Re: Octave on Intel-macintosh: experiences?

[A S HODEL]

On May 23, 2006, at 3:44 PM, A S Hodel wrote:

I'm looking at purchasing one of the new intel-inside macintosh 

computers.  Does anyone on the list have any experience with their 

performance with Octave and/or other high performance computing 

software relative to G5 or dual G5 computers?

I wrote this question a bit over a week ago and received replies that encouraged me to proceed with the purchase of a macbook pro (currently the fastest mac available, but that ought to change this August).  I've briefly proofread the summary below and I hope that this will be useful to other readers of the address@hidden mailing list.

I have octave/octave-forge running now through the use of http://hpc.sourceforge.net/ .

Darwinports http://darwinports.opendarwin.org/ lists octave as an available package, but the installation fails due to lack of gcc3.3 compiler.  Off topic: while tracking this down I read that Apple is apparently not going to support gcc3.3 on intel macs.  However, the header failures that occured with gcc4.0 under my PowerPC macs now merely generates a header warning message with other C++ code I'm using in a simulation.  Thus, the use of gcc_select to use gcc3.x on intel-macs does not appear to be a requirement in some situations.

Darwinports easily installs gnuplot and imagemagick.

Fink has an info file for octave, but octave does not appear in the list of packages (fink list | grep octave yields nothing).  [Off topic] I was unable to install tetex under darwinports due to download problems (file not found), but was able to install tetex under fink.

Fortran It appears that there are currently no f77/g77/g95 compilers in fink or darwinports.  There must be some wizardry in package compilation, which explains why it's a binary distribution.  I  have both g95 and gfortran installed as intel-binaries (easily found with a google search for, e.g., g95 intel mac).

mkoctfile currently fails on one simulation that I'm running.  It's apparently looking for libraries in the darwinports directory structure.  This may be due to my use of binaries, darwinports, and fink all at the same time so that I could get close to the functionality of my original system.  Alternatively, it may be due to my use of fortran legacy code in the .oct file.  I haven't tried the oregonator .oct file example., which is strictly C++ code.

Conclusion: It appears that the HPC port is the sole option for Octave on intel-Macintosh at present.  Fortunately, it works well.  I have not yet been able to compile a .oct file, but that may be due to the presence of fortran code in the project. 

The beauty of the HPC port is that I had my machine up and running in just a few hours and ran some rough time comparisons on a set of monte carlo dispersions (C++ source called from Octave through python; don't ask why) and the 2.18GHz Macbook is roughly 2ce as fast as my home iMac G5 and 4-5 times as fast as my current desktop G4.  

intel mac: 45 minutes

iMac G5: 1 hr 15 minutes

iCube G4 (upgraded processor at 1.2 GHz): 2.5 hours or more.

Once the mkoctfile/fortran problem is resolved,  I can completely migrate to the new machine.  (That's a status report, not a request nor an offer of technical support ... I'm up to my neck as it is!)

Thank you to Joseph Slater and Jonathan Shea for their helpful input.

Yours,

A S Hodel

http://www.eng.auburn.edu/~hodelas