoVolunteer tester
Send message
Joined: 7 Jul 08
Posts: 165
ID: 25183
Credit: 875,354,536
RAC: 114,800
               
|
|
I had the same problem with 4.3.7, it runs fine if you just let it run. Problem is only when I need to stop and restart it. It can not resume from the checkpoint file. It does find the checkpoint file, but it claims that the genefer version that created the checkpoint file "Geneferxxx" is not compatible, and it starts the work unit from the beginning ("Geneferxxx" = Genefer+"some strange ascii characters").
Same happens with my GTX275 and GTX285 cards in Win7_64 and WinXP_32, and GTX480 in Linux_64.
4.3.6. works fine.
--
Lumiukko |
|
|
Crun-chiVolunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
|
|
So I will try 4.3.6 :)
My 560 Ti complete in 1:15 hours and all is ok if you leave from start: then one WU is over, second is started.
So I find solution. I clear checkpoint ,and all start ok
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
|
|
I had the same problem with 4.3.7, it runs fine if you just let it run. Problem is only when I need to stop and restart it. It can not resume from the checkpoint file. It does find the checkpoint file, but it claims that the genefer version that created the checkpoint file "Geneferxxx" is not compatible, and it starts the work unit from the beginning ("Geneferxxx" = Genefer+"some strange ascii characters").
Same happens with my GTX275 and GTX285 cards in Win7_64 and WinXP_32, and GTX480 in Linux_64.
4.3.6. works fine.
--
Lumiukko
Yes, I went back to 4.3.6 also. I was running it on Mint Linux 64/GTX 470. It just does not work after you try to restart after a "control c".
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
This implies a bug that I introduced between 4.3.6 and 4.3.7. I'll have to look into it. |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
This implies a bug that I introduced between 4.3.6 and 4.3.7. I'll have to look into it.
Mark,
The bug would be in GeneferCUDA rather than PRPNet, right? Perhaps just rolling back to the older version of GeneferCUDA would be sufficient?
I never switched to 4.3.7, not because of this bug, but because it seemed that the newer version of GeneferCUDA included in 4.3.7 was slower than the version in 4.3.6, based upon what I was reading on the forums.
Mike
____________
My lucky number is 75898524288+1 |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
               
|
This implies a bug that I introduced between 4.3.6 and 4.3.7. I'll have to look into it.
Also note that geneferCUDA was updated as well. It would be interesting to see if the geneferCUDA in the 4.3.7 package produces the same error with prpclient in the 4.3.6 package.
In the 4.3.6 PRPClient package:
GeneferCUDA - Version 0.97 - based on Genefer 2.2.0 (Windows and Linux64 only)
GenefX64 - Version 2.2.0 (Windows x64); Version 2.2.1 (Linux x64, MacIntel x64)
Genefer - Version 2.2.0 (Windows, MacPPC); Version 2.2.1 (Linux, MacIntel)
Genefer80 - Version 2.2.0 (Windows); Version 2.2.1 (Linux, MacIntel) In the 4.3.7 PRPClient package:
GeneferCUDA - Version 1.01 - based on Genefer 2.2.1 (Windows, Linux64, MacIntel)
GenefX64 - Version 2.2.1 (Windows x64, Linux x64, MacIntel x64)
Genefer - Version 2.2.1 (Windows, MacPPC, Linux, MacIntel)
Genefer80 - Version 2.2.1 (Windows, Linux, MacIntel) I also recall an issue specific with the Windows build of geneferCUDA 2.2.1. Are these errors happening only on Windows machines?
____________
|
|
|
Crun-chiVolunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
|
|
[2011-12-22 01:19:30 CEST] GFN262144: PRPNet server is version 4.3.7
[2011-12-22 01:37:04 CEST] GFN262144: No data in file [genefer.log]. Is genefer broken?
