Speed of matlab simulation in mac os

Hi,
I have got two computers for my project. one is 2.9GHz P 4 running
windows XP 1Gb RAM. and another is dual processor (with each processor
running at 2.5 GHz), 4 GB RAM mac os. My problem is, when i run the
same matlab project on these two computers the speed of similation in
first PC is arround 2.8 times faster than the mac os. I have used the
same simulation parameters in both. I tried my best to understand why
the simulation in mac os is slower than in PC.I would appreciate if
somebody can solve my problem :How can I make the simulation run faster
in mac os ?
Rijan Bhandari

rijanbhandari@yahoo.com wrote:
> Hi,
> I have got two computers for my project. one is 2.9GHz P 4 running
> windows XP 1Gb RAM. and another is dual processor (with each processor
> running at 2.5 GHz), 4 GB RAM mac os. My problem is, when i run the
> same matlab project on these two computers the speed of similation in
> first PC is arround 2.8 times faster than the mac os. I have used the
> same simulation parameters in both. I tried my best to understand why
> the simulation in mac os is slower than in PC.I would appreciate if
> somebody can solve my problem :How can I make the simulation run faster
> in mac os ?
> Rijan Bhandari
>

www.macintouch.com has extensive discussions on performance among
Pentiums, PowerPC Macs, and the Dual-core Intel Macs.

Rob Lake

In article <1141678602.756831.42130@z34g2000cwc.googlegroups.c om>,
rijanbhandari@yahoo.com wrote:

> I tried my best to understand why
> the simulation in mac os is slower than in PC.I would appreciate if
> somebody can solve my problem :How can I make the simulation run faster
> in mac os ?

I would try to get a reduction - can you profile and figure out the
exact part of the simulation that is slower? If it is a specific
function or set of parameters to a function, perhaps we can suggest a
replacement.

In addition to other steps, contact MathWorks technical support.
Assuming that your code is decent, it should run at acceptable speed on
all platforms, or they should have warnings about slow things in the
docs with suggested replacements. Having that reduction I mentioned a
bit ago will help them answer the question.

And platforms do differ - a good LAPACK implementation on a G5 blows
away a similarly clocked Pentium, because the G5 has amazing memory
bandwidth, and a keen vectorized float unit. If the test centers on
integer performance, you will usually get better results on an x86
machine, as they tend to have better integer units. Add in the
excitement of different clock speeds, multi core vs multi chip, quality
of multithread implementations, and it becomes a bit murkier.

Thus, while the best MW tech support can say about certain problems is
'that's how it is', there should be a fair number of problems that they
can say something useful about.

Scott

--
Scott Ellsworth
scott@alodar.nospam.com
Java and database consulting for the life sciences

In article <scott-9CB29D.14365009032006@news.west.cox.net>,
Scott Ellsworth <scott@alodar.com> wrote:

> In article <1141678602.756831.42130@z34g2000cwc.googlegroups.c om>,
> rijanbhandari@yahoo.com wrote:
>
> > I tried my best to understand why
> > the simulation in mac os is slower than in PC.I would appreciate if
> > somebody can solve my problem :How can I make the simulation run faster
> > in mac os ?
>
> I would try to get a reduction - can you profile and figure out the
> exact part of the simulation that is slower? If it is a specific
> function or set of parameters to a function, perhaps we can suggest a
> replacement.
>
> In addition to other steps, contact MathWorks technical support.
> Assuming that your code is decent, it should run at acceptable speed on
> all platforms, or they should have warnings about slow things in the
> docs with suggested replacements. Having that reduction I mentioned a
> bit ago will help them answer the question.
>
> And platforms do differ - a good LAPACK implementation on a G5 blows
> away a similarly clocked Pentium, because the G5 has amazing memory
> bandwidth, and a keen vectorized float unit. If the test centers on
> integer performance, you will usually get better results on an x86
> machine, as they tend to have better integer units. Add in the
> excitement of different clock speeds, multi core vs multi chip, quality
> of multithread implementations, and it becomes a bit murkier.
>
> Thus, while the best MW tech support can say about certain problems is
> 'that's how it is', there should be a fair number of problems that they
> can say something useful about.
>
> Scott

Matlab uses a just in time compiler to increase execution speed, and the
quality of executed code probably varies across platforms.
--
See my photos at http://www.kjbeath.com.au/photos/index.html
Desktop pictures as well.