AMBER Archive (1999)

Subject: Re: Problem with bench test on PIII450

From: David Case (case_at_scripps.edu)
Date: Fri Aug 13 1999 - 13:54:48 CDT

On Fri, Aug 13, 1999, Claudio Chuaqui wrote:

>
> I have been porting AMBER4.1 to a dual PentiumIII 450....
>
> Both in serial and parallel, under every possible compiler/optimization
> directive combination, using pgf77, g77 and even fort77,
> I obtain the following (identical) differences at the first time step
> for the dnawat/Ewald test case.:
>

Yes, this problem is familiar to us. Even for Amber 6 (coming out this
fall) we are unable to exactly reproduce Ewald energies on Intel chips. We
believe it is likely to be related to the implementation of the erfc
functionality in system libraries, but are still trying to pin it down.

It is of interest that "traditional" Unix boxes (e.g. from SGI, Compaq(DEC),
HP, Cray T3E, Sparc Solaris) all give identical results, that are slightly
different from those returned by Intel processors (whether running Windows
or Linux, and also seemingly independent of compilers or optimization.)
We would be happy to hear if anybody has results on Intel boxes using
Solaris, and/or any other good suggestions.

As far as we can tell, PME simulations on Intel chips still seem to be
"good": they conserve energy, give good derivatives, etc. The results are
just very slightly different than for other platforms. In the past, few
members of the Amber development team were using Intel boxes, so this
problem has perhaps not received the attention it deserved.

...thanks for your input....dac 

-- 

==================================================================
David A. Case                     |  e-mail:      case_at_scripps.edu
Dept. of Molecular Biology, TPC15 |  fax:          +1-858-784-8896
The Scripps Research Institute    |  phone:        +1-858-784-9768
10550 N. Torrey Pines Rd.         |  WWW:                         
La Jolla CA 92037  USA            |    http://www.scripps.edu/case
==================================================================