MathGroup Archive 1995

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

Search the Archive

Re: Comparison of MMA on Various Machines

  • Subject: [mg2737] Re: Comparison of MMA on Various Machines
  • From: braunsha at EMBL-Heidelberg.DE (Gerhard Braunshausen)
  • Date: Mon, 11 Dec 1995 22:03:16 -0500
  • Approved: usenet@wri.com
  • Distribution: local
  • Newsgroups: wri.mathgroup
  • Organization: Wolfram Research, Inc.


----- Begin Included Message -----

In article <4a3el9$qnc at dragonfly.wri.com>, sinan at u.washington.edu  40Mhz40Mhz40Mhz(Sinan
Karasu) wrote:

> In article <49dp82$8a6 at dragonfly.wri.com>,
> Gerhard Braunshausen  <braunshausen at embl-heidelberg.de> wrote:
> >Could anyone offer an explanation for the vast differences
> >in speed particularly in comparing the PowerMac 9500 series
> >with the SparcStation 20 series?
> >
> >To my knowledge, the Sparcstations were supposed to be very
> >powerful numbercrunchers. So why do they perform so poorly
> >with Mathematica?
> >
> >Thanks for any hints
> 
> I tried Mathematica on an UltraSparc, the results were very good.
> They were somewhat better than PowerMac 9500.
> I suspect that Mathematica memory management is written for Mac.
> I also suspect that Wolfram Research uses gcc as the compiler
> for minimum hassle when supporting various platforms.
> 
> Maybe it is time for Sun to donate some compilers and an 
> UltraSparc machine to Wolfram Research.
> 

and ianc at wri.com wrote: 

Mathematica has not been yet been optimised for the Ultra 
Sparc. I am glad to hear that it nonetheless performed 
well on that platform. A future release of Mathematica 
will be optimised for Ultra Sparc.

----- End Included Message -----


I am very interested in the UltraSparc performance as we will be
purchasing one soon.

To clarify the performance rates announced by sinan at u.washington.edu 
it would be helpful to know which UltraSparc was tested:
Model 140 (147Mhz) or Model 170 or 170E (167Mhz)? 

SUN lend us a Model 140 for benchmarking which showed
a 3 times faster floating point speed on an FFT program we were using.

However, with applications involving mainly data shuffling back and forth
in memory, a factor 10 was obtained.

And in general, running  propriatary applications like khoros workspaces
we obtained an overall factor of 2.

Unfortunately, we did not get mathematica to run on it.
Does one need a separate license for getting mathematica to run on UltraSparc?
We have a distributed site license, which should run also on
UltraSparc, although not optimised for the 64bit architecture.
However, the Kernel wouldn't start up properly and the FrontEnd neither.

Does anyone have suggestions towards a solution?

Thanks to all

--gerhard



  • Prev by Date: Re: your mail
  • Next by Date: Re: Comparison of MMA on Various Machines
  • Previous by thread: Re: Comparison of MMA on Various Machines
  • Next by thread: Re: Comparison of MMA on Various Machines