Linux CUDA Support


log in

Advanced search

Message boards : Number crunching : Linux CUDA Support

1 · 2 · Next
Author Message
Space45
Volunteer tester
Send message
Joined: 11 Dec 05
Posts: 5
Credit: 499,720
RAC: 0
United States
Message 1339191 - Posted: 17 Feb 2013, 18:11:58 UTC

Hi,

I'm not sure if this is the right forum, but has there any indication if/when CUDA support for Linux will be released for SETI@home?

Thanks,

Space45
____________

Profile Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,717,219
RAC: 1,273
United States
Message 1339194 - Posted: 17 Feb 2013, 19:06:48 UTC - in response to Message 1339191.
Last modified: 17 Feb 2013, 19:24:17 UTC

Hi,

I'm not sure if this is the right forum, but has there any indication if/when CUDA support for Linux will be released for SETI@home?

Thanks,

Space45


We've had CUDA for a while. At least if you have a 64bit Linux system.

You'll need to install the cuda app yourself into your projects/seti folder and use an appropriate app_info.xml file.

Here's the app you need. Others can help guide you with the appinfo.xml if the readme with the app is not enough.

Good luck. :-)



Now if you are asking when there will be stock Linux CUDA support, someone else will have to answer that.
____________
-Dave #2

3.2.0-33

Space45
Volunteer tester
Send message
Joined: 11 Dec 05
Posts: 5
Credit: 499,720
RAC: 0
United States
Message 1339261 - Posted: 17 Feb 2013, 23:35:35 UTC - in response to Message 1339194.

Hi,

I'm not sure if this is the right forum, but has there any indication if/when CUDA support for Linux will be released for SETI@home?

Thanks,

Space45


We've had CUDA for a while. At least if you have a 64bit Linux system.

You'll need to install the cuda app yourself into your projects/seti folder and use an appropriate app_info.xml file.

Here's the app you need. Others can help guide you with the appinfo.xml if the readme with the app is not enough.

Good luck. :-)



Now if you are asking when there will be stock Linux CUDA support, someone else will have to answer that.


Ok, thanks.

Does anyone know anything about stock Linux CUDA support?
____________

Terror Australis
Volunteer tester
Send message
Joined: 14 Feb 04
Posts: 1670
Credit: 204,056,383
RAC: 24,072
Australia
Message 1339514 - Posted: 20 Feb 2013, 3:09:15 UTC - in response to Message 1339261.

Ok, thanks.

Does anyone know anything about stock Linux CUDA support?

Yep, there isn't any, and AFAIK none is planned.

As with most things Linux it's DIY

This thread will give you an idea of what is involved plus some tips and tricks. It's a bit old but all that may be required is to update the file names in the app_info.xml file.

I found that to get CUDA to work on Linux, as well as the Nvidia driver I also had to download and install the CUDA Toolkit file.

What brand of Linux are you using ? The Nvidia driver and Toolkit should be available from your Distro repository.

T.A.

Profile Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,717,219
RAC: 1,273
United States
Message 1339598 - Posted: 20 Feb 2013, 6:40:57 UTC
Last modified: 20 Feb 2013, 6:41:18 UTC

Yea I'd use the repository verison of the NVIDIA driver, not the newer ones as man they will give you problems. I think I'm at 295-ish which is Ubuntu/Mint's "nvidia-current".
I have no issues with that driver and the app I linked to earlier with kernel 3.2.0
____________
-Dave #2

3.2.0-33

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3612
Credit: 48,142,408
RAC: 35,108
United States
Message 1339704 - Posted: 20 Feb 2013, 15:54:33 UTC

Now if we can get a Linux developer who can build Jason's CUDA app, it can be brought in as a stock app when SETI updates to v7.
____________

Profile Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,717,219
RAC: 1,273
United States
Message 1339731 - Posted: 20 Feb 2013, 17:43:18 UTC - in response to Message 1339704.

Now if we can get a Linux developer who can build Jason's CUDA app, it can be brought in as a stock app when SETI updates to v7.


I pointed the Debian Boinc Maintainers here, in case they are looking for something to do. :-)
____________
-Dave #2

3.2.0-33

Profile ML1
Volunteer tester
Send message
Joined: 25 Nov 01
Posts: 8345
Credit: 4,093,339
RAC: 1,153
United Kingdom
Message 1340007 - Posted: 21 Feb 2013, 20:44:32 UTC - in response to Message 1339704.
Last modified: 21 Feb 2013, 20:45:25 UTC

Now if we can get a Linux developer who can build Jason's CUDA app, it can be brought in as a stock app when SETI updates to v7.

