[gmx-users] platform dependent fate of a md simulation

[Mark Abraham]

On 11/01/2012 9:18 PM, Ester Chiessi wrote:
> Hi,
> running a md simulation with the same tpr file ( using Gromacs 4.5.3 ) 
> on three different platforms ( A: /HP xw8600 Workstation; B: PC with 
> //Intel CORE 2 DUO processor; C: /IBM Power6 cluster ), I obtained the 
> following results:
> on A and B platforms the run is fine,
> on C the run crashes at the first step, returning LINCS errors.
> Did anyone observe this behaviour? What does it mean?
> I wonder if my system is really wrong or not.
> I noted this difference in the corresponding md.log files:
> Platform A RMS relative constraint deviation after constraining: 3.56e-06
> Platform B RMS relative constraint deviation after constraining: 3.48e-06
> Platform C RMS relative constraint deviation after constraining: 3.58e-06
> Really many thanks for any comment.
>

See http://www.gromacs.org/Documentation/Terminology/Reproducibility for 
general discussion. Probably A and B are getting lucky with the 
numerical integration of a poorly conditioned system, and C is not.  For 
advice on fixing the problem, see 
http://www.gromacs.org/Documentation/Terminology/Blowing_Up

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20120111/4e32ef1e/attachment.html>