[BUG] Have a trouble when I calculate MD(lammps) by using pbs to submit tasks

[baihuyu12]

Bug summary

I meet a trouble when I use graph.gh to calculate task by lammps.

I use pbs to submit my task on two local machines, one machine (hyper-thread) can successfully work, but the other (no hyper-thread) can't. Two machine have the same hardware and software settings. Once I submit the task, the other machine will be down. Only I delete the task, and the other machine can be opened again.

I don't know why?

And I also use 'export OMP_NUM_THREADS=50 && nohup mpirun /home/***/deepmd-kit/bin/lmp < in.data >output 2>&1 &' to submit task, only one thread can be used which is computationally inefficient.

How can I implement multi-threaded on the other machine(no hyper-thread)?

DeePMD-kit Version

2.1.1

TensorFlow Version

2.7.0

How did you download the software?

Offline packages

Input Files, Running Commands, Error Log, etc.

#!/bin/bash
#PBS -N In-12
#PBS -l nodes=node01:ppn=20
#PBS -l walltime=12000:00:00

cd $PBS_O_WORKDIR
NP=cat $PBS_NODEFILE|wc -l
mpirun -machinefile $PBS_NODEFILE -np $NP /home/software/deepmd/bin/lmp < in.data

Steps to Reproduce

anything is ok, Only submit isn't ok.

Further Information, Files, and Links

No response

[njzjz]

You did not limit the number of threads when you started multiple processes. These processes may compete for resources.