|
From: | Vincent Ustach |
Subject: | Re: [ESPResSo-users] Espresso installation- errors with more than 3 tasks- GPU + MPI |
Date: | Fri, 11 Nov 2016 11:19:20 -0800 |
Hi,This seems to be an issue of your job submission: it seems your GPU is a consumable resource and you have to request it in your jobscript. However this doesn't explain why you can run with less than 4 cores.CheersKai SzuttorInstitute for Computational PhysicsUniversität Stuttgart
Allmandring 3
Room 1.084
70569 Stuttgart, Germany
Phone: +49(0)711/685-67707 Hi all,I am having trouble updating Espresso on a cluster. I am trying to install version 3.3.0-2099-gbeb1108The autotools configure and make seem to go fine, and I can run a script with ntasks = 1, 2, or 3, however, with ntasks => 4, Espresso complains:
all CUDA-capable devices are busy or unavailable
while executing
"cuda setdevice $device"
(file "freely_rotating_chains.tcl" line 10)
srun: error: surface138: task 0: Exited with exit code 1I am submitting batch scripts on a slurm cluster using
export ESPRESSO_SOURCE="/g/g14/ustach2/espresso/may2016/ espressomd/build2/"
FILE1="freely_rotating_chains.tcl"
module load cudatoolkit/7.5
srun --ntasks=4 $ESPRESSO_SOURCE/Espresso $FILE1 -j 15 40 . 0 &My installation steps are
1) git clone git://git.savannah.nongnu.org/espressomd.git
2) cd espressomd
3) ./bootstrap.sh
4) mkdir build2
5) cd build2
6) mv myconfig.hpp path/espresso/may2016/espressomd/build2/
7) use boost-mpi-1.55.0
8) use gcc-4.9.3p
9) use openmpi-gnu-debug-1.4.3
10) module load cudatoolkit/7.5
11) ../configure --with-mpi=yes --with-fftw=yes --with-cuda=/opt/cudatoolkit-7.5 --with-boost=/usr/local/tools/ boost-mpi-1.55.0/
11) make
12) make installI have attached the Espresso script, my configure log and the myconfig.hppAny advice on working through this problem is greatly appreciated.Many thanks,Vincent--Vincent Ustach
University of California, Davis<freely_rotating_chains.tcl><config.log><myconfig.hpp><config.status>
[Prev in Thread] | Current Thread | [Next in Thread] |