Forum > Discussion Forum
CPU work fetch seems to be inhibited by GPU one
Raistmer:
My host downloaded much GPU work for SETI beta, then it took some ATI GPU work for SETI main and then it decided to download ~10 tasks for CPU only.
No CPU tasks running in panic mode, transfers are absent, it's quad-core host... Why not asking for CPU work (when manually updated it refuses to ask for work!).
BOINC 6.10.58 (Jason's modded one).
What flags should I enable to bug-report this issue ?
Claggy:
Haven't you been following this thread?:
Shorties estimate up from three minutes to six hours after today's outage!
and this thread:
Request issues
Claggy
Raistmer:
--- Quote from: Claggy on 17 Sep 2011, 02:04:43 pm ---Haven't you been following this thread?:
Shorties estimate up from three minutes to six hours after today's outage!
and this thread:
Request issues
Claggy
--- End quote ---
Yes, I'm quite out of sync now - hard weeks were...
Could someone summarize those discussions in few words, please?
There was another David's brilliant server update? Most what I need to know - what can I do on my side to force work fetches go ?
Looks like this host runs w/o CPU work few days already #$%^^#$# >:(
Jason G:
--- Quote from: Raistmer on 17 Sep 2011, 02:02:08 pm ---What flags should I enable to bug-report this issue ?
--- End quote ---
For our internal boinc, enable <dcf_debug> . What you'll probably find is that normal Project DCF is completely 'wacked out', and aDCF in the modded version is at least periodically shoved hard against the lower stop (0.02) inhibiting work fetch due to there still being overestimates.
There is nothing much to do but wait for David to fix the server side estimates, though I am having some success building a GPU cache, with further enhancement to aDCF (chopping out server side dependencies & making convergence more rapid.) I'm currenlty talking to others about it in GPUUG, and if it takes too long for David to fix it, I might have to consider making the improved modded variant available to some to smooth David's transitions.
Jason
Raistmer:
--- Quote from: Jason G on 17 Sep 2011, 02:08:44 pm ---
--- Quote from: Raistmer on 17 Sep 2011, 02:02:08 pm ---What flags should I enable to bug-report this issue ?
--- End quote ---
For our internal boinc, enable <dcf_debug> . What you'll probably find is that normal Project DCF is completely 'wacked out', and aDCF in the modded version is at least periodically shoved hard against the lower stop (0.02) inhibiting work fetch due to there still being overestimates.
There is nothing much to do but wait for David to fix the server side estimates, though I am having some success building a GPU cache, with further enhancement to aDCF (chopping out server side dependencies & making convergence more rapid.) I'm currenlty talking to others about it in GPUUG, and if it takes too long for David to fix it, I might have to consider making the improved modded variant available to some to smooth David's transitions.
Jason
--- End quote ---
Well, as I understand it's known issue already... OK, will try to run re-schedule....
Navigation
[0] Message Index
[#] Next page
Go to full version