mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2015-11-11, 17:52   #1
Smokingenius
 
Nov 2015

2×5 Posts
Default Issues With Starting CUDALucas

Greetings,

First I would like to say I am new to the GIMPS project and to the GPU subthread so I apologize if this topic has been covered already.

I downloaded this morning the CUDALucas 4.2 for CUDA 4.2 and installed the 4.2 toolkit from Nvidia's website. Unfortunately it appears the toolkit might be missing something or I got the wrong hardware for this project because I get the following error when I attempt to run the CUDALucas exe (see attached screenshot). I am running a GTX 970M in my laptop that I am trying to get this working on. If this was successful I was hoping to get this working on a spare GTX 980 I have currently sitting on the sidelines but have room for in my primary PC.

I appreciate any troubleshooting assistance that can be provided.
Attached Thumbnails
Click image for larger version

Name:	CUDALucas_error 1.PNG
Views:	95
Size:	8.7 KB
ID:	13403  
Smokingenius is offline   Reply With Quote
Old 2015-11-11, 19:53   #2
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

24·33·23 Posts
Default

http://sourceforge.net/projects/cuda...s/CUDA%20Libs/

Download your choice of library packages, unzip, and place the appropriate (32 or 64 bit) versions of cudart*.dll and cufft*.dll in the folder with your CUDALucas executable.

EDIT: The current version, of CUDALucas is 2.05.1. Get it here, which is also at Sourceforge.
Also, I think you may need later versions of the libraries for the 900 series GPUs.

Last fiddled with by kladner on 2015-11-11 at 19:57
kladner is offline   Reply With Quote
Old 2015-11-11, 22:00   #3
Smokingenius
 
Nov 2015

128 Posts
Default

Greetings,

Thanks for your help! That did the trick. I updated also to the latest version 2.05.1, and used 6.5 libraries. I feel really embarrassed I missed those library files.

I did have to move to the latest CUDA install which was 7.5. Definitely works a lot better on Win 10 with a GTX 970 than that 4.2 CUDA code I was running.

Once again thanks for all the assistance.


I did have another question. Will the CUDALucas build partial saves just like GIMPS? Only reason I ask is so that it can be started and stopped if I need to turn off the machine, etc, or needed the cycles for a more intensive process.
Smokingenius is offline   Reply With Quote
Old 2015-11-12, 14:38   #4
Smokingenius
 
Nov 2015

2·5 Posts
Default

NVM, I got it figured out. The iteration save option was in the ini file, and I just had to turn it on.

Thanks for all the help, GPU is going strong. Working on getting the 980 in a spare rig I got and just use it for prime crunching.
Smokingenius is offline   Reply With Quote
Old 2015-11-12, 17:21   #5
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

993610 Posts
Default

Glad you got it sorted! More crunch power is always welcome!

Welcome to the project!

Last fiddled with by kladner on 2015-11-12 at 17:22
kladner is offline   Reply With Quote
Old 2015-11-12, 19:25   #6
Smokingenius
 
Nov 2015

2×5 Posts
Default

Thanks, I had previously helped out with Seti and Folding, but being a CE/EE those didn't appeal to me. Mersenne primes are definitely up my alley, even as a hobby, and when I am not using my computers for my own work and I can be crunching numbers for GIMPS. :)


I did have another configuration question for CUDALucas. Currently I have one stance running, and looking at the current utilization of the GPU, it isn't much. I have seen several posts about launching multiple instances of CUDALucas to maximize efficiency. Is there a way to have a single instance use multiple workers, like the bass GIMPS software, just by adding workers to the todo file or is the software does not have that capability yet?
Smokingenius is offline   Reply With Quote
Old 2015-11-13, 01:32   #7
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

21BB16 Posts
Default

Quote:
Originally Posted by Smokingenius View Post
Currently I have one stance running, and looking at the current utilization of the GPU, it isn't much.
What CPU/GPU (hardware) are you running? (no details, just say the name).
Are you getting anything (output) close to this list (scroll down for table) for your specific card?
Can you post some time_per_iteration and exponent so we have an idea if you are using it right?
Do you use a small (below 2000) value for screen/file output? (the parameter of the -c switch, in that case time is lost with screen/file output, use a value as 10000 for scree, or higher, and 200k or higher, usually 1M, for checkpoint file output).

Other things:
Can you make sure you have the last cudaLucas (2.05 or so).
Do you use "polite" function (and which value?) for details check the .ini file inside the zip, and adapt it for your needs.
The "multiple instances" does not apply, unless you either use a very old card, or a very old software version. New cudaLucas used adequately should "maximize" almost any (nVidia) card I know about, with ONE instance (one worker) on ONE exponent. The "multiple" thingies are done for every iteration (thousands of little guys inside of GPU struggling together in parallel to do that squaring/multiplication as fast as possible). That is where the speed comes from.

Last fiddled with by LaurV on 2015-11-13 at 01:41
LaurV is offline   Reply With Quote
Old 2015-11-13, 04:55   #8
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

24·33·23 Posts
Default

He's running a 970M, and hopes to get a 980 going. No info, yet, on the systems running the GPUs.
kladner is offline   Reply With Quote
Old 2015-11-13, 17:46   #9
Smokingenius
 
Nov 2015

10102 Posts
Default

Greetings,

So I do have the latest version of CUDALucas, 2.05.1, for x64 architecture. I am using a 970 GTX at the moment, on an ASUS ROG MOBO and an I7 Haswell.

Currently I am averaging 11.6 ms/iteration on the 970.

I did do some testing with multiple instances and I saw more than a double in performance loss. Ended up having around 30 ms/iteration when running 2. I was doing this with 2 different exponents.

Reason I was going this route was b/c in the normal GIMPS software, say the one I am running on the I7, I have 3 workers, each running an exponent. The graphics card obviously has thousands of little vector processors currently crunching numbers. Is there a way to have CUDALucas run more than 1 worker on the same card, or does each card its own worker? I believe it appears to be the later, per the information you posted.
Smokingenius is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Starting mprime at boot daxmick PrimeNet 18 2019-03-17 01:12
mfaktc not starting in Mac OSX bayanne GPU Computing 0 2014-05-10 14:38
Disk starting to go Chuck Hardware 8 2013-05-20 06:40
Starting new bases MrOzzy Conjectures 'R Us 104 2010-03-18 22:11
mprime starting spaz Software 9 2009-05-03 06:41

All times are UTC. The time now is 09:39.

Thu Aug 13 09:39:53 UTC 2020 up 6:15, 0 users, load averages: 1.18, 1.17, 1.23

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.