Modified SETI MB CUDA + opt AP package for full GPU utilization

Message boards : Number crunching : Modified SETI MB CUDA + opt AP package for full GPU utilization
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 11 · 12 · 13 · 14 · 15 · 16 · 17 . . . 25 · Next

AuthorMessage
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851566 - Posted: 10 Jan 2009, 4:40:37 UTC - in response to Message 851564.  
Last modified: 10 Jan 2009, 4:51:49 UTC

http://img339.imageshack.us/my.php?image=setiye2.png
yeah i actually suspended all AP tasks. i think the MB CUDA is working fine, except that my cpu is not being used at all. i may just add a wcg project, so that my cpu has something to do, but i would really prefer to have cpu+gpu crunching seti
thanks for reply


Hi,

You should resume the AP tasks, they won't start as long as theey are suspended. Can you show a shot of Boinc with the CUDA running and the AP task not suspened, so we can see what else is going on if anything.
ID: 851566 · Report as offensive
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851569 - Posted: 10 Jan 2009, 4:44:05 UTC - in response to Message 851537.  
Last modified: 10 Jan 2009, 5:00:46 UTC

OK.
Now I have running 4 threads of seti (one off 0.04% CUDA) and 2 of rosetta, so is this correct?

No, try setting your ncpus back to 5 in your cc_config.xml file, you shouldn't have 5 cpu tasks running.


CUDA one is really fast though..... shoudnt there be 2 or does it uuse sli?

It appears Boinc may be using both your cards as one, or is only seeing one, which has been known to happen. I'm not clear myself why, though I do know there have been others that have gotten two gpu cards to run at one time doing two different tasks, but I don't think it had anything to do with the cc_config.xml file. What does it show in the log in Boinc Manger when you start it up, regarding the CUDA detected?

EDIT:This thread is one of the ones I read where they had gotten two gpu cards to work at one point, but didn't go into too much details, other than loading the driver, how they got it to work.
ID: 851569 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851603 - Posted: 10 Jan 2009, 8:00:45 UTC - in response to Message 851569.  
Last modified: 10 Jan 2009, 8:01:45 UTC

There is new version of package available here:
http://lunatics.kwsn.net/gpu-crunching/modified-seti-mb-cuda-opt-ap-package-for-full-gpu-utilize.msg12576.html#msg12576

It incorporates triplet overflow bug-fix provided by CUDA MB developer.
Should give invalid overflow on lesser amount of tasks now.
Other bugs still remain, debugging in progress, please, be patient.
ID: 851603 · Report as offensive
e6600

Send message
Joined: 25 Aug 06
Posts: 4
Credit: 514,792
RAC: 0
United States
Message 851676 - Posted: 10 Jan 2009, 14:04:23 UTC - in response to Message 851566.  

http://img339.imageshack.us/my.php?image=setiye2.png
yeah i actually suspended all AP tasks. i think the MB CUDA is working fine, except that my cpu is not being used at all. i may just add a wcg project, so that my cpu has something to do, but i would really prefer to have cpu+gpu crunching seti
thanks for reply


Hi,

You should resume the AP tasks, they won't start as long as theey are suspended. Can you show a shot of Boinc with the CUDA running and the AP task not suspened, so we can see what else is going on if anything.

ok im sorry, i was trying to run seti MB GPU + MB CPU at same time.
thanks
ID: 851676 · Report as offensive
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851708 - Posted: 10 Jan 2009, 15:38:04 UTC - in response to Message 851676.  

ok im sorry, i was trying to run seti MB GPU + MB CPU at same time.
thanks

I'm sorry I didn't understand that's what you meant. It not possible at this time to run cpu MB and gpu MB at same time. They hope to have this available in the future.
ID: 851708 · Report as offensive
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851709 - Posted: 10 Jan 2009, 15:39:32 UTC

Question, what does

major = 1
minor = 0

or

major = 1
minor = 1

in the stderr stand for?
ID: 851709 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851714 - Posted: 10 Jan 2009, 15:49:26 UTC - in response to Message 851709.  

Question, what does

major = 1
minor = 0

or

major = 1
minor = 1

in the stderr stand for?


Think revision version of CUDA (so called computational ability or smth like this - check correct spelling on nVidia site).
In general 1.0 - 8800 cards. Lack of some very useful abilities. All with 1.1 and up have mostly stuff that needed fore real work.
ID: 851714 · Report as offensive
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851715 - Posted: 10 Jan 2009, 15:50:47 UTC - in response to Message 851714.  

Question, what does

major = 1
minor = 0

or

major = 1
minor = 1

in the stderr stand for?


