Thursday, September 30, 2010

5 seconds to get million digits of pi - easy benchmark

Here is a great way to benchmark a new system, or pick the leader from a pool of desktop systems.

Okay so you are planning to do some mathematical exploration and you have at your disposal a laptop, and a couple of desktops.

Which one to use?

You can be very scientific about this (more later), or you can just press a few keys on each and pick the winner.

  • apt-get install python-gmpy
  • cd /usr/share/doc/python-gmpy/examples/test
  • python test_large.py
5 seconds for the desktop Amd 7850 which is in a year old Asus P2-M3A3200
25 seconds for the laptop, an Intel T5550 based Dell and is 2 years old
My other desktop came inbetween those two, just slightly beating the laptop.



The winning system is running Debian Squeeze 64 bit and Python 2.6.6:



I mentioned earlier about this being a quick but unscientific way of choosing a suitable machine for you mathematical task.

You will perhaps have a particular task in mind. If it is linear algebra based then you might want to run some sample tests using octave, or scilab, or sage.

Here is a useful link to benchmarks that include some linear algebra/matrix operations:

Whilst looking around for benchmark results, I came across this 2007 post, that includes a 300x300 matrix timing test, and here are my results:



(optional) more technical specs for the Asus:
  • Chipset is AMD 780G
  • Socket is AM2+
  • Onboard Graphics are Radeon HD 3200
  • Memory installed is PC2-6400 DDR2-800 

( The above is by no means cutting edge, but for a £200 desktop system, as the benchmark shows, results are not too shabby )

(optional) Matrix commands in text format if you would like to repeat that 300x300 test in your own program:

M = rand(300,300);
t=0.25; T=2;
tic; inv(M) * (expm(-t*M) - expm(-T*M)); toc;

( I hope I made it clear from the link earlier that those commands appeared on a forum at wilmott.com
In particular blondie and spacemonkey posters used those commands for timing tests )

( The install commands such as apt-get which I have used to install gmpy are specific to my platform GNU/Linux. I am unable to provide help with other platforms, and suggest you might want to find a more suitable howto, that gives useful benchmarks including an install guide for your platform. )

1 comment:

  1. Octave does include some standard benchmarks, if you wish to delve a bit deeper, into a particular aspect of performance.

    If you really want to stress your system, then maybe install scilab and run this:
    http://sourceforge.net/projects/distance/

    "Euclidean distance transform algorithms used in Image Processing and computational geometry"

    ReplyDelete