Mathematica 9 is now available
Services & Resources / Wolfram Forums
-----
 /
MathGroup Archive
2005
*January
*February
*March
*April
*May
*June
*July
*August
*September
*October
*November
*December
*Archive Index
*Ask about this page
*Print this page
*Give us feedback
*Sign up for the Wolfram Insider

MathGroup Archive 2005

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

Search the Archive

Re: Re: Does Mathematica 5 use multiple processors on SPARC?

  • To: mathgroup at smc.vnet.net
  • Subject: [mg55241] Re: [mg55168] Re: Does Mathematica 5 use multiple processors on SPARC?
  • From: Igor Antonio <igora at wolf-ram.com>
  • Date: Thu, 17 Mar 2005 03:29:57 -0500 (EST)
  • Organization: Wolfram Research, Inc.
  • References: <d13j25$qjl$1@smc.vnet.net> <200503150521.AAA09693@smc.vnet.net>
  • Reply-to: igora at wolf-ram.com
  • Sender: owner-wri-mathgroup at wolfram.com

Dave wrote:
> Actually, more to the point, can it use multiple-CPUs on SPARC on 
> version 5.1? I just installed a copy from the univeristy today, and see 
> it is 5.1

In general, no.  The MathKernel is single-threaded. However, operations 
that use BLAS libraries will take advantage of multiple processors since 
they are multi-threaded (with the exception of the ones we already 
discussed).

> I just tried multipling two 5000x5000 matrices of random numbers 
> together and found it did not seem to be using more than one CPU, which 
> is should be able to do on this problem, as it can be done in parallel 
> quite easily.

That's where Parallel Computing Toolkit comes in...

> 
> Interestingly it run out of memory when trying to multiply two 10,000 x 
> 10,000 matrices, which given they should only take up 800MB each, and I 
> have 4GB RAM plus a few GB swap, I'm surprised it could not manage this. 
> There was nothing much on the computer apart from Mathematica running, 
> so why it run out of RAM is beyond me.
> 

That depends on various things.  What version of Mathematica are you 
running?  Mathematica 5.1.0 and earlier had a 2GB limit for packed 
arrays.  With 10K x 10K matrix, you will need roughly 800MB each, as you 
said.  You will need enough memory to store 3 instances of them (M1, M2, 
and the result of M1.M2).  Plus the BLAS libraries will need scratch 
space to do their computations.  I'm not exactly sure about what you're 
doing, but if you're unpacking the arrays for some computation, then 
each array would take up 10000^2 * 24 bytes ( ~ 2.2GB).

Try it in Mathematica 5.1.1

-- 


Igor Antonio
Wolfram Research, Inc.
http://www.wolfram.com

To email me personally, remove the dash.


  • Prev by Date: Re: Sum
  • Next by Date: Re: numerical, bessel
  • Previous by thread: Re: Does Mathematica 5 use multiple processors on SPARC?
  • Next by thread: Copy As menu command