namd performance on altix machines

From: Michel Espinoza-Fonseca (
Date: Fri Sep 09 2005 - 07:10:48 CDT

Hi all,

I'm wondering if any of you have tested the new version of NAMD compiled
for Altix machines. I was particularly perplexed when I experienced the
following: when I ran a simulation (~30,000 atoms, using PBC and PME) on
25 processors, the performance was quite good (~12 ns/day), but when I
increased the number of processors to 50, I got exactly the same
performance (or even less)!

I'm not quite sure if this problem is due to the bad scalability of the
binary compiled for this architecture or a problem with our system (I
doubt it).

It'll be interesting to see what others have experienced. It'll
enormously help the program to be improved. Particularly, I'm very
satisfied with NAMD -- it's my main simulations engine, so it's nothing
against the program, just the opposite.


This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:41:07 CST