MATLAB Benchmarks - R2007b (and possibly R2008a)

By xterm at Tue, Mar 18 2008 11:47am |

I have a Mac Pro (Aug 2006) running at 2.0 GHz, 6 GB RAM, 256 MB nVidia GeForce 7300GT. I noticed that my MATLAB benchmarks on R2007b are terribly slow, specifically in the sparse computation and the 3D.

With -nojvm and running from X11, bench(5) yields:

0.1645 0.4358 0.3819 1.3560 0.7507 1.8111
0.1393 0.4144 0.2934 1.3461 0.6431 1.7227
0.1487 0.4233 0.2901 1.3510 0.5671 1.7996
0.1491 0.4287 0.2924 1.3819 0.5670 1.7477
0.1438 0.4299 0.2914 1.3667 0.6366 1.7147

for an average of:

0.1491 0.4264 0.3099 1.3603 0.6329 1.7592.

The 4th and 6th columns are the surprisingly slow columns in question. Running with jvm doesn't really change these numbers considerably (unscientific claim). What could possibly be going wrong with this machine? Is MATLAB jut not optimized even in R2007b for Intel Macs? Does the situation improve in R2008a?

Thanks.

‹ error trying to exec `as': execvp: No such file or directory Anyone using TextMate for sequence data? ›
Wed, 2008-04-02 01:57 — verga

Indeed, very slow...

I also noted that "bench" gives ridiculous slow performance, even in version R2008a.
This is in contrast with Mathematica benchmark and compiled programs, when compared to other systems.

System:
Mac OS X 10.5.2; MacBookPro, Intel Core 2 Duo 2.2 GHz; GeForce 8600M GT

BENCH results Matlab R2008a
LU FFT ODE Sparse 2D 3D
0.1855 0.3078 0.3506 0.4764 1.2291 1.2417

System:
Mac OS X 10.5.2; MacPro, 2 x 3 GHz quad core Intel Xeon; ATY Radeon X1900

BENCH results Matlab R2008a
0.1461 0.2536 0.2561 0.8539 0.9214 1.3298

Alberto Verga
Aix-Marseille Université

Wed, 2008-04-02 16:36 — wcuk

yup

I also have a mac pro and benched it in R2007b with very poor results. I think part of it is that Matlab for the mac is 32 bit. I get a much faster bench if I run R2007b in 64bit Ubuntu under vmware fusion. This is embarrassing considering the Linux virtual machine is using 2 of the 8 cores.

Thu, 2008-04-03 13:13 — CrimsonS

I just wanted to say that

I just wanted to say that you can't compare the results of bench run on two different versions of MATLAB. It even says so in the help info:
"This benchmark is intended to compare performance of one particular version of MATLAB on different machines. It does not offer direct comparisons between different versions of MATLAB. The tasks and problem sizes change from version to version."

Though I'll admit that MATLAB is rather slow, not to mention painfully buggy, on OS X... But, what can you expect from an application which only costs a few tens of thousands of dollars :-P

Thu, 2008-04-10 18:11 — xterm

R2008A Results

Good reminder regarding versions. I just installed R2008a with the following results, on my MBP 2.2 GHz, OS X 10.5.2, 4 GB RAM, no jvm:

0.1479 0.3103 0.2915 0.4456 0.7183 1.6583

These are an avg. over 5 trials.