[2011-12-22 02:23:10 CEST] PRPNet Client application v4.3.7 started
[2011-12-22 02:23:10 CEST] User name Crun-chi at email address is xxxxxxxxxx
[2011-12-22 02:23:10 CEST] GFN262144: Getting work from server prpnet.mine.nu at port 11002
[2011-12-22 02:23:12 CEST] GFN262144: PRPNet server is version 4.3.7
[2011-12-22 03:33:17 CEST] GFN262144: 1*414050^262144+1 is not prime. ...
And one more thing: sometimes when I start client I got this messages:
[2011-12-22 20:25:58 CEST] Server: GFN262144, Candidate: 1*414196^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
[2011-12-22 20:26:54 CEST] Server: GFN262144, Candidate: 1*414214^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
[2011-12-22 20:27:49 CEST] Server: GFN262144, Candidate: 1*414222^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
Does that mean that GENEFER CUDA cannot check that numbers?
In message http://www.primegrid.com/forum_thread.php?id=3249&nowrap=true#34615 Rogue say not to use GFN262144 with GENEFERCUDA. What to use?
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Crun-chiVolunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
|
This implies a bug that I introduced between 4.3.6 and 4.3.7. I'll have to look into it.
Mark,
The bug would be in GeneferCUDA rather than PRPNet, right? Perhaps just rolling back to the older version of GeneferCUDA would be sufficient?
I never switched to 4.3.7, not because of this bug, but because it seemed that the newer version of GeneferCUDA included in 4.3.7 was slower than the version in 4.3.6, based upon what I was reading on the forums.
Mike
How much slower?
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
How much slower?
I don't remember exactly, but I think it was a few percent. You would have have to find the post about it. I may have commented about it, so searching for just my posts within the last 30 days or so should find it. I'll post a link if I can find it.
____________
My lucky number is 75898524288+1 |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
How much slower?
I don't remember exactly, but I think it was a few percent. You would have have to find the post about it. I may have commented about it, so searching for just my posts within the last 30 days or so should find it. I'll post a link if I can find it.
Ok, found it:
This message has the benchmark timings.
Looks like about a 10% slowdown, which is pretty significant.
Of course, you can just run the benchmarks from both versions of GeneferCUDA yourself and see what you get on your hardware.
____________
My lucky number is 75898524288+1 |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
And one more thing: sometimes when I start client I got this messages:
[2011-12-22 20:25:58 CEST] Server: GFN262144, Candidate: 1*414196^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
[2011-12-22 20:26:54 CEST] Server: GFN262144, Candidate: 1*414214^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
[2011-12-22 20:27:49 CEST] Server: GFN262144, Candidate: 1*414222^262144+1 Program: genefercuda.exe Residue: roundofferr Time: 0 seconds
Does that mean that GENEFER CUDA cannot check that numbers?
Yes, but the round off error is more typical as the application reaches its testing limits. For N=262144 this is around b=815k.
Other GPUs have experienced this but the cases are so infrequent that we haven't been able to establish a pattern yet.
____________
|
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
|
On my GTX 460, with the ranges relevant to us highlighted:
C:\prpclient-4.3.6-windows-gpu\test2>genefercuda097 -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)
Copyright (C) 2010, Shoichiro Yamada (CUDA)
A program for finding large probable generalized Fermat primes.
Generalized Fermat Number Bench
2009574^8192+1 Time: 394 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 419 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 451 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 585 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 716 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 949 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.54 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 2.42 ms/mul. Err: 1.48e-001 5946413 digits
380742^2097152+1 Time: 3.85 ms/mul. Err: 3.63e-001 11703432 digits
C:\prpclient-4.3.6-windows-gpu\test2>genefercuda221 -b
GeneferCUDA 2.2.1 (CUDA) based on Genefer v2.2.1
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, 2011 Mark Rodenkirch, David Underbakke (v2.2.1)
Copyright (C) 2010, 2011, Shoichiro Yamada (CUDA)
A program for finding large probable generalized Fermat primes.
