Mac, NVidia GPU and Seti@home


log in

Advanced search

Message boards : Number crunching : Mac, NVidia GPU and Seti@home

1 · 2 · 3 · 4 . . . 5 · Next
Author Message
enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1329118 - Posted: 19 Jan 2013, 15:09:55 UTC

Is there any update on getting Macs to run CUDA application for Seti@home? CUDA FAQ suggests anytime soon (and has for several years.... ). Brief attempt at compiling the seti_cuda branch under SVN/branches suggests that it is targeting OSX 10.4 or earlier, and has Windows specific code (log results in client/cuda appears to be hardcoded to "Windows.h" and there's a lot of conflicting strXXX functions). Many of the helper utilities are much older releases.

Is there a more recent code branch for CUDA and is it viable for Mac OSX?

Profile Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,691,439
RAC: 1,289
United States
Message 1329123 - Posted: 19 Jan 2013, 15:47:55 UTC

Linux64 users have Cuda, perhaps there is something else you could attempt at compiling, seeing as how Macs are *nix systems.

But to answer your question, who knows. ;-)
____________
-Dave #2

3.2.0-33

Josef W. Segur
Volunteer developer
Volunteer tester
Send message
Joined: 30 Oct 99
Posts: 4206
Credit: 1,030,619
RAC: 266
United States
Message 1329157 - Posted: 19 Jan 2013, 18:27:59 UTC - in response to Message 1329118.
Last modified: 19 Jan 2013, 18:28:56 UTC

Is there any update on getting Macs to run CUDA application for Seti@home? CUDA FAQ suggests anytime soon (and has for several years.... ). Brief attempt at compiling the seti_cuda branch under SVN/branches suggests that it is targeting OSX 10.4 or earlier, and has Windows specific code (log results in client/cuda appears to be hardcoded to "Windows.h" and there's a lot of conflicting strXXX functions). Many of the helper utilities are much older releases.

Is there a more recent code branch for CUDA and is it viable for Mac OSX?

There is more recent code for CUDA builds in /branches/sah_v7_opt/Xbranch (the Windows CUDA apps under test at SETI Beta are built from that code). AFAIK nobody has attempted to update that code for Macs.
Joe

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1329966 - Posted: 21 Jan 2013, 22:57:26 UTC - in response to Message 1329157.
Last modified: 21 Jan 2013, 22:58:23 UTC

Thanks! After some effort I have seti@home compiled for NVidia/Cuda on Mac OSX 10.8.2. Appears to be processing WU happily though haven't compared time against CPU.

One thing I noticed is that Seti and Einstein will only run one at a time. Does a single NVidia card (8800GT in this case) only represent a single GPU as far as BOINC processing goes? Thought GPUs have multiple cores but BOINB only appears to run single task requiring GPU.

Profile ivan
Volunteer tester
Avatar
Send message
Joined: 5 Mar 01
Posts: 579
Credit: 131,508,836
RAC: 112,954
United Kingdom
Message 1329991 - Posted: 21 Jan 2013, 23:45:22 UTC - in response to Message 1329966.

Thanks! After some effort I have seti@home compiled for NVidia/Cuda on Mac OSX 10.8.2. Appears to be processing WU happily though haven't compared time against CPU.

One thing I noticed is that Seti and Einstein will only run one at a time. Does a single NVidia card (8800GT in this case) only represent a single GPU as far as BOINC processing goes? Thought GPUs have multiple cores but BOINB only appears to run single task requiring GPU.

While GPUs have multiple processing units, they get their power by processing data in parallel, not by processing multiple jobs. Some of the newer Nvidia cards (Fermi-class and later, IIRC) can process more than one job at a time (depending on memory capacity), but I don't believe your 8800GT can.
____________

TBar
Volunteer tester
Send message
Joined: 22 May 99
Posts: 1177
Credit: 41,646,747
RAC: 111,376
United States
Message 1329998 - Posted: 22 Jan 2013, 0:02:15 UTC - in response to Message 1329966.

