mersenneforum.org CUDALucas (a.k.a. MaclucasFFTW/CUDA 2.3/CUFFTW)
 Register FAQ Search Today's Posts Mark Forums Read

 2011-02-17, 17:44 #419 Svenie25     Aug 2008 Good old Germany 3·47 Posts Thanks a lot. I will give it a try, then I start the next exponent tomorrow. Hopefully there will be no blackout. ;)
2011-02-17, 19:47   #420
aaronhaviland

Jan 2011
Dudley, MA, USA

1118 Posts

This simple patch lets the program try to resume from a checkpoint file if it exists, instead of having to specify the checkpoint file manually. If the primary c# checkpoint file does not exist, it will try the backup t# file, before finally starting the exponent from scratch.

You can still specify the checkpoint file manually, if needed.

(Patch against cudalucas-1.1)

Code:
orion@anubis:~/projects/gimps/cudalucas/cudalucas.1.1-new$rm c216091 t216091 orion@anubis:~/projects/gimps/cudalucas/cudalucas.1.1-new$ ./CUDALucas 216091
CUDALucas: Could not find a checkpoint file to resume from
^C
orion@anubis:~/projects/gimps/cudalucas/cudalucas.1.1-new$./CUDALucas 216091 CUDALucas: Resuming from checkpoint file c216091 caso 2 ^C orion@anubis:~/projects/gimps/cudalucas/cudalucas.1.1-new$ ./CUDALucas t216091
CUDALucas: Resuming from checkpoint file t216091
caso 2
^C
Attached Files
 checkpoint.diff.gz (526 Bytes, 107 views)

 2011-02-18, 03:43 #421 msft     Jul 2009 Tokyo 61010 Posts Nice patch,Thank you!
 2011-02-19, 13:00 #422 Karl M Johnson     Mar 2010 6338 Posts What does n mean in this string ? Code: Iteration $iter$ M( $exp$ )C, $checkpoint$, n = 8388608, CUDALucas v1.0 I remember it has something to do with error rate. Last fiddled with by Karl M Johnson on 2011-02-19 at 13:01 Reason: Yes.
2011-02-19, 16:25   #423
Brain

Dec 2009
Peine, Germany

331 Posts
FFT length..?

Quote:
 Originally Posted by Karl M Johnson What does n mean in this string ? Code: Iteration $iter$ M( $exp$ )C, $checkpoint$, n = 8388608, CUDALucas v1.0 I remember it has something to do with error rate.
I thought it was FFT length..? Divide it by 1024...

 2011-02-19, 17:14 #424 Karl M Johnson     Mar 2010 3·137 Posts Right. I remember, when I started working on a certain exponent, it used 4 million something n, then reported some error, and made n higher to this value.
2011-02-20, 13:24   #425
msft

Jul 2009
Tokyo

2·5·61 Posts

Thank you, kjaget & aaronhaviland
Attached Files
 CUDALucas.1.2.tar.bz2 (28.1 KB, 235 views)

 2011-02-20, 21:41 #426 drh     Jan 2011 Cincinnati, OH 22·52 Posts Compatible hardware? I picked up a system a while back, HP xw8400, has dual x5355 Quad CPU's, and a Quadro FX 3450 NVIDIA card. Can this system run CUDALucas? Thanks, Doug
 2011-02-20, 22:38 #427 Ken_g6     Jan 2005 Caught in a sieve 6138 Posts Wikipedia says, "Chip also used in GeForce 6800". Which means you can't even run CUDA, let alone CUDALucas.
2011-02-20, 22:46   #428
drh

Jan 2011
Cincinnati, OH

6416 Posts

Quote:
 Originally Posted by Ken_g6 Wikipedia says, "Chip also used in GeForce 6800". Which means you can't even run CUDA, let alone CUDALucas.
Thanks for the link. Guess this card is worthless for GIMPS :(

2011-02-23, 08:22   #429
alexhiggins732

Mar 2010
Brick, NJ

67 Posts

Quote:
 Originally Posted by kjaget I've done some minor porting to get this to build and run under Windows using MSVC 2008. There was nothing major that needed to change, just some fixes to get the code to use prototypes when building, use binary mode when opening checkpoint files, and a few #defines in the makefile. ... I've included both the source code and a 64-bit windows binary. I had no luck getting a 32-bit binary working, but I didn't try that hard since my system is 64-bit. My hope is that I didn't break anything for Linux builds, but without a system to test on I'm not sure how I was. In any case, there weren't that many changes so I don't think I did too much damage :) Attachment 6166 I don't have any specific benchmarks, but a 26,xxx,xxx doublecheck takes about 2.5 days on a GTX 275. I think that's in line with Linux performance, more or less. If you can give me instructions on how to get per-iteration timing I'll try it out. Enjoy, Kevin
I modified the makefile to get this to compile on x32 the test is running, but I think it is giving incorrect results.

Cuda info:
Quote:
 CUDA device info name GeForce GTX 260 compute capability 1.3 maximum threads per block 512 number of multiprocessors 27 (216 shader cores) clock rate 1242MHz CUDA version info binary compiled for CUDA 3.10 CUDA driver version 3.20 CUDA runtime version 3.10
This doesn't match the check files posted.
Quote:
 CUDALucas.1.2>cudalucas 18760031 -t Iteration 10000 M( 18760031 )C, 0xfffb000040009ffe, n = 1048576, CUDALucas v1.2
And with smaller exponents, I get numbers close to 0xfffffffffffffffd. Eg,

Quote:
 CUDALucas.1.2>cudalucas 216149 -t Iteration 10000 M( 216149 )C, 0xfffffffffffffffd, n = 524288, CUDALucas v1.2
Just to verify 1) the check files posted contain the residue after the first 100,000 iterations and 2) the final residue should math those on the Gimps Exponent status lookup page.

Quote:
 http://www.mersenne.org/report_expon...&B1=Get+status Verified LL36BF by "Unknown" Verified LL6BF by "John Sweeney" Verified LL6BF by "Marc-Etienne Vargenau" Verified LLD0AEFBD176BE36BF by "John Sweeney" Verified LLD0AEFBD176BE36BF by "Mike Gach" Verified LLD0AEFBD176BE36BF by "Ken Kriesel" Verified LLD0AEFBD176BE36BF by "Eivind Triel"
Any hints were to start looking in the code? I see the original versions ran on 32 bit linux, so has this been overhauled/optimized along the way to longer work on 32 bit flavors?

 Similar Threads Thread Thread Starter Forum Replies Last Post LaurV Data 131 2017-05-02 18:41 Brain GPU Computing 13 2016-02-19 15:53 Karl M Johnson GPU Computing 15 2015-10-13 04:44 fairsky GPU Computing 11 2013-11-03 02:08 Rodrigo GPU Computing 12 2012-03-07 23:20

All times are UTC. The time now is 18:53.

Fri Feb 26 18:53:38 UTC 2021 up 85 days, 15:04, 1 user, load averages: 2.07, 2.48, 2.64