Generalized Fermat Number Bench
2009574^8192+1 Time: 424 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 428 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 454 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 594 us/mul. Err: 1.88e-001 395325 digits
874718^131072+1 Time: 726 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 1.04 ms/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.67 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 2.57 ms/mul. Err: 1.52e-001 5946413 digits
380742^2097152+1 Time: 4.04 ms/mul. Err: 3.63e-001 11703432 digits
309258^4194304+1 Time: 7.62 ms/mul. Err: 1.47e-001 23028076 digits
251196^8388608+1 Time: 15 ms/mul. Err: 1.41e-001 45298590 digits
The first benchmark above is from the version of GeneferCUDA included in 4.3.6, and the second is the version included in 4.3.7.
____________
My lucky number is 75898524288+1 |
|
|
Crun-chiVolunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
|
|
Yes I can confirm it : I also got few % better result with old GeneferCUDA ( Geforce 560 TI) that uses Cuda 3.2. Will wait until first WU is finished to see how difference is really big :)
Thanks for tip Michael
Thumb up for that!
GENEFER CUDA 436
Generalized Fermat Number Bench
2009574^8192+1 Time: 316 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 321 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 338 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 416 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 508 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 674 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.04 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 1.62 ms/mul. Err: 1.48e-001 5946413 digits
380742^2097152+1 Time: 2.47 ms/mul. Err: 3.63e-001 11703432 digits
GENEFER CUDA 437
Generalized Fermat Number Bench
2009574^8192+1 Time: 337 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 334 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 342 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 423 us/mul. Err: 1.88e-001 395325 digits
874718^131072+1 Time: 518 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 724 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.1 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 1.74 ms/mul. Err: 1.52e-001 5946413 digits
380742^2097152+1 Time: 2.59 ms/mul. Err: 3.63e-001 11703432 digits
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
Crun-chiVolunteer tester
Send message
Joined: 25 Nov 09
Posts: 3233
ID: 50683
Credit: 151,443,349
RAC: 73,965
|
|
Update : from average 4180 to 3985 , so about 200 seconds faster.
____________
92*10^1585996-1 NEAR-REPDIGIT PRIME :) :) :)
4 * 650^498101-1 CRUS PRIME
2022202116^131072+1 GENERALIZED FERMAT
Proud member of team Aggie The Pew. Go Aggie! |
|
|
LumiukkoVolunteer tester
Send message
Joined: 7 Jul 08
Posts: 165
ID: 25183
Credit: 875,354,536
RAC: 114,800
|
This implies a bug that I introduced between 4.3.6 and 4.3.7. I'll have to look into it.
Also note that geneferCUDA was updated as well. It would be interesting to see if the geneferCUDA in the 4.3.7 package produces the same error with prpclient in the 4.3.6 package.
In the 4.3.6 PRPClient package:
GeneferCUDA - Version 0.97 - based on Genefer 2.2.0 (Windows and Linux64 only)
GenefX64 - Version 2.2.0 (Windows x64); Version 2.2.1 (Linux x64, MacIntel x64)
Genefer - Version 2.2.0 (Windows, MacPPC); Version 2.2.1 (Linux, MacIntel)
Genefer80 - Version 2.2.0 (Windows); Version 2.2.1 (Linux, MacIntel) In the 4.3.7 PRPClient package:
GeneferCUDA - Version 1.01 - based on Genefer 2.2.1 (Windows, Linux64, MacIntel)
GenefX64 - Version 2.2.1 (Windows x64, Linux x64, MacIntel x64)
Genefer - Version 2.2.1 (Windows, MacPPC, Linux, MacIntel)
Genefer80 - Version 2.2.1 (Windows, Linux, MacIntel) I also recall an issue specific with the Windows build of geneferCUDA 2.2.1. Are these errors happening only on Windows machines?
4.3.7 package in Linux machine seems to be working fine.
in Windows:
Genefercuda from 4.3.7 does not work with PRPclient 4.3.7.
Genefercuda from 4.3.7 does not work with PRPclient 4.3.6.
Genefercuda from 4.3.6 does work with PRPclient 4.3.7.