If Terror-A speaks to me really nicely, I might do some of his assumed DIY in the FLOSS way such that everyone thereafter can have their own single click custom CUDA install for s@h for Boinc.

(You never know, I might even get round to that before Berkeley ramp up for the extra number crunching onslaught. Wait a moment... Is that why they haven't automated that bit? ;-) )


The rest of various Linux systems work very nicely with or without the DIY :-p

Happy freedom crunchin',
Martin
____________
See new freedom: Mageia4
Linux Voice See & try out your OS Freedom!
The Future is what We make IT (GPLv3)

Terror Australis
Volunteer tester
Send message
Joined: 14 Feb 04
Posts: 1670
Credit: 204,056,383
RAC: 24,072
Australia
Message 1340097 - Posted: 22 Feb 2013, 5:02:34 UTC - in response to Message 1340007.
Last modified: 22 Feb 2013, 5:18:18 UTC

If Terror-A speaks to me really nicely, I might do some of his assumed DIY in the FLOSS way such that everyone thereafter can have their own single click custom CUDA install for s@h for Boinc.

(You never know, I might even get round to that before Berkeley ramp up for the extra number crunching onslaught. Wait a moment... Is that why they haven't automated that bit? ;-) )


The rest of various Linux systems work very nicely with or without the DIY :-p

Martin, I always speak nicely to you. Even when I'm disagreeing with you. :)

In my experience there is always a bit of DIY involved with Linux. Take SAH CUDA for example, it's all 3rd party (without Crunch3r's and Aaron's efforts there wouldn't even be an app), and then after you install it you have to tweak "Niceness" etc. to make it crunch as effectively as "the other OS". :P

Not knocking Linux, just stating a fact.

I eagerly await the outcome of your FLOSSing

T.A.

Amauri
Volunteer tester
Send message
Joined: 18 May 08
Posts: 26
Credit: 465,203
RAC: 283
Brazil
Message 1340130 - Posted: 22 Feb 2013, 7:25:00 UTC - in response to Message 1340097.

In my experience there is always a bit of DIY involved with Linux. Take SAH CUDA for example, it's all 3rd party (without Crunch3r's and Aaron's efforts there wouldn't even be an app), and then after you install it you have to tweak "Niceness" etc. to make it crunch as effectively as "the other OS". :P

Not knocking Linux, just stating a fact.

I eagerly await the outcome of your FLOSSing

T.A.

Milkyway, Einstein, Poem, WCG... All of them have stock Linux GPU apps, why is so difficult to have one here? x41g is running fine, why not make it a stock app for v6 and v7?

Profile jason_geeProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4951
Credit: 72,892,314
RAC: 11,247
Australia
Message 1340143 - Posted: 22 Feb 2013, 8:06:08 UTC - in response to Message 1340130.
Last modified: 22 Feb 2013, 8:13:26 UTC

In my experience there is always a bit of DIY involved with Linux. Take SAH CUDA for example, it's all 3rd party (without Crunch3r's and Aaron's efforts there wouldn't even be an app), and then after you install it you have to tweak "Niceness" etc. to make it crunch as effectively as "the other OS". :P

Not knocking Linux, just stating a fact.

I eagerly await the outcome of your FLOSSing

T.A.

Milkyway, Einstein, Poem, WCG... All of them have stock Linux GPU apps, why is so difficult to have one here? x41g is running fine, why not make it a stock app for v6 and v7?


With v7 close, the project hasn't been focussed on updating v6 apps, despite many known issues having been resolved in more mature builds. With the project having limited resources, development was handed to us external volunteer devs after direct involvement from nVidia having gone by the wayside & to my knowledge AMD not having had any on the table.

The Cuda v7 x41zc apps are intended for stock on the switch to v7. For Windows this has been under third party public beta, test Mac builds have been made in Alpha, and it's likely someone will provide any minor patches necessary prior to v7 going live for one or more stock Linux builds.

So come v7 introduction, I'm expecting apps for all three major platforms will either come out at once (as stock), or fairly closely staggered as any last minute issues are found & fixed. IMO it's highly unlikely any of the now dated v6 GPU apps will be upgraded on main, or additional platforms added, until then.

