Performace lost due to PME ranks

GROMACS version: 5.1.4
GROMACS modification: Yes/No
Here post your question

for -np 64

  1. I am using “gmx_mpi mdrun -v -deffnm postpulleq_0.98 -cpo postpulleq_0.98.cpt -cpt 180”
    When I run it, gromacs defult divides np into 32 (8x2x2) plus 32 (8x4X1) for dd and PME repsectively .
    i.e I am getting
    “Initializing Domain Decomposition on 64 ranks
    Dynamic load balancing: auto
    Will sort the charge groups at every domain (re)decomposition
    Initial maximum inter charge-group distances:
    two-body bonded interactions: 0.415 nm, LJ-14, atoms 32239 32248
    multi-body bonded interactions: 0.415 nm, Fourier Dih., atoms 32239 32248
    Minimum cell size due to bonded interactions: 0.457 nm
    Maximum distance for 5 constraints, at 120 deg. angles, all-trans: 0.218 nm
    Estimated maximum distance required for P-LINCS: 0.218 nm
    Guess for relative PME load: 0.46
    Will use 32 particle-particle and 32 PME only ranks
    This is a guess, check the performance at the end of the log file
    Using 32 separate PME ranks, as guessed by mdrun
    Scaling the initial minimum size with 1/0.8 (option -dds) = 1.25
    Optimizing the DD grid for 32 cells with a minimum initial size of 0.571 nm
    The maximum allowed number of cells is: X 12 Y 10 Z 70
    Domain decomposition grid 8 x 2 x 2, separate PME ranks 32
    PME domain decomposition: 8 x 4 x 1”

NOTE: 7.1 % performance was lost because the PME rank had less work to do than the PP ranks. You might want to decrease the number of PME ranks or decrease the cut-off and the grid spacing.

  1. so decreased the pme ranks using use manual “-dd xxxx -dlb on -npme p” (where, p= 28 or 30 or any other value less than 32). Now gromacs is dividing it as (p x 1 x 1) and that causes more load in other directions. Now I am getting

NOTE: 13.1 % performance was lost because the PME rank had more work to do than the PP ranks.

I would like to know how to break alloted PME ranks into the desired value and not just (px1x1).
Is it possible??