Thanks! After some effort I have seti@home compiled for NVidia/Cuda on Mac OSX 10.8.2. Appears to be processing WU happily though haven't compared time against CPU.

One thing I noticed is that Seti and Einstein will only run one at a time. Does a single NVidia card (8800GT in this case) only represent a single GPU as far as BOINC processing goes? Thought GPUs have multiple cores but BOINB only appears to run single task requiring GPU.

You do understand no one else has been able to do that. It would be great if it also works in Lion & SL. I can test that for ya... This is big news if it works consistently and results are verified. Thanks for your work.

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1330005 - Posted: 22 Jan 2013, 0:32:47 UTC - in response to Message 1329991.

Considering upgrading to something more recent (460/560 appear to work on ML) but wanted to understand whether it would help. Sounds like it might, even if just to process faster.

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1330009 - Posted: 22 Jan 2013, 0:36:15 UTC - in response to Message 1329998.

It appears to work on ML but would be happy to work with someone to prove it is actually doing right thing.....there doesn't appear to be a test suite as part of Makefile/package. Had to "hack" code to get to work as a lot of Windows specific pieces and #ifdef's but will see what I can pull together.

No longer have SL or L (only one machine) and did optimise for ML. May be possible to take it back to these versions.

TBar
Volunteer tester
Send message
Joined: 22 May 99
Posts: 1177
Credit: 41,646,747
RAC: 111,376
United States
Message 1330013 - Posted: 22 Jan 2013, 0:46:44 UTC - in response to Message 1330005.

Considering upgrading to something more recent (460/560 appear to work on ML) but wanted to understand whether it would help. Sounds like it might, even if just to process faster.

If your Mac 8800 is like my Mac 8800, it is already running at around 95% load with one task. All running multiple copies accomplish is to force the card to run at a higher load and take around twice or thrice the amount of time to accomplish the task. Since your at 95% load with one task, raising it to 100% will see very little, if any, gain in time. That's also why I'm not interested in forcing my AMD 6850 to run more than one task. I'm able to have it run at 90-95% load with one task and running 2 would have very little benefits, beside having it run hotter and noisier...

What SETI files are you using with the Mac? 609, 610? Have any verified?

TBar
Volunteer tester
Send message
Joined: 22 May 99
Posts: 1177
Credit: 41,646,747
RAC: 111,376
United States
Message 1330022 - Posted: 22 Jan 2013, 1:35:27 UTC - in response to Message 1330009.

It appears to work on ML but would be happy to work with someone to prove it is actually doing right thing.....there doesn't appear to be a test suite as part of Makefile/package. Had to "hack" code to get to work as a lot of Windows specific pieces and #ifdef's but will see what I can pull together.

No longer have SL or L (only one machine) and did optimise for ML. May be possible to take it back to these versions.

If we could decide on a transfer method, I'll see if I can get it to work. I have a SETI account in Snow Leopard and also have Lion installed on another partition. I have an 8800 & AMD 6850 working in OSX. Do you have a .Mac account? You can also transfer files in Skype, but, it might take me a while to remember how to use Skype...

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1330026 - Posted: 22 Jan 2013, 2:06:57 UTC - in response to Message 1330013.

Using locally compiled branch of source, modified to work on Mac. seti_enhanced 6.0.3 from Lunatics from Seti SVN. Two units have just verified and will monitor for others.

How are you finding out GPU load stats on Mac? GPU-Z doesn't work and CUDA-Z is only showing GFlops performance.

TBar
Volunteer tester
Send message
Joined: 22 May 99
Posts: 1177
Credit: 41,646,747
RAC: 111,376
United States
Message 1330027 - Posted: 22 Jan 2013, 2:26:56 UTC - in response to Message 1330026.
Last modified: 22 Jan 2013, 3:18:02 UTC

I'm not running SETI on my OSX systems at present. I'm running XP on an early 2008 MacPro. Those numbers are from the same hardware in XP. I switched to running the XP I installed back in 2008, on my Mac, when I realized I could do 10x more work using my existing hardware. I would happily go back to OSX on my Mac if I could use the GPUs. The 8800 is kinda lightweight now, the 6850 is much better at SETI tasks. Someone needs to get AMD AstroPulse working on a Mac.