Here is the restart of Genefercuda from 4.3.7 with PRPclient 4.3.6. Win7x64 GTX275 (with debuglevel=1):
[2011-12-24 10:02:43 FST] GFN524288: AddWorkUnitToList for 88172^524288+1
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitCompleted for 88172^524288+1. Returning false (not c
ompleted, main)
[2011-12-24 10:02:43 FST] PRPNet Client application v4.3.6 started
[2011-12-24 10:02:43 FST] User name Lumiukko at email address is xxxxx@xxxx.fi
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitInProgress for 88172^524288+1. Returning true (in pr
ogress, main)
[2011-12-24 10:02:43 FST] GFN524288: HaveInProgressTest for 88172^524288+1. Returning true (in prog
ress)
[2011-12-24 10:02:43 FST] GFN524288: Returning work. currentworkunits=1, completedworkunits=0, quit
Option=2
[2011-12-24 10:02:43 FST] GFN262144: CanDoAnotherTest. Returning false (none in progress)
[2011-12-24 10:02:43 FST] GFN262144: CanDoAnotherTest. Returning false (all done)
[2011-12-24 10:02:43 FST] GFN524288: Returning work. currentworkunits=1, completedworkunits=0, quit
Option=3
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitCompleted for 88172^524288+1. Returning false (not c
ompleted, main)
[2011-12-24 10:02:43 FST] GFN524288: CanDoAnotherTest for 88172^524288+1. Returning true (not done)
[2011-12-24 10:02:43 FST] GFN262144: Returning work. currentworkunits=0, completedworkunits=0, quit
Option=3
[2011-12-24 10:02:43 FST] in FindAnyIncompleteWork
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitInProgress for 88172^524288+1. Returning true (in pr
ogress, main)
[2011-12-24 10:02:43 FST] GFN524288: HaveInProgressTest for 88172^524288+1. Returning true (in prog
ress)
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitCompleted for 88172^524288+1. Returning false (not c
ompleted, main)
[2011-12-24 10:02:43 FST] GFN524288: CanDoAnotherTest for 88172^524288+1. Returning true (not done)
[2011-12-24 10:02:43 FST] Work: GFN524288, HaveInProgressTest = 1, CanDoAnotherTest = 1
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitCompleted for 88172^524288+1. Returning false (not c
ompleted, main)
[2011-12-24 10:02:43 FST] GFN524288: CanDoAnotherTest for 88172^524288+1. Returning true (not done)
[2011-12-24 10:02:43 FST] GFN524288: ProcessWorkUnit looking for work. In Progress=0
[2011-12-24 10:02:43 FST] GFN524288: IsWorkUnitCompleted for 88172^524288+1. Returning false (not c
ompleted, main)
[2011-12-24 10:02:43 FST] GFN524288: ProcessWorkUnit will test 88172^524288+1
GeneferCUDA 2.2.1 (CUDA) based on Genefer v2.2.1
Copyright (C) 2001-2003, Yves Gallot (v1.3)
Copyright (C) 2009, 2011 Mark Rodenkirch, David Underbakke (v2.2.1)
Copyright (C) 2010, 2011, Shoichiro Yamada (CUDA)
A program for finding large probable generalized Fermat primes.
Start test of file 'work_GFN524288.in' - 10:02:43
Checkpoint saved by genefer CUDA. CUDA This version cannot restart with itCheckpoint saved by genef
er ╠²↑. This version cannot restart with it
Base prime factor(s) taken : 47, 67
Resuming N-1 prime test of 88172^524288+1 at bit 1869 [0.02%]
Using generic reduction Core2 type-3 FFT length 896K, Pass1=896, Pass2=1K, a = 3
88172^524288+1, bit: 10000 / 8613005 [0.11%]. Time per bit: 107.171 ms.
--
Lumiukko
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
in Windows:
Genefercuda from 4.3.7 does not work with PRPclient 4.3.7.
Genefercuda from 4.3.7 does not work with PRPclient 4.3.6.
Genefercuda from 4.3.6 does work with PRPclient 4.3.7.
Points to a Windows GeneferCUDA - Version 1.01 - based on Genefer 2.2.1 issue and not a PRPNet issue. Thanks for confirming this.
____________
|
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
         
|
|
I compiled the source for GeneferCUDA_1.03 from mersenneforum and have problems too.
