Changed cache cause Boinc stop using 2nd GPU
15 Jan 2015 8:53:21 UTC
Topic 197929
(moderation:
Have been running E@H on a dual AMD 7870 Windows 8.1 setup for several months. No problems. 2 work units per GPU and 2 free CPU cores.
Changed the cache or work buffer from a setting of .75 and 1.5 to 3.00 and 3.5 and GPU 1 stopped working on work units. Since then I have tried:
Adjusting work buffer up and down.
changing "On multiprocessor systems, use at most.".
Changing "Activity" to all three settings.
removing app_config.xml and cc_config.xml.
changing most options and back.
Installing newest version of Boinc. Had 7.2.42 now 7.4.36.
rebooting
Turning off Crossfire and back on.
Played a game that supports Crossfire, both GPUs were used as usual.
AMD driver 14.12
Windows 8.1
app_config.xml
einsteinbinary_BRP4G
0.5
0.5
cc_config.xml
0
0
0
http://boinc.berkeley.edu/download.php?xml=1
http://boinc.berkeley.edu/download.php
0
0
0
0
300
10
2000
8
2
0
0
0
-1
http://www.google.com/
0
0
0
0
0
10.000000
1
0
30
0
0
0.000000
0
0
0
1
0
0
0
1 on 4th line from bottom.
On web site settings are all OK even tried changing to another location changing from Arecibo to Perseus.
In Event log:
Quote:
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] ATI instance 0; 0.500000 pending for p2030.20140630.G51.28-00.85.N.b1s0g0.00000_2864_4
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] ATI instance 0; 0.500000 pending for p2030.20140114.G200.83-00.90.C.b2s0g0.00000_3472_2
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] ATI instance 0: confirming 0.500000 instance for p2030.20140630.G51.28-00.85.N.b1s0g0.00000_2864_4
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] ATI instance 0: confirming 0.500000 instance for p2030.20140114.G200.83-00.90.C.b2s0g0.00000_3472_2
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] Assigning 0.500000 of ATI free instance 1 to p2030.20141110.G64.91+00.40.N.b4s0g0.00000_1824_3
1/15/2015 2:38:50 AM | Einstein@Home | [coproc] Assigning 0.500000 of ATI instance 1 to p2030.20140114.G200.59-01.36.C.b6s0g0.00000_1088_2
The tasks assigned to GPU 1 say "Ready to Start".
Event log shows both GPUs.
Again everything was fine E@H using both GPUs, 2 work units each, until I changed the work buffer. Games still use both GPUs. Hopefully I have remembered everything I have tried. At a loss of what to do next except abort, uninstall., delete directories and re-install.
Ideas?
Changed cache cause Boinc stop using 2nd GPU
)
I think the problem is with your app_config.xml file
You say 'einsteinbinary_BRP4G' but have 'Binary Radio Pulsar Search (Perseus Arm Survey) v1.39 (BRP5-opencl-ati)' units in your cache. Boinc has no clue what to do, you are telling it to run the BRP4G units, but it only has BRP5 units! My suggestion is to go to the webpage and specify two units at once and get rid of the app_config.xml file, restart Boinc and see if it works then.
Your cc_config.xml file is a whole other story, you seem to be telling to do the defaults in most cases.
Sorry I was not clear. The
)
Sorry I was not clear.
The app_config.xml is what I was using before trouble started. I am not using it now. I changed applications to see if it would make a difference and since there did not seem to be many Arecibo available at the time.
WHen I got up all GPUs are working. Every work unit is marked "Running High Priority" even the WCG tasks. Due dates run from the 19-29th. But at least it is all working.
Thanks for reply.
Cheers!
RE: Sorry I was not
)
Ahhh that makes sense then....
As for running at 'high priority' you might want to back off your cache size just a little bit and see if it stops doing that. Boinc thinks that if it doesn't run the units at 'high priority' it won't finish them before the deadlines. I see you are running some OLD version of the Boinc software, you may try upgrading and see if the new versions of Boinc do better on your machines. You can get any version, including the new betas, here:
http://boinc.berkeley.edu/dl/?C=M;O=D
RE: RE: Sorry I was not
)
Be aware that the newest BOINCs - including the currently-recommended v7.4.36 - do still run tasks in 'Earliest Deadline First' mode if the scheduling algorithm finds it necessary, but don't shout it from the rooftops as 'high priority' in the task list.
And it is doing it again.
)
And it is doing it again. This time I made no changes or adjustments.
It keeps repeating:
Tried rebooting and pausing and resuming. Instance 0, device 0, only one working. Instance, device 1 idle.
EDIT:
If I suspend WCG E@H uses both GPUs. Had set for WCG to use 90% CPU. Tried 80 and 20. No good. E@H will use 2 gpus only when WCG suspended. Again all has been fine for some weeks and I made no changes. Confused.
EDIT:
It seems to have something to do with how I am trying to limit the number of CPU cores on WCG. It is working now with 4 CPU core for WCG and 2 CPU cores for E@H along with 2 E@H WUs per GPU. Both GPUs now working. Still confused.
Seems ok now. Used a
)
Seems ok now. Used a app_config file to control number of cores WCG can use.