Re: [MLUG] Anyone got a fast modern computer

[Stephen Montgomery-Smith <EMAIL:PROTECTED>]

Jack Smith wrote:
> On Fri, 2008-07-04 at 14:16 -0500, Stephen Montgomery-Smith wrote:
>> I have this multithreaded program written for Linux.  I would really 
>> like to try it on one of those quad core 64 bit systems.  Really just 
>> for timing tests.  I would expect it to take about 30 minutes.  Is 
>> anyone willing to indulge me?  I would send you a tar file, and you 
>> would do a simple "make; time ./the-program" and send me back the results.
>>
>> Thanks, Stephen
> 
> I use a Core 2 Quad Q6600 with 8 GB RAM at work but unfortunately it is
> running Windows XP x86_64 and the only thing I am _not_ allowed to do is
> install a different OS. And naturally, I need Linux to do my work as
> there is quite a bit of data analysis and programming involved and there
> is simply no substitute for most of the stuff I'd do with a short bash
> script. Yes, I know there are the gnuwin32 tools and cygwin and I've
> tried them but I need a 64-bit environment to work correctly. I thus
> have my trusty Debian installation on Virtualbox Open Source Edition. It
> works well but the OSE version only supports one CPU so I'd be no help.
> 
> I do have a pair of dual-core 64-bit systems at home though running
> Debian amd64. One is a socket 939 AMD Athlon 64 X2 4200+ (2.2 GHz/2x512
> KB L2) desktop with 4 GB RAM and the other is a Core 2 Duo U7500 laptop
> (1.067 GHz/2 MB unified L2) with 3 GB RAM. They are multi-core 64-bit
> processors...would it still help even though they are not quad-core
> units and the clock speeds are rather low by today's standards?

Well, let me put it out for anyone to try.  The program is at 
http://www.math.missouri.edu/~stephen/software/spherical.  This is what 
to do:

tar xvfz spherical-1.15.tar.gz
cd spherical-1.15
make

If the make doesn't work, delete the "-march=native" in CFLAGS in 
Makefile.  Edit timetest.sh so that NR_THREADS is equal to the number of 
processors/cores that you have.  Then run the test:

./timetest.sh

Then send me the results, which will be in record.txt, along with specs 
of your computer, and what value you used for NR_THREADS.

It is the 4th trial that really interests me.  On my dual core 1.85GHz 
laptop I can get it to run in 65 minutes using Linux Fedora version 9. 
I suspect it makes a lot of difference if you are using a rather recent 
gcc (4.2, 4.3 or higher), and that you are using a 64 bit version of the 
OS.  It goes a lot slower under FreeBSD on the same computer (like 3 
times slower).

I had a friend who ran it on his apple 8 core system.  It took 38 
minutes.  On Mike's machine, it took about 54 minutes.  He is using gcc 
4.1, so he wasn't able to use the "-march=native" option.

Finally, if you can get it to run on Windows, that would also interest 
me.  It is rather standard C, except it does use pthreads quite a bit.

Let me also say that this isn't simply some kind of benchmarking test, 
but is a program doing real numerical analysis, seeing how fibers align 
in fluids.

There is a sense in which I am merely curious as to how fast I can get 
this program to run, but I don't think my desire for this knowledge is 
entirely academic.  So anyone who wants to try it out would be doing me 
a nice favor.

Stephen

_______________________________________________
members mailing list
EMAIL:PROTECTED
http://mlug.missouri.edu/mailman/listinfo/members