[gmx-users] Is there any fundamental difference between the output of serial, gpu and parallel gromacs run

[sudipta]

Hi All,

We were trying to test a simple lipid bilayer system using gromacs with
martini model and polarizable CG water to validate the model. We have
tested it by several ways such as varying gromacs version, precision,
different platform,  different initial configuration, increasing length of
trajectory etc. However, we were always getting exactly same results and
didn't reach the exact reported value as published in plos computational
biology, e1000810,vol 6, issue6, 2010. We are getting a small error
compared to their value. However, we used the same system, same model, same
md simulation parameter, same length of trajectory (initially then we have
varied). May be the authors have used MPI version of gromacs. However, we
have not tested with MPI version of gromacs. Is there any sources of error?
Apparently, there is no difference but is there any inherent operation that
changes our data. I am just curious. We are puzzled about this.

Any suggestions, comments are welcome.

Thanks in advance
Sudipta