[gmx-users] Nitty Gritty details of compiling to use both cores of an X86 core-duo

Trevor Marshall trevor at trevormarshall.com
Thu Feb 22 06:32:38 CET 2007

Thanks very much to the folk who did all the hard work on GROMACS. 
Yesterday I did a simulation of the nuclear receptor VDR for 1 nanosecond, 
and, using Vega-ZZ for analysis,  was able to identify the entrance tunnel 
for the ligand. This is something I have been trying to do for several 
years... Kudos to the GROMACS team!... The next step is to show the ligand 
enter and migrate to the binding pocket.

I have already used Autodock 3 to identify  a number of moderate affinity 
ligand binding sites adjacent to the ligand entry tunnel. Several of these 
look likely to induce ligand entry in an MD simulation, so I am needing to 
figure out how to get the computing equipment together capable of running 
50 nsec or so of simulation (I think 50nsec, do you agree with this estimate?)

At the moment I am using an overclocked E4300 Core 2 Duo (the just-released 
Intel chip). With a 4 ps step I don't get torsional violations, and I can 
simulate one nanosecond every 3.2 hours of time on just one half of the CPU.

But I would like to get both halves of the CPU doing the calculation, and 
am having trouble with the compilation switches to achieve that. I have an 
SMP Linux kernel (v2.6.19) which allocates multiple jobs nicely between the 
cores (for example, when I am running twelve single-threaded Autodock jobs 
simultaneously, everything works fine) but when I tried to compile the FFTW 
and GROMACS sources (with gcc 4.1) the Configure script complains that this 
compiler doesn't recognize the MPI commands.

Is it possible to compile using gcc 4.1 so that I get an SMP system using 
both cores of my CPU? Do I need a different compiler? What switches or 
script tweaks do I need?


Trevor G Marshall, PhD
Dept of Biological Sciences and Biotechnology, Murdoch University, Western 
Director, Autoimmunity Research Foundation, Thousand Oaks, California

More information about the gromacs.org_gmx-users mailing list