ABP2 CUDA applications

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043163
RAC: 586162

Ok, I'll forward this to the

Ok, I'll forward this to the devs as this really looks strange.

Thanks for reporting this,
HB

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20603702
RAC: 0

Thank you Bikeman, Looking

Thank you Bikeman,

Looking forward to some updates.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043163
RAC: 586162

RE: Thank you

Message 96305 in response to message 96304

Quote:

Thank you Bikeman,

Looking forward to some updates.

I just got a note from Bruce Allen, pointing out that the new CUDA app versions will perform a check of CUDA memory resources before starting, and if there isn't enough on-board memory available, the app start will be retried a couple of minutes later (earlier version would just error out in this situation).

So theoretically this could be the case. I wonder whether there is anything to this effect in the stderr.txt file (where the logging messages are started) in the respective slot directory (one of the /Library/Application\ Support/BOINC\ Data/slots/ directories).

So, is the ABP1 task never started at all (state "ready to start") or is it started but doesn't progress ? (state "waiting to run" etc) ?

CU
HB

Elphidieus
Elphidieus
Joined: 20 Feb 05
Posts: 245
Credit: 20603702
RAC: 0

RE: I just got a note

Message 96306 in response to message 96305

Quote:

I just got a note from Bruce Allen, pointing out that the new CUDA app versions will perform a check of CUDA memory resources before starting, and if there isn't enough on-board memory available, the app start will be retried a couple of minutes later (earlier version would just error out in this situation).

So theoretically this could be the case. I wonder whether there is anything to this effect in the stderr.txt file (where the logging messages are started) in the respective slot directory (one of the /Library/Application\ Support/BOINC\ Data/slots/ directories).

So, is the ABP1 task never started at all (state "ready to start") or is it started but doesn't progress ? (state "waiting to run" etc) ?

CU
HB

You meant ABP2...?

They never started. It just says "Ready to start". They were never in a state of "Waiting to run". No error messages from BOINC. The ABP2 app was never shown active on Activity Monitor. The GT120 was only shown using 2 - 5% graphics usage from Hardware Monitor driving a 24-inch display.

GPV67
GPV67
Joined: 26 Mar 05
Posts: 11
Credit: 19231568
RAC: 4089

RE: RE: Thank you

Message 96307 in response to message 96305

Quote:
Quote:

Thank you Bikeman,

Looking forward to some updates.

I just got a note from Bruce Allen, pointing out that the new CUDA app versions will perform a check of CUDA memory resources before starting, and if there isn't enough on-board memory available, the app start will be retried a couple of minutes later (earlier version would just error out in this situation).

So theoretically this could be the case. I wonder whether there is anything to this effect in the stderr.txt file (where the logging messages are started) in the respective slot directory (one of the /Library/Application\ Support/BOINC\ Data/slots/ directories).

So, is the ABP1 task never started at all (state "ready to start") or is it started but doesn't progress ? (state "waiting to run" etc) ?

CU
HB


Hi,
Thanks for the update.

Tasks are "ready to start" and there are no others CUDA apps running.
No entry for Einstein ABP2 in slots directories.
It looks like STSP 5.07 application units (ABP2) are never handled by the scheduler.
During that time, CPU units (S5R6) are successfully computed.

GPV

 

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1079
Credit: 341280
RAC: 0

RE: Tasks are "ready to

Message 96308 in response to message 96307

Quote:

Tasks are "ready to start" and there are no others CUDA apps running.
No entry for Einstein ABP2 in slots directories.
It looks like STSP 5.07 application units (ABP2) are never handled by the scheduler.
During that time, CPU units (S5R6) are successfully computed.

GPV


Did you try a reboot to clear graphics memory?

Can you otherwise reduce GPU menory usage (desktop resolution...)?

Gruß,
Gundolf

Computer sind nicht alles im Leben. (Kleiner Scherz)

GPV67
GPV67
Joined: 26 Mar 05
Posts: 11
Credit: 19231568
RAC: 4089

RE: RE: Tasks are "ready

Message 96309 in response to message 96308

Quote:
Quote:

Tasks are "ready to start" and there are no others CUDA apps running.
No entry for Einstein ABP2 in slots directories.
It looks like STSP 5.07 application units (ABP2) are never handled by the scheduler.
During that time, CPU units (S5R6) are successfully computed.

GPV


Did you try a reboot to clear graphics memory?

Can you otherwise reduce GPU menory usage (desktop resolution...)?

Gruß,
Gundolf


Hi,
Yes, computer rebooted.
Even with very low resolution (640*480) and all others projects suspended, no ABP2 unit started.
In slots directories, no trace of attempt to start ABP2 unit.
What is the amount of Graphical memory needed by 5.07 application? (GeForce 8800GT is a 512MB card)

 

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 686043163
RAC: 586162

Hi all, EDIT: it requires

Hi all,

EDIT: it requires around 430 MB free graphics mem or some such. So it will depend on the video mode whether this will work on a 512 MB board.

Anyway, let's try to get a bit more data on this one.

If you put the following text into a file cc_config.xml in the BOINC working directory (/Library/Application Support Data/BOINC on Macs) and restart BOINC. There should be some more log messages explaining hopefully why boinc does what it does (or fails to do).

EDIT: corrected type:
[pre]


1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

[/pre]

Standard Disclaimer: PLEASE DO NOT post the full output here as it may contain confidential account information (cookies, account identifiers...whatever). But I'm sure you will notice a smoking gun if there is one in the output.

CU
H-B

GPV67
GPV67
Joined: 26 Mar 05
Posts: 11
Credit: 19231568
RAC: 4089

Hi Bikeman, I got this error

Hi Bikeman,
I got this error message after created cc_config.xml file by copying and paste your lines:
Dim 17 jan 22:31:32 2010 Unrecognized tag in cc_config.xml:
Dim 17 jan 22:31:32 2010 Missing end tag in cc_config.xml

GPV

 

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2139
Credit: 2752710342
RAC: 1460874

Blimy! The whole kitchen

Message 96312 in response to message 96310

Blimy! The whole kitchen sink.....

You'll get masses of output with all of those set. Don't run it like that for more than ten minutes, then disable all except the first three by changing the 1 to 0 (those are ones and zeros). Then restart BOINC, and take stdoutdae.txt away into a dark corner to see if you can find a smoking gun.

If you do find something, just re-enable the single debug log that showed it, and take another 10-minutes worth (or until it shows again), and just post the essentiial snippet here. We can always ask for a longer run if the first proves inconclusive.

And if all else fails, there are even more logging flags we can show you to try - the client configuration page hasn't been updated with all the new v6.10 goodies.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.