[gmx-users] Gromacs slow for 23000 atom DPPC bilayer on a (1 x 4) node: 50 ps in 10 hours

[Mark Abraham]

maria goranovic wrote:
> An update:
> Using a 4th order PME spline resulted in ~ 15 % increase in efficiency.
> 
> Thank you, Carsten.

As you can see in some of the original PME papers, tweaking the 
parameters can give about an order of magnitude variation in the 
approximation error for given computational cost, or similar variation 
in cost for a given error. pme_order = 6 works best on my hardware for 
error that I found acceptable.

Mark