I don't have a GPU - should I resign from Binary Radio Pulsar Search ?
23 Jun 2012 15:46:35 UTC
Topic 196417
(moderation:
Comparing a run time of typical 31 hours at a Linux notebook (in power saving mode of course) to about 800 seconds of a nVidea GF 580 I' wondering whether it is worth to focus on non-GPU-optimised projects ?
I don't have a GPU - should I resign from Binary Radio Pulsar Se
)
Hello,
This will be a matter of preference. BRP4 tasks generally take much longer to process on CPUs compared to other applications like Gamma-ray. My personal preference would be to run Gamma-ray and S6LV tasks via CPU and to run BRP4 tasks on GPUs only.
Jeroen
From the project's point of
)
From the project's point of view, E@H still gets about 20..25% of all BRP4 results from CPUs, so CPUs make quite a significant contribution. Also because CPUs are considered the gold standard wrt accuracy of the results, it's good to have a significant population of CPUs participating to help detect problems on the GPUs if they should arise.
Cheers
HB
P.S.: As for running the BOINC apps on a laptop in power saving mode, this raises an interesting point: is this really more efficient? Let's say that the BRP4 app will take twice as long to finish in this power saving mode. To be more efficient, the notebook would have to run on less than half the power consumption, otherwise you would over all consume more electricity to crunch a unit. For the CPU alone, this will usually be true because modern CPUs do not just lower the clock frequency, but also the voltage in power saving mode, and power consumption scales with the square of the voltage. But there are of course other parts in the notebook that consume power and that are not affected so much by the power saving mode).
Of course there are other good reasons to run BOINC in power saving mode on a notebook (less noise, less heat dissipated), I just wanted to note that from a power consumption point of view, it is worth taking an actual measurement of the power consumption at the wall socket to check the efficiency of the power saving mode operation.
Cheers
HB
RE: (less noise, less heat
)
Well - I should better say, I run my Gentoo Linux with cpu governor "ondemand".
By this I don't waste electrical power nor CPU cycles - BOINC is simply the app which gets the spare CPU power.
And the fan noise is the main argument for it, b/c I often run the notebook in the docking station.
In Albert@home I have two
)
In Albert@home I have two BRP3SSE results on my CPU running Linux. One is validated and the other is waiting a wingman using BRP3cuda32.
Tullio
RE: RE: (less noise, less
)
Whether this works as intended depends on the configuration of your Linux box. I'm not sure how your Gentoo handles this, but several Linux distributions use settings that ignore processes running in "nice" mode, as the CPU jobs of BOINC. This means that the CPU clock frequency will not get increased when BOINC jobs are running, and on-demand basically has no effect for the BOINC CPU jobs, they would run as fast as in the "unconditional" power saving mode.
But if the Linux OS is configured to consider "nice" jobs as ordinary jobs that should increase CPU freuency, BOINC jobs will run faster even in on-demand mode , again with the drawback of loud fans and lots of heat.
See e.g. http://www.archivum.info/boinc_dev@ssl.berkeley.edu/2009-01/00227/Re-%28boinc_dev%29-Power-saving-vs-Boinc-vs-user-control-amp-confusion.html
Cheers
HB
tfoerste@n22 ~ $ sudo cat
)
export PATH=/bin:/sbin:/usr/bin:/usr/sbin
#
# power saving
#
F=/sys/devices/system/cpu/cpufreq/ondemand/ignore_nice_load
if [[ ! -f $F ]]; then
modprobe -v acpi_cpufreq
fi
echo 1 > $F || echo "$F doesn't exist !"
:-)
Nice! HB
)
Nice!
HB