Re: Infiniband and PME gridsize

From: Brian Bennion (brian_at_youkai.llnl.gov)
Date: Tue Dec 14 2004 - 12:06:36 CST

Can't address number 1 (wish I could!! I would love to try that hardware)

Number 2
>From my discussions with the developers, the transforms are optimized for
small factorials of the pme gridsizes. In other words, 99.2 doesn't
factor well. However, 90 does a bit better 3x3x2x5. Ideally a number
that factors to prime numbers less than 5 is "optimal". This may have
changed though. So chose a value that encompasses your system and can be
factored well...they don't have to be equal to your system dimensions.

Regards
Brian

On Tue, 14 Dec 2004, Gianluca Interlandi wrote:

> Dear NAMD users,
>
> I have two independent questions:
>
> 1) Are there any experiences about the scalability running NAMD on
> infiniband? I have been running NAMD on myrinet and, as an example, for
> a system with 51000 atoms I get a seedup of 14 on 16 CPUs (dual opteron
> 1800+).
>
> 2) What is recommended for a PME gridsize? One of my boxes has a
> sidelength of 99.2 A. What is the "ideal" gridsize in this case?
>
> Many thanks,
>
> Gianluca
>
> -----------------------------------------------------
> Gianluca Interlandi gianluca_at_bioc.unizh.ch
> +41 (0) 1 635 55 93
> +41 (0)79 655 73 35
> http://biocroma.unizh.ch/gianluca/
>
> PhD-Student of the Department of Biochemistry
> at the University of Zurich Irchel, Switzerland
> -----------------------------------------------------
>

*****************************************************************
**Brian Bennion, Ph.D. **
**Computational and Systems Biology Division **
**Biology and Biotechnology Research Program **
**Lawrence Livermore National Laboratory **
**P.O. Box 808, L-448 bennion1_at_llnl.gov **
**7000 East Avenue phone: (925) 422-5722 **
**Livermore, CA 94550 fax: (925) 424-6605 **
*****************************************************************

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:39:03 CST