I've just discovered that in the computing preferences for s...@h (and others), there's a new GPU parameter:
Suspend GPU work while computer is in use? Enforced by version 6.7+ That was set to "yes". I've just now set it instead to "no". ...And the s...@h link appears to be down :-( I'll see what happens to any further GPU WUs. Happy crunchin', Martin Martin wrote: > Martin wrote: >> Martin wrote: >>> Martin wrote: >>>> David Anderson wrote: >>>>> Please set <cpu_sched_debug> and send message log >>>> That's going to be a lot of output... >>>> >>>> One brief snapshot is attached showing a s...@h task restart by the >>>> scheduler. >>> I've posted a 907kBytes log on: >>> >>> http://www.ml1.co.uk/misc/cpu_sched_debug02.txt >>> >>> >>> As it happens, s...@h is showing "waiting to run" at the moment. >> The s...@h CUDA has settled into a steady repeated cycle of periods of >> compute and then no-compute. >> >> For anyone interested, there's today's log uploaded to: >> >> http://www.ml1.co.uk/misc/cpu_sched_debug03.txt >> >> Note: 4 MBytes of log! > > Has that been of any interest/help? > > > One other thing to notice is that the scheduler is rather busy > 'rescheduling'. > > One thought is that the *rescheduling* triggers should be less frequent. > Either that or some mechanism of including hysteresis may well be needed > to avoid scheduling oscillation for marginal conditions/decisions with > larger numbers CPU cores (GPUs, processing devices)... > > Or is that already included? :-) -- -------------------- Martin Lomas m_boincdev ml1 co uk.ddSPAM.dd -------------------- _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
