Technical limitation in replica exchange MD simulation

User discussions
1

GROMACS version: 2024
GROMACS modification: No
Dear All,

I need to perform a replica exchange MD simulation for a chromophore system. When I prepare the system with 30 chromophore molecules and 68 replicas, the simulation finishes successfully. However, for a system containing 100 chromophore molecules and 124 replicas, I encounter the following error:

Fatal error:
Not enough memory. Failed to allocate 18446744073103078171 aligned elements of
size 4 for grid->grid
(called from file
/home/conda/feedstock_root/build_artifacts/gromacs_1720551832028/work/src/gromacs/ewald/pme_grid.cpp,
line 620)

It seems that the required memory exponentially increases with the number of replicas. My question is: Is there a technical limitation on the number of replicas for REMD simulations? If not, what could be the problem?
FYI: I‌ am using Gromacs version 2024

Best regards,
Bahareh

2

what is your configuration for REMD simulation to run,

3

I have a box containing 100 chromophores. I prepared 124 replicas with temperatures ranging from 100 to 800 K. I used a geometric progression to divide the temperature range into 124 values. Please let me know if you need more information.

4

what command you used and your system configuration, how many cores you assigned and how many RAM you got? also provide your mdp file. specifically your grid spacing.

5

Hi,

The following is the script I am using for running my job using slurm. From it you can find all necessary information:
#!/bin/bash
#SBATCH --job-name=remd
#SBATCH --mem=60000
#SBATCH --nodes=8
#SBATCH --tasks-per-node=62
#SBATCH --time=1259200

module load gromacs
mpirun -np 496 gmx_mpi mdrun -v -s remd -multidir repl* -replex 1000

and here is my mdp file for one case:
title = OPLS Lysozyme NPT equilibration
; Run parameters
integrator = md
nsteps = 200000000
dt = 0.001
; Output control
nstxout = 100000
nstvout = 100000
nstenergy = 100000
nstlog = 100000
; Bond parameters
continuation = yes
constraint_algorithm = lincs
constraints = h-bonds
lincs_iter = 1
lincs_order = 4
; Nonbonded settings
cutoff-scheme = Verlet
ns_type = grid
nstlist = 10
rcoulomb = 1.2
rvdw = 1.2
DispCorr = EnerPres
; Electrostatics
coulombtype = PME
pme_order = 4
fourierspacing = 0.16
; Temperature coupling is on
tcoupl = V-rescale
tc-grps = UNL
tau_t = 0.1
ref_t = 100

pcoupl = Parrinello-Rahman
pcoupltype = isotropic
tau_p = 2.0
ref_p = 1.0
compressibility = 4.5e-5
refcoord_scaling = com
pbc = xyz
gen_vel = no

6
  1. why you are using 1 fs, time step? do you need to save this much data? or in this short time? try to reduce those, if i remember correctly, i never used more than 20 GB in my setting.
  2. what is your node configuration? can you try using this way:
#SBATCH --ntasks=496
#SBATCH --cpus-per-task=1