Cube vs. Truncated Octahedron Performance

From: Alexander Predeus (predeus_at_gmail.com)
Date: Thu Oct 21 2010 - 17:04:42 CDT

Dear all,

I am studying a large protein-DNA complex. The size of the system
solvated in cubic box was ~460k atoms. I have decided to use truncated
octahedron, which turned out to be only 360k atoms. But to my
disappointment, when I ran the (truncated octahedron) system on TACC
Ranger using 512 cores, it actually performed WORSE then the cube!

Is it because the system is smaller and scales worse now (due to
number of patches etc), or am I overlooking something important? I
have turned "wrapNearest" option on, so the cell looks like a proper
truncated octahedron; all the energies look reasonable too. Are there
any other options to play with to improve the performance?

 All the best,

-- Alex Predeus,
Michigan State University

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 05:23:19 CST