AMBER Archive (2005)

Subject: Re: AMBER: Build for single precision

From: Robert Duke (rduke_at_email.unc.edu)
Date: Thu Dec 15 2005 - 10:23:49 CST


David -
Other amber folks will have to comment on the appropriateness for the "main
product"; for pmemd I don't support a single precision mode (and don't use
the _REAL_ define). This was a decision based on results quality and
reproducibility; generally with single precision builds of dynamics code
there is enough variability in results due to rounding error that it is hard
to even validate the software as correct. So the issue is not so much that
you have more than single precision accuracy in results; you don't.
However, if you go through all the calculations with all the intermediates
in single precision, you end up with results that have significantly less
precision than single precision.
Regards - Bob

----- Original Message -----
From: "David Caliga" <caliga_at_srccomputers.com>
To: <amber_at_scripps.edu>
Sent: Thursday, December 15, 2005 10:31 AM
Subject: AMBER: Build for single precision

> Folks,
>
> I want to build amber8 to use single and not double precision? What do I
> change in the build process to get the preprocessor to set _REAL_ to
> REAL*4.
>
> Thanks,
> David
> -----------------------------------------------------------------------------
> David Caliga
> Application Technology Manager
> SRC Computers, Inc.
> Tel: 719 785 5117
> Fax: 719 262 0223
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu