----- Original Message -----
From: "Nick Arnett" <[EMAIL PROTECTED]>
To: "Brin-L" <[EMAIL PROTECTED]>
Sent: Friday, December 07, 2001 4:50 PM
Subject: RE: Quick irregulars question


>
>
> > -----Original Message-----
> > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On
> > Behalf Of Dan Minette
>
> [snip]
>
> > I've been running the same number crunching code for about 17
> > years now.  I
> > started out running on a $200,000 machine that ran at half a
> > megaflop.  I'm
> > now running on a $750 computer that runs at about 375 megaflops.  So, I
> > spent $2/mflop in 2001, and $400,000 in 1984.  That's amazing.
>
> Any idea when the calculation will be completed?
>
> ;-)
>
> Nick

ROTFLMAO.

Well the code is called MCNP for Monte Carlo Neutron Photon.  It is a
FORTRAN program that started development in the '50s and is still being fine
tuned.  It models the propagation of neutrons and photons through solid
formation using the basic physics of scattering.

It has all sorta neat acceleration techniques in order to get the answer
faster than simply modeling the transport of a zillion particles through the
earth formation. (For example, in 1 second a 1 curie source puts out about
37 billion gamma rays. Early on, its use was limited, since computers were
very slow. In order to use it at all, the problem had to be simplified and
the acceleration techniques had to be worked out carefully.

When I started, I worked months to be able to get partial runs that would go
overnight and full runs that would take about a week.  Now, I can do far
less optimization and get good results in anywhere from 10 minutes to
overnight for the most complicated problems and the most exact answers. (I
just didn't do those problems even 10 years ago.)

Its a neat benchmark because its close to the the exact same code run over
the last 20 years.  One of the main reasons I'm in business for myself is
that I now have, in my house, far more power than the old 10 million dollar
Cray XMPs.

Dan M.

Reply via email to