Re: NAMD on Linux cluster with Myrinet

From: Gengbin Zheng (
Date: Mon Oct 18 2004 - 13:28:36 CDT

Like I mentioned before in many circumstances, Intel C++ compiler 8 (7
was fine) is buggy for running NAMD and you have to try your luck since
different architectures the compiler works in slightly different way in
compiler optimization and things may break in mysterious way. New intel
patch may fix some of the problems and that's why NAMD *started* to work
just recently on NCSA's xeon clusters with icc. I assume NCSA ia64 with
icc works fine since I have not heard much of complain recently.


Alberto Santana wrote:

>I managed to compiled NAMD with gm using icc v8.0. The BPTI example runs OK
>but the ApoA1 (as created by Jim) does not, it complains about atoms not
>having unique coordinates, right after that the "atoms are moving too fast"
>error and it aborts. I then tried the MPI-GM version compiled with icc v8.0
>and got exactly the same behavior. The ApoA1 example aborts because the atoms
>were moving too fast. At this moment I don't trust neither one of those 2
>I gave g++ a try and Voila!!!, everything runs smoothly. It would have been
>nice to have them compiled with icc if IT COULD create a GOOD executable.
>I wonder how the people at NCSA compiled theirs.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:37:55 CST