Checkpointing does not seem to work. The checkpoint-file has ever a size of zero.
Maybe this causes the hieroglyphs in the message:
"Checkpoint saved by genefer CUDA. CUDA This version cannot restart with it
Checkpoint saved by genefer ╠²↑. This version cannot restart with it" ???
[add]
Here is my source of genefercuda.1.03.1.tar.bz2 with a linux64-executable for sm_13, sm_20 and sm_21. It was compiled against Cuda4.0.
I solved only all compiler warnings about undefined identifier, not referenced variables and even more messages with "result of call is not used" in version 1.03...
The sources for the older version GeneferCUDA_0.98 or the new llrCUDA_0.80 in mersenneforum produce the same problems...
____________
Best wishes. Knowledge is power. by jjwhalen
 |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
I compiled the source for GeneferCUDA_1.03 from mersenneforum and have problems too.
Checkpointing does not seem to work. The checkpoint-file has ever a size of zero.
Maybe this causes the hieroglyphs in the message:
"Checkpoint saved by genefer CUDA. CUDA This version cannot restart with it
Checkpoint saved by genefer ╠²↑. This version cannot restart with it" ???
Sorry folks, we have made no progress in resolving this issue. The Windows geneferCUDA in the 4.3.7 package has been returned to the 0.97 version until we can get this issue corrected.
As for the slowdown, it appears to be caused by compiling it under CUDA 4.0 as opposed to the previous CUDA 3.2.
If anyone is able to help with these issues, please let us know.
____________
|
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
As for the slowdown, it appears to be caused by compiling it under CUDA 4.0 as opposed to the previous CUDA 3.2.
That's what I feared.
Is there any benefit to using CUDA 4.0 instead of 3.2? My guess would be that it has support for newer cards not supported under 3.2, perhaps correcting that funky memory problem with the 550TI. The slowdown might be due to that fix. Lots of speculation there, but if it's correct, it may be worthwhile compiling a separate version for the 550TI. That won't work long term, of course.
For now, does anyone have a link where you can you download the CUDA 3.2 tools? I could only find the 4.0.
____________
My lucky number is 75898524288+1 |
|
|
|
|
|
http://developer.nvidia.com/cuda-toolkit-32-downloads
____________
|
|
|
|
|
|
Hi ,rroonnaalldd
Here is my source of genefercuda.1.03.1.tar.bz2 with a linux64-executable for sm_13, sm_20 and sm_21. It was compiled against Cuda4.0.
Thank you for your work.
|
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
|
Hi ,rroonnaalldd
Here is my source of genefercuda.1.03.1.tar.bz2 with a linux64-executable for sm_13, sm_20 and sm_21. It was compiled against Cuda4.0.
Thank you for your work.
Shoichiro, do you have an idea why checkpointing in 1.03 (and maybe 0.98 too) does not work?
@All
Does anybody have or found a better [HOW TO] Setup CUDA Compiler on Code::Blocks for integrating CUDA in Code::Blocks???
I use C::B in linux64 remote via Xming X Server on my Windows-based VMware VMserver1-Host...
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
|
|
Shoichiro, do you have an idea why checkpointing in 1.03 (and maybe 0.98 too) does not work?
My personal opinion.
Please try.
944 #ifdef WIN32
945 // SetConsoleCtrlHandler(HandlerRoutine, TRUE);
946 #else
|
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14011
ID: 53948
Credit: 433,209,141
RAC: 977,735
|
Hi ,rroonnaalldd
Here is my source of genefercuda.1.03.1.tar.bz2 with a linux64-executable for sm_13, sm_20 and sm_21. It was compiled against Cuda4.0.
Thank you for your work.
