MathGroup Archive 2003

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Speed improvements in Mathematica 5 ??

  • To: mathgroup at smc.vnet.net
  • Subject: [mg42424] Re: Speed improvements in Mathematica 5 ??
  • From: Bill Rowe <listuser at earthlink.net>
  • Date: Sun, 6 Jul 2003 06:57:05 -0400 (EDT)
  • Sender: owner-wri-mathgroup at wolfram.com

On 7/5/03 at 3:10 AM, kowald at molgen.mpg.de (Axel Kowald) wrote:

> I just downloaded the demoversion of Mathematica 5 and tried to find the 
> advertised speed improvements. I got a benchmark notebook from 
> http://smc.vnet.net/timings40.html and run it on my old Mathematica 4.1 and the 
> new Mathematica 5 on the same machine (1.8GHz, win2k, 768MB RAM).

> To my surprise Mathematica 5 was only 1.6% better than the 4.1 version !!

> Any comments ?

Several. First, you might want to look at the benchmarking notebook that can be found at http://www.scientificweb.de/mathstef3.html

This notebook takes longer to run giving improved resolution of benchmarks. Using this notebook I see an overall improvement by a factor of 1.27 between version 4.2 znd version 5.0 running on a 800 MHz Apple PowerBook with Mac OS 10.2.6 and 1 GB of RAM.

But this number doesn't really tell the full story. No single numbe can really tell you how much improvement you will see unless you are only doing one particular type of computation.

For example using the benchmarks above I see the time to compute an FFT over 4194304 random values went from 13 seconds to 4.3 seconds. I see an even greater improvement in the timing for polynomial regression with the time going from 40.9 seconds to .56 seconds.

OTOH, there is a significant slow down in the computation of Laplace transforms with the time increasing from 12.4 seconds to 111.3 seconds. The point is whether you see version 5.0 as significantly faster than previous versions depends on what you will be doing. 

Clearly, these benchmarks show significant improvement in some of the numerics, much more than what was indicated by Wolfram. OTOH, some symbolic computations have slowed.

One final comment. Obviously, these results were obtained on my machine. What may not be obvious is the relative ranking might be quite different in a different computing environment.


  • Prev by Date: Re: Matrix functions
  • Next by Date: Re: NestWhile and FindRoot[Integrate]
  • Previous by thread: Speed improvements in Mathematica 5 ??
  • Next by thread: Re: Re: Speed improvements in Mathematica 5 ??