Think revision version of CUDA (so called computational ability or smth like this - check correct spelling on nVidia site).
In general 1.0 - 8800 cards. Lack of some very useful abilities. All with 1.1 and up have mostly stuff that needed fore real work.

that's what I was afraid of.
ID: 851715 · Report as offensive
Profile Crunch3r
Volunteer tester
Avatar

Send message
Joined: 15 Apr 99
Posts: 1546
Credit: 3,438,823
RAC: 0
Germany
Message 851724 - Posted: 10 Jan 2009, 16:01:36 UTC - in response to Message 851715.  

Question, what does

major = 1
minor = 0

or

major = 1
minor = 1

in the stderr stand for?


Think revision version of CUDA (so called computational ability or smth like this - check correct spelling on nVidia site).
In general 1.0 - 8800 cards. Lack of some very useful abilities. All with 1.1 and up have mostly stuff that needed fore real work.

that's what I was afraid of.


to be more precise ...
1.0 G80-GPUs (Geforce 8800 GTS 320/640, GTX und Ultra)
1.1 Geforce 8800 GS,Geforce 8800 GT, Geforce 8800 GTS 512/1024 MiB and 9xxx,gtx 2xx




Join BOINC United now!
ID: 851724 · Report as offensive
Profile Byron S Goodgame
Volunteer tester
Avatar

Send message
Joined: 16 Jan 06
Posts: 1145
Credit: 3,936,993
RAC: 0
United States
Message 851725 - Posted: 10 Jan 2009, 16:04:57 UTC - in response to Message 851724.  



to be more precise ...
1.0 G80-GPUs (Geforce 8800 GTS 320/640, GTX und Ultra)
1.1 Geforce 8800 GS,Geforce 8800 GT, Geforce 8800 GTS 512/1024 MiB and 9xxx,gtx 2xx

Thanks, there was a list around here somewhere but I can never find it when I wanted it.
ID: 851725 · Report as offensive
Profile Geek@Play
Volunteer tester
Avatar

Send message
Joined: 31 Jul 01
Posts: 2467
Credit: 86,146,931
RAC: 0
United States
Message 851727 - Posted: 10 Jan 2009, 16:10:05 UTC


Compute capability can be checkedd with a utility CUDA-Z.

I read somewhere that a compute capability of 1.1 is required to use Cuda on Boinc/Seti MultiBeam.
Boinc....Boinc....Boinc....Boinc....
ID: 851727 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851733 - Posted: 10 Jan 2009, 16:15:10 UTC - in response to Message 851715.  
Last modified: 10 Jan 2009, 16:21:36 UTC


that's what I was afraid of.


(I was wrong, 8800GT - compute capability 1.1)

Probably 1.1 required indeed.

Number of
Multiprocessors
Compute
Capability
GeForce GTX 280 30 1.3
GeForce GTX 260 24 1.3
GeForce 9800 GX2 2x16 1.1
GeForce 9800 GTX, 9800 GTX+, 8800 GTS 512 16 1.1
GeForce 8800 Ultra, 8800 GTX 16 1.0
GeForce 9800 GT, 8800 GT, 9800M GTX 14 1.1
GeForce 9600 GSO, 8800 GS, 8800M GTX,
9800M GT
12 1.1
GeForce 8800 GTS 12 1.0
GeForce 9600 GT, 8800M GTS, 9800M GTS 8 1.1
GeForce 9700M GT 6 1.1
GeForce 9500 GT, 8600 GTS, 8600 GT,
9700M GT, 9650M GS, 9600M GT, 9600M GS,
9500M GS, 8700M GT, 8600M GT, 8600M GS
4 1.1
GeForce 8500 GT, 8400 GS, 8400M GT,
9500M G, 9300M G, 8400M GS, 9400 mGPU,
9300 mGPU, 8300 mGPU, 8200 mGPU,
8100 mGPU
2 1.1
ID: 851733 · Report as offensive
Profile Björn

Send message
Joined: 29 Jul 01
Posts: 22
Credit: 1,058,001
RAC: 0
Germany
Message 851781 - Posted: 10 Jan 2009, 17:37:35 UTC - in response to Message 851603.  

There is new version of package available here:
http://lunatics.kwsn.net/gpu-crunching/modified-seti-mb-cuda-opt-ap-package-for-full-gpu-utilize.msg12576.html#msg12576

It incorporates triplet overflow bug-fix provided by CUDA MB developer.
Should give invalid overflow on lesser amount of tasks now.
Other bugs still remain, debugging in progress, please, be patient.


I got a parse error when starting up Boinc. It works though, just want to let you know that you have "<app_version>" right before "</app_info>".

