On Wed, Aug 20, 2014 at 1:56 PM, Achim Gratz <strom...@nexgo.de> wrote: > Ole Tange <tange <at> gnu.org> writes: >> This was measured on my 4 hyperthreaded cpus (= 8 "cores") forced to >> run 1.7GHz, so the timings may be lower if you have a faster cpu or if >> you do not use hyperthreading. > > If you post the script, I could give you times for an Opteron 6328 on Cygwin > on Windows Server2012R2 (aside from clock frequency this is a very similar > CPU to yours: 4 piledriver modules with two cores each).
It is now in git: parallel/doc/boxplot-runtime Don't waste too much time on it, though (It has a lot of dependencies outside Cygwin). I will be surprised if the trend is not the same, and the actual timings (milliseconds) will likely just be a factor off. If it is run on a machine that does other things or which has a dynamic cpu frequency you will see a bigger spread. In my initial tries that spread was so big it completely obscured the trend. /Ole