Shoichiro,
What was the reason for switching to the CUDA 4.0 toolkit? I've done some tests, and the 4.0 toolkit produces code that's about 10% slower than 3.2:
3.2:
Generalized Fermat Number Bench
2009574^8192+1 Time: 409 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 433 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 467 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 587 us/mul. Err: 1.72e-001 395325 digits
874718^131072+1 Time: 718 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 958 us/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.55 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 2.43 ms/mul. Err: 1.56e-001 5946413 digits
380742^2097152+1 Time: 3.86 ms/mul. Err: 3.63e-001 11703432 digits
309258^4194304+1 Time: 6.81 ms/mul. Err: 1.56e-001 23028076 digits
251196^8388608+1 Time: 13.6 ms/mul. Err: 1.41e-001 45298590 digits
4.0:
Generalized Fermat Number Bench
2009574^8192+1 Time: 425 us/mul. Err: 3.82e-001 51636 digits
1632282^16384+1 Time: 428 us/mul. Err: 2.53e-001 101791 digits
1325824^32768+1 Time: 457 us/mul. Err: 1.88e-001 200622 digits
1076904^65536+1 Time: 592 us/mul. Err: 1.88e-001 395325 digits
874718^131072+1 Time: 725 us/mul. Err: 3.47e-001 778813 digits
710492^262144+1 Time: 1.05 ms/mul. Err: 4.21e-001 1533952 digits
577098^524288+1 Time: 1.68 ms/mul. Err: 2.01e-001 3020555 digits
468750^1048576+1 Time: 2.58 ms/mul. Err: 1.52e-001 5946413 digits
380742^2097152+1 Time: 4.08 ms/mul. Err: 3.63e-001 11703432 digits
309258^4194304+1 Time: 7.68 ms/mul. Err: 1.47e-001 23028076 digits
251196^8388608+1 Time: 15.1 ms/mul. Err: 1.41e-001 45298590 digits
Both were compiled from the same source code (0.99), except for changing #define CPU_TARGET "CUDA3.2" to #define CPU_TARGET "CUDA4.0" and building with the 4.0 toolkit instead of the 3.2 toolkit.
Mike
P.S. Thank you for the great work creating GeneferCUDA!!!
____________
My lucky number is 75898524288+1 |
|
|
|
|
|
Hi,
What was the reason for switching to the CUDA 4.0 toolkit? I've done some tests, and the 4.0 toolkit produces code that's about 10% slower than 3.2:
I provided 64 bit linux version.
About 64 bit linux it is no reason. |
|
|
|
|
|
Hi ,rroonnaalldd
Could you give me information about this issue.
Thank you,
|
|
|
|
|
I compiled the source for GeneferCUDA_1.03 from mersenneforum and have problems too.
Checkpointing does not seem to work. The checkpoint-file has ever a size of zero.
Maybe this causes the hieroglyphs in the message:
"Checkpoint saved by genefer CUDA. CUDA This version cannot restart with it
Checkpoint saved by genefer ╠²↑. This version cannot restart with it" ???
Sorry folks, we have made no progress in resolving this issue. The Windows geneferCUDA in the 4.3.7 package has been returned to the 0.97 version until we can get this issue corrected.
As for the slowdown, it appears to be caused by compiling it under CUDA 4.0 as opposed to the previous CUDA 3.2.
If anyone is able to help with these issues, please let us know.
Please see.
http://mersenneforum.org/showpost.php?p=284535&postcount=99
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
I compiled the source for GeneferCUDA_1.03 from mersenneforum and have problems too.
Checkpointing does not seem to work. The checkpoint-file has ever a size of zero.
Maybe this causes the hieroglyphs in the message:
"Checkpoint saved by genefer CUDA. CUDA This version cannot restart with it
Checkpoint saved by genefer ╠²↑. This version cannot restart with it" ???
Sorry folks, we have made no progress in resolving this issue. The Windows geneferCUDA in the 4.3.7 package has been returned to the 0.97 version until we can get this issue corrected.
As for the slowdown, it appears to be caused by compiling it under CUDA 4.0 as opposed to the previous CUDA 3.2.
If anyone is able to help with these issues, please let us know.
Please see.
http://mersenneforum.org/showpost.php?p=284535&postcount=99
Thanks Michael for the checkpoint fix! :)
____________
|
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
|
|
I did some minor cosmetics (i*2) to ((i)*2) in v1.051.
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
