Inverting a 50 by 50 random matrix, aka "the JKT 
benchmark", has long been used in J.  An example
from my notebooks:

no. 5968433 filed 17.35.28  thu 13 feb 1992
from eem
to   kei rhui
cc   dhs jkt

The timing David mentioned was for Joey's benchmark
on the NeXT computer.  The timing I reported early
last year was 101.141, for NeXT using the 68030.  
The timing obtained last night was for J4.1 on
a NeXT using a 68040, and was 19.67.
...

On this machine (2.8 GHz Celeron) just now:

   6!:2 '%. 50 50 [EMAIL PROTECTED] 1000'
0.00307246



----- Original Message -----
From: Joey K Tuttle <[EMAIL PROTECTED]>
Date: Monday, January 15, 2007 9:42 am
Subject: Re: [Jgeneral] Invert max  and LAPACK

> At 12:25  -0500 2007/01/15, John Randall wrote:
> >Joey K Tuttle wrote:
> >>  On my G5 iMac -
> >>
> >>      x=: 1000 1000 [EMAIL PROTECTED] 0         NB. random real matrix
> >>      6!:2 'y=: %. x'             NB. invert it
> >>  13.7781
> >>
> >>      >./|, (=i.1000) - x +/ .* y  NB. how accurate?
> >>  1.97126e_13
> >>
> >
> >Random matrices tend to be numerically stable.  Unstable ones 
> crop up all
> >the time.
> >
> >    hilbert=:]\ %@(i.&.<:@+:)
> >    x=:hilbert 10
> >    y=:%.x
> >    >./|, (=i.10) - x +/ .* y
> >0.108042
> >
> >Best wishes,
> >
> >John
> >
> 
> I agree/understand - but the original enquiry seemed
> to be implying some kind of size limit... I was just
> noting that awhile back I found it amusing/interesting
> that I could invert a 1000 x 1000 matrix on my desktop
> machine that is not all that much bigger than a laptop -
> even with its 20 inch screen... (thinking back to the
> days of machines like the IBM 3090 mainframe... :)
> 
> - joey
> 
> -------------------------------------------------------------------
> ---
> For information about J forums see http://www.jsoftware.com/forums.htm
> 

----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to