From: Ramin Ekhteiari (ramin_ekh_at_yahoo.com)
Date: Wed May 14 2014 - 07:19:10 CDT
Hi dears,
I am trying to run NAMD-CUDA version on a batch system, when I do it without PME facility, I face with no problem but when employing PME I face with this problem **FATAL ERROR: PME offload requires exactly one CUDA device per process*
This is my lsf file
#!/bin/bash
# UHeM 2014-05-14 v0.2
#BSUB -q tesla
#BSUB -o %J.out
#BSUB -e %J.err
#BSUB -P stbknm
#BSUB -m gpu
#BSUB -n 4
#BSUB -J NAMD
export file=$PWD/step5_production.inp
source /RS/progs/namd/NAMD_CVS-2014-05-13_Linux-x86_64-multicore-CUDA/ENVARS
export bindir=/RS/progs/namd/NAMD_CVS-2014-05-13_Linux-x86_64-multicore-CUDA
export LD_LIBRARY_PATH=".:$LD_LIBRARY_PATH"
declare -i Ncpu
export Ncpu=0
export nodefile=.nodelist
echo group main > $nodefile
export nodes=$LSB_HOSTS
for node in $nodes
do
echo "host $node" >> $nodefile
Ncpu=$Ncpu+1
done
$bindir/charmrun +p$Ncpu +idlepoll $bindir/namd2 $file >& cuda.log
rm -f $nodefile
Would be grateful for any suggestion.
Cheers,
Ramin.
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:20:48 CST