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Space45
Volunteer tester
Send message
Joined: 11 Dec 05
Posts: 5
Credit: 499,720
RAC: 0
United States
Message 1340304 - Posted: 22 Feb 2013, 17:29:11 UTC

This is a feature that I'm very interested in. Is there any way I could help?
____________

Profile jason_geeProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4951
Credit: 72,892,314
RAC: 11,247
Australia
Message 1340315 - Posted: 22 Feb 2013, 17:38:24 UTC - in response to Message 1340304.

This is a feature that I'm very interested in. Is there any way I could help?


Quite possibly, if you're familiar with building applications under Linux ;). Check out & take a look at the sources [ https://setisvn.ssl.berkeley.edu/svn/branches/sah_v7_opt/Xbranch ], which will be right-ish for Pre-V7 builds, and require some modification for update to V7.

I've already been given some patches to apply for the Mac build, which will probably cover at least some of what's needed. Just finding time for that as I can, while the Mac build is in alpha testing an Crunchers Anonymous.
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Space45
Volunteer tester
Send message
Joined: 11 Dec 05
Posts: 5
Credit: 499,720
RAC: 0
United States
Message 1340356 - Posted: 22 Feb 2013, 19:18:10 UTC - in response to Message 1340315.

This is a feature that I'm very interested in. Is there any way I could help?


Quite possibly, if you're familiar with building applications under Linux ;). Check out & take a look at the sources [ https://setisvn.ssl.berkeley.edu/svn/branches/sah_v7_opt/Xbranch ], which will be right-ish for Pre-V7 builds, and require some modification for update to V7.

I've already been given some patches to apply for the Mac build, which will probably cover at least some of what's needed. Just finding time for that as I can, while the Mac build is in alpha testing an Crunchers Anonymous.


As a matter of fact, I am good at Linux development. :) I'll see what I can do with the sources. I presume that CUDA support is in the main branch?

However, my Linux machine only has an Intel GPU (Ivy Bridge). I've found an old GeForce 7300GS with 256MB of RAM. Will this work with BOINC and SETI@home?

Also, where could I find the Mac patches?

Thanks,

Space45
____________

Profile jason_geeProject donor
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4951
Credit: 72,892,314
RAC: 11,247
Australia
Message 1340375 - Posted: 22 Feb 2013, 20:06:23 UTC - in response to Message 1340356.

As a matter of fact, I am good at Linux development. :) I'll see what I can do with the sources. I presume that CUDA support is in the main branch?

However, my Linux machine only has an Intel GPU (Ivy Bridge). I've found an old GeForce 7300GS with 256MB of RAM. Will this work with BOINC and SETI@home?

Also, where could I find the Mac patches?

Thanks,

Space45


Great! :D Ok, 7300 is pre_Cuda I believe ( see https://developer.nvidia.com/cuda-gpus. Even though you;d be unable to run it, I suspect this wouldn't prevent you attempting a build, with the appropriate Cuda toolkit in place etc, & spotting more obvious makefile issues etc.

I have the preliminary Mac alpha patches here, which I was given privately. I'll be committing to that svn branch as soon as I can go through them carefully (hopefully in a day or so). I'm not expecting a lot of complications, as from what I can see the app works on at least some OSX variants.

Jason
____________
"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change."
Charles Darwin

Profile WilliamProject donor
Volunteer tester
Avatar
Send message
Joined: 14 Feb 13
Posts: 1580
Credit: 9,459,119
RAC: 8,587
Message 1340380 - Posted: 22 Feb 2013, 20:20:13 UTC - in response to Message 1340356.
Last modified: 22 Feb 2013, 20:21:57 UTC

This is a feature that I'm very interested in. Is there any way I could help?


Quite possibly, if you're familiar with building applications under Linux ;). Check out & take a look at the sources [ https://setisvn.ssl.berkeley.edu/svn/branches/sah_v7_opt/Xbranch ], which will be right-ish for Pre-V7 builds, and require some modification for update to V7.

I've already been given some patches to apply for the Mac build, which will probably cover at least some of what's needed. Just finding time for that as I can, while the Mac build is in alpha testing an Crunchers Anonymous.


As a matter of fact, I am good at Linux development. :) I'll see what I can do with the sources. I presume that CUDA support is in the main branch?


Great :)

However, my Linux machine only has an Intel GPU (Ivy Bridge). I've found an old GeForce 7300GS with 256MB of RAM. Will this work with BOINC and SETI@home?


