VASP Forum

Hello,

I am running a spin-polarized geometry optimization with VASP 6.3.2 on our HPC cluster, but the calculation repeatedly terminates with an OOM (Out of Memory) error after several ionic steps.

System
VASP version: 6.3.2
Number of atoms: 57
Spin-polarized calculation (ISPIN = 2)
DFT+U
Cluster hardware:
1 node
2 × Intel Xeon E5-2680 v3
128 GB RAM per node
Job script
#!/bin/bash
#SBATCH --job-name=R
#SBATCH --partition=cpu
#SBATCH --nodes=1
#SBATCH --ntasks=24
#SBATCH --cpus-per-task=1
#SBATCH --time=200:00:00

mpirun -np 24 vasp_std
INCAR
ISTART = 1
ISPIN = 2
LREAL = .FALSE.
ENCUT = 400
LWAVE = .FALSE.
LCHARG = .FALSE.
ADDGRID = .TRUE.
ISMEAR = 0
SIGMA = 0.05
NELM = 90
NELMIN = 6
EDIFF = 1E-4

NSW = 500
IBRION = 2
ISIF = 2
EDIFFG = -2E-3

LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = 2 -1 -1 -1 -1
LDAUU = 5.9 0 0 0 0
LDAUJ = 0 0 0 0 0

MAGMOM = 18*3.0 22*0.6 10*0.6 6*3.0 1*0.0
KPOINTS
Monkhorst-Pack
3 3 1
Output near the crash
12 F= -.34971991E+03 E0= -.34971527E+03 dE =-.173245E+00

...

DAV: 7 -0.349822961716E+03

Primary job terminated normally, but 1 process returned
a non-zero exit code.

mpirun noticed that process rank 12 exited on signal 9 (Killed).

slurmstepd: error: Detected 1 oom_kill event in StepId=201923.batch.
Some of the step tasks have been OOM Killed.

The calculation is progressing normally until one MPI process is killed because of insufficient memory.

I would appreciate any advice on the likely cause. In particular, I would like to know whether this is mainly related to:

the number of MPI ranks (24),
memory distribution among MPI processes,
any INCAR settings (e.g., LREAL = .FALSE. or ADDGRID = .TRUE.),
or another issue.

Thank you very much for your help.

Dear ghadakhedr,

To me this looks like it could be a memory issue. But I can only guess since I don't have your input and output files. So I would like to ask you to upload your INCAR, POSCAR, KPOINTS and OUTCAR file. Then I can examine further.
Meanwhile you could run:

 grep -A 8 "total amount of memory" OUTCAR

This will output the memory consumption on MPI rank 0. From this you can deduce the total amount of memory which is used by your calculation. You just have to multiply the used amount of memory on rank 0 times the total number of MPI ranks you are using.

All the Best Jonathan