Re: speed of octave

[Eric S. Carlson]

I noticed the thread also on the python newsgroup. There was one additional
issue noted there and that concerns the distribution of MATLAB with the MKL
libraries. For many years it was true that Atlas outperformed INTEL's stuff,
but the latest MKL stuff seems to be quite good. The python folks noted that
MKL could be linked to python, and I presume that MKL could also be linked
to octave. Since I have done neither and have not verified this as a source
of the performence gain, I will simply pass the info on.

>From a philosophical perspective, you can only hope that for $2000+ and
probably $1000/year per copy that there would be some benefit.

I have done many benchmarks on comparable machines, and it is rare on
anything besides looping that MATLAB significantly outperforms octave, and
it is often true that octave wins. JIT may be great for taking care of some
things (like numerical loops that should have been vectorized to begin
with), but not all. For example, I wrote programs with loops to read and
write ascii stl files in MATLAB, and later versions without the loops were
100's of times faster.

While not as good as JIT (yet), python looping is faster than octave's as
well. Some time in the future  I may try to dig into the how each system
does the looping. For now, I will happily vectorize, teach my students to
vectorize, and write oct files for those rare things that can not be
vectorized.

Cheers, Eric Carlson  
-- 
View this message in context: 
http://www.nabble.com/speed-of-octave-tf4394980.html#a12653754
Sent from the Octave - General mailing list archive at Nabble.com.