By the way, I happen to get no new tasks when using your app_info. Do you know why this is?
ID: 851781 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851795 - Posted: 10 Jan 2009, 18:19:20 UTC - in response to Message 851781.  
Last modified: 10 Jan 2009, 18:25:32 UTC

There is new version of package available here:
http://lunatics.kwsn.net/gpu-crunching/modified-seti-mb-cuda-opt-ap-package-for-full-gpu-utilize.msg12576.html#msg12576

It incorporates triplet overflow bug-fix provided by CUDA MB developer.
Should give invalid overflow on lesser amount of tasks now.
Other bugs still remain, debugging in progress, please, be patient.


I got a parse error when starting up Boinc. It works though, just want to let you know that you have "<app_version>" right before "</app_info>".

By the way, I happen to get no new tasks when using your app_info. Do you know why this is?


Hm, probably just because that error in app_info.xml... Will check it.

ADDON:
Checked. That parse error is really harmless. But I added 607 app_version to differentiate from 606 - maybe that's the reason ?.. Anyway, V4a of package added with repaired app_info.xml.
Feel free to update. And thanks for spotting that glitch.
ID: 851795 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 851802 - Posted: 10 Jan 2009, 18:39:58 UTC - in response to Message 851795.  
Last modified: 10 Jan 2009, 18:48:40 UTC

Was just starting to say I got the parse error too but You beat me to it . Thanks for the great work Raistmer.


Ok, I got the V4a and now I'm getting this error 1/10/2009 1:45:28 PM|SETI@home|[error] State file error: bad application name


Hope it is harmless too.


PROUD MEMBER OF Team Starfire World BOINC
ID: 851802 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851821 - Posted: 10 Jan 2009, 19:27:08 UTC - in response to Message 851802.  
Last modified: 10 Jan 2009, 19:50:39 UTC

Was just starting to say I got the parse error too but You beat me to it . Thanks for the great work Raistmer.


Ok, I got the V4a and now I'm getting this error 1/10/2009 1:45:28 PM|SETI@home|[error] State file error: bad application name


Hope it is harmless too.


OMG, will check again :)

BTW, there is speedup from new build:

MB_r396mod_CUDA.exe -verb -st / 03no08aa.27103.267279.4.11.253.wu :
Started at : 20:10:21.323
Ended at : 20:24:16.329
834.974 secs Elapsed
107.251 secs CPU time

setiathome_6.06_windows_intelx86__cuda.exe -verb -st / 03no08aa.27103.267279.4.11.253.wu :
Started at : 20:24:16.485
Ended at : 20:39:32.579
916.063 secs Elapsed
119.294 secs CPU time

EDIT: SETI@home|[error] State file error: bad application name may be worse than prev error.
Please, check what version number your SETI MB tasks have - 6.06 or 6.07 ?

ADDON: try to add these lines to app_info.xml
(stop BOINC before editing)

<version_num>607</version_num>
<plan_class>cuda</plan_class>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.040000</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>MB_r396mod_CUDA.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
</app_version>
<app_version>
ID: 851821 · Report as offensive
Maik

Send message
Joined: 15 May 99
Posts: 163
Credit: 9,208,555
RAC: 0
Germany
Message 851870 - Posted: 10 Jan 2009, 21:09:43 UTC - in response to Message 851802.  
Last modified: 10 Jan 2009, 21:12:09 UTC

Ok, I got the V4a and now I'm getting this error 1/10/2009 1:45:28 PM|SETI@home|[error] State file error: bad application name


There is an additional <app_version> (NOT </app_version>) at the end of the app_info.xml in the v4 package...
ID: 851870 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851880 - Posted: 10 Jan 2009, 21:24:06 UTC - in response to Message 851870.  

Well, Ok.
I will reissue to version 4b :) Will return 607 back as on my host that app_info didn't prevent downloading of new tasks. They were branded as 605 though - some enigmatic event for me cause there is 606 and 607 app versions in app_info.xml.
But they downloaded and in processing now.
So V4b come :)
ID: 851880 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851887 - Posted: 10 Jan 2009, 21:31:50 UTC - in response to Message 851880.  

Ok, 606 and 607 versions in app_info.xml from V4a were incomplete.
So try v4b - repaired one.
ID: 851887 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 851925 - Posted: 10 Jan 2009, 23:15:35 UTC - in response to Message 851887.  

For all who started to recive "invalid device function " CUDA errors with new build:
Please, upgrade your video driver at least to 180.48 (or more recent).
178.xx versions not work with new build whereas 180.48 works just as before.
ID: 851925 · Report as offensive
Previous · 1 . . . 11 · 12 · 13 · 14 · 15 · 16 · 17 . . . 25 · Next

Message boards : Number crunching : Modified SETI MB CUDA + opt AP package for full GPU utilization


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.