Saturday, December 29, 2012

BOINC: CPU Time vs CPU Utilization

I mentioned I was going to run this experiment awhile back, and while I have been running it, I forgot exactly how my computer initially behaved, so switching a few settings back, I can now give a report on whether it is best to cut back on CPU Utilization or cut back on Percentage of CPU time BOINC uses.

Results:  I have found that in terms of both temperature control, and credits earned in a lengthy amount of time ( lets just say 1 month), that it is better to use all CPU's but throttle back the CPU time.

Hardware: This is on my Dell Laptop running  an Intel i3 Processor with 2 cup's but hyperthreading enabled causing BOINC to view it as 4. Ubuntu 12.04 is my OS.

First set up: Using 50% of the processors, I could only crunch two tasks at once, but it did enable me to crunch projects quicker, as I could increase my CPU utilization to nearly 50% before the temp became unstable ( I like to keep the temp of my CPU below 80C, though at 50% of CPU time it would often flirt above that line, especially if any other task on the computer was running).

Second set up: Using 100% of processors, so I can crunch four tasks at once, in theory this places any stable CPU temps, at a utilization of anywhere over 25% a slight victory.  I recalled 30% being stable in the past, which is why I was keen to try something else, as projects seemed to take forever to complete.  But now ( maybe because it is winter and the temps are in general cooler), I can get stable temperatures while using 35% of CPU time.

If your goal is to finish each task as quickly as possible the second set up is not for you.  But if the goal is to do as much crunching as possible, even if each task does take longer, the second set up seems ideal.  I have not quite thought of an exact way to measure these performance factors, but assuming 50% of CPU's at 50% utilization is identical to 100% of CPU's at 25% utilization, I in theory while running at 35% utilization have a 40% increase in performance over an extended length of time. 

The 40% is calculated, based on CPU time spent on each project.  Two cores, each with 1 project spending 1/2 of every second on a project amounts to 1 second of crunching each second.   While when crunching 4 projects at 35% means each second we advance the combined four projects by 1.4 seconds.  Thus 1.4 is 40% better than 1.

World Community Grid Signature