Is it possible to disable Hyperthreading? Or a way via torque PBS script to use only physical cores available in the node.
My problem is when I run my MPI job with node=1:ppn=12 it runs with a good speed.(1st iteration take less than 1min) but when I use node=2:ppn=12 the speed is horrible (1st iteration takes more than 6hrs)
My Torque PBS JOB SCRIPT
#!/bin/bash
#PBS -N Vcrlx
#PBS -q batch
#PBS -l nodes=2:ppn=12
#PBS -l walltime=20:00:00
#PBS -m bea
#PBS -j oe
cd $PBS_O_WORKDIR
mpirun -np 24 pw.x < vcrlx.in > vcrlx.out
Tried solutions
I emailed them about this Hyperthreading problem but no response yet (10days)
As my code Quantum Espresso can run with MPI, OpenMP and with Mixed MPI+OpenMP. I tried my best to understand all these but failed to get desired results. Or maybe I misunderstood because of very poor background knowledge of these things.
I tried many combinations like following but none of them worked
#!/bin/bash
#PBS -N Vcrlx
#PBS -q batch
#PBS -l nodes=2:ppn=12::mpiprocs=1:ompthreads=24 <--------------
#PBS -l walltime=24:00:00
#PBS -m abe
#PBS -j oe
cd $PBS_O_WORKDIR
mpirun -np 24 pw.x < vcrlx.in > vcrlx.out
its output
Parallel version (MPI & OpenMP), running on 48 processor cores
Number of MPI processes: 24
Threads/MPI process: 2
MPI processes distributed on 2 nodes
R & G space division: proc/nbgrp/npool/nimage = 24
Please help me to run the job without Hyperthreading. Thanks