BTW, it looks like we have the same Mac, here's mine in OSX Computer 6796479. Back in 2008 I couldn't see ever needing 8 cores, hence, I chose the 4 core version. Now that I need 8 cores, it's actually time for a new Mac.

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3595
Credit: 47,354,447
RAC: 1,779
United States
Message 1330044 - Posted: 22 Jan 2013, 3:13:17 UTC - in response to Message 1330009.

It appears to work on ML but would be happy to work with someone to prove it is actually doing right thing.....there doesn't appear to be a test suite as part of Makefile/package. Had to "hack" code to get to work as a lot of Windows specific pieces and #ifdef's but will see what I can pull together.

No longer have SL or L (only one machine) and did optimise for ML. May be possible to take it back to these versions.


There is a test suite for OSX, just not there.

Urs made a bench program that I ran on my iMac before it died.
http://www.arkayn.us/forum/index.php?topic=7.0
____________

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4920
Credit: 72,626,488
RAC: 2,357
Australia
Message 1330183 - Posted: 22 Jan 2013, 19:21:30 UTC - in response to Message 1330009.
Last modified: 22 Jan 2013, 19:30:29 UTC

It appears to work on ML but would be happy to work with someone to prove it is actually doing right thing.....there doesn't appear to be a test suite as part of Makefile/package. Had to "hack" code to get to work as a lot of Windows specific pieces and #ifdef's but will see what I can pull together.

No longer have SL or L (only one machine) and did optimise for ML. May be possible to take it back to these versions.


Great News!. If you could send me either diff patches or modified source files via email ( contact at jgopt dot org ) I'll happily look them over for V7 multibeam operational compliance, consult with you on any tweaks & proper crediting in the documentation, then apply them straight into Berkeley for a probable stock beta project release. My own schedule is pretty tight at the moment, limiting me to mostly weekends.Also I don't have a timeline on V7 release from Eric, though I believe another step forward like this will make things easier/better when Eric comes knocking.

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

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1330282 - Posted: 23 Jan 2013, 2:33:02 UTC - in response to Message 1330044.

Running benchmark (and after restarting machine to clear some memory issues) appears to be 300% faster than the optimized CPU code on an 8800GT. Will post Jason some diff when I get a chance (but these were heavily hacked so probably better to go through will a little more rigour...)

enewman
Volunteer tester
Send message
Joined: 27 Jun 01
Posts: 15
Credit: 6,255,912
RAC: 127
United States
Message 1330335 - Posted: 23 Jan 2013, 10:58:17 UTC - in response to Message 1330183.

Jason

Code is in your inbox.

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3595
Credit: 47,354,447
RAC: 1,779
United States
Message 1330404 - Posted: 23 Jan 2013, 16:46:12 UTC - in response to Message 1330335.

And the build in on my site.
____________

Profile jason_gee
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 24 Nov 06
Posts: 4920
Credit: 72,626,488
RAC: 2,357
Australia
Message 1330422 - Posted: 23 Jan 2013, 17:28:45 UTC - in response to Message 1330335.

Jason

Code is in your inbox.


Thanks! Should be able to go through that & update Berkeley's sources on the weekend.
____________
"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 Ex
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 1,691,439
RAC: 1,289
United States
Message 1330444 - Posted: 23 Jan 2013, 18:35:38 UTC

Awesome work guys! Thanks to enewman for this, and thanks to our awesome devs for making sure not to let anything slip by.
____________
-Dave #2

3.2.0-33

B-Man
Volunteer tester
Send message
Joined: 11 Feb 01
Posts: 253
Credit: 147,366
RAC: 0
United States
Message 1335168 - Posted: 6 Feb 2013, 15:57:26 UTC

I also must say congratulation. I am looking forward to this.
____________

1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Number crunching : Mac, NVidia GPU and Seti@home

Copyright © 2014 University of California