256MiB would be enough on a Windows system, with a bit of fiddling - I've never seen a Linux memory footprint but as Linux is usually better with its resources it should work.

However, what doesn't work (as was pointed out to me, that's 'before my time') is the card itself. A GeForce 7300 is pre-cuda, so can't be used for this purpose. You'd need at least a 8400GS or so. there's a list here. They are pretty cheap to buy these days, even new, but they are not very cost effective regarding output/watts used.

Also, where could I find the Mac patches?


Not publicly available yet - you'll have to wait for Jason to commit them to Berkeley's SVN repository or handed privately.

William the Silent

edit: I see the boss beat me to it.
____________
A person who won't read has no advantage over one who can't read. (Mark Twain)

Profile Raistmer
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 16 Jun 01
Posts: 3386
Credit: 46,225,544
RAC: 7,755
Russia
Message 1340390 - Posted: 22 Feb 2013, 20:52:22 UTC - in response to Message 1340356.
Last modified: 22 Feb 2013, 20:52:35 UTC


However, my Linux machine only has an Intel GPU (Ivy Bridge). I've found an old GeForce 7300GS with 256MB of RAM. Will this work with BOINC and SETI@home?
,

Space45


Then you can try to checkout and build OpenCL apps instead.
They could work on Ivy Bridge GPU though I have troubles with MultiBeam for now. AstroPulse works under windows on Ivy Bridge GPU, it would be interesting how Linux port would behave.
____________

AlProject donor
Send message
Joined: 3 Apr 99
Posts: 481
Credit: 51,785,133
RAC: 23,555
United States
Message 1340478 - Posted: 24 Feb 2013, 18:28:44 UTC

I know resources are pretty sparse at this point, I wonder if we had another fund drive, and see how much we could raise to try and support one of our top notch Linux developers, maybe raising enough for them to be able to take a sabbatical from work for a few weeks/months, to knock out these versions for both 6 & 7? I know we probably have fund drive fatigue right now, but imagine how well things would go if they had that type of (hopefully) uninterrupted time to devote to one project. It would be spectacular. Not likely to happen, but would be interesting if someone with deep pockets who happened to be a Linux fan thought it was a great idea. Just tossin' it out there.. ;-)
____________

Terror Australis
Volunteer tester
Send message
Joined: 14 Feb 04
Posts: 1670
Credit: 204,056,383
RAC: 24,072
Australia
Message 1340499 - Posted: 24 Feb 2013, 19:10:22 UTC

While the code cutters are listening. :)

In my experiments with Linux I found that the current x41g linux app fell over when a large number of video cards were involved.

I built a machine using 7 GTS250 GPU's. Under Windows XP 32 bit I could quite happily feed the cards using an E7200 2 core processor (GPU crunching only). Under Windows the crunching times per unit of each card were normal for a GTS250. Processor usage was around 5% to 10% per core.

I then installed a clean build of Mageia 64 bit Linux on the same hardware. I installed BOINC with the 41g Linux app and a known good app_info.xml file and known good settings for "Niceness" and other parameters.

The system fell in a heap. Processor usage climbed to around 80% per core and crunching times for nearly tripled. I then swapped out the E7200 for a quad core CPU. This made a difference but CPU usage was still high and crunching times were still 150% to 200% above normal.

Through my investigations I came to the conclusion that the problem was due to congestion on the PCIE bus, but as I'm a "hardware man" I was unable to ascertain whether this was a problem with the app or a Linux problem.

As many of the top crunchers are now running 8 GPU's in the one machine, if the Linux app is to be competitive this is something that will need to be looked at.

Thanks to all for your efforts with the Linux app. I'm not knocking, just bringing a problem to your attention.

T.A.

Space45
Volunteer tester
Send message
Joined: 11 Dec 05
Posts: 5
Credit: 499,720
RAC: 0
United States
Message 1340560 - Posted: 24 Feb 2013, 21:52:16 UTC

I'm trying to compile the Xbranch linked earlier under Linux. However, I keep getting dependency errors about BOINC C++ headers. Since I'm using Ubuntu, I have installed the boinc-dev package.

However, I've noticed that the boinc header files are included with quotes. Am I supposed to have the BOINC header files in the local tree?
____________

1 · 2 · Next

Message boards : Number crunching : Linux CUDA Support

Copyright © 2014 University of California