This release of GAMESS-US is available on gadi. To run it user must load the appropriate environment via
$ module load gamess/2020-06-R1
The module provides the following environment:
setenv GMSPATH /apps/gamess/2020-06-R1 prepend-path PATH /apps/gamess/2020-06-R1/. conflict gamess module load openmpi/4.0.2 setenv GAMESS_BASE /apps/gamess/2020-06-R1 setenv GAMESS_ROOT /apps/gamess/2020-06-R1 setenv GAMESS_VERSION 2020-06-R1
See here for more information about modules.
Below is an example of a parallel GAMESS-US job runscript gamess_submit.pbs to be run under PBS on gadi. It uses a fictitious project a99, using 4 cpus, and requiring 30 minutes of wall clock time, and 4GB of memory. The program uses the $PBS_JOBFS scratch directory for intermediate files, and for our example this is set to 1GB.
#!/bin/bash #PBS -P a99 #PBS -l walltime=30:00 #PBS -l ncpus=4 #PBS -l mem=4gb #PBS -l jobfs=1gb #PBS -l storage=scratch/a99 #PBS -l wd module load gamess/2020-06-R1 INPUT="molecule1.inp" OUTPUT=${INPUT%.*}.log rungms $INPUT $PBS_NCPUS >& $OUTPUT
To submit the job to the queueing system:
$ qsub gamess_submit.pbs
If you specify ncpus > the cores-per-node number please keep it multiple of of the cores-per-node number (multiple of 48 for the gadi normal queue).
Some examples of input decks can be found in the directory $GMSPATH/tests. Documentation on input to GAMESS is in $GMSPATH/docs-input.txt.
Four versions of the major GAMESS binary are provided. All of them are compiled with Intel Parallel Studio 2020.x.xxx compilers and linked to the matched MKL libraries. For all of them the network DDI interface build in MPI 'mixed' mode using OpenMPI library, version 4.0.2. The distinctions between the binaries as follows:
The default version of the executable is gamess.00.x. To use an alternative executable, you must provide its version number as the third command-line argument to the script 'rungms', e.g.,
to use gamess.02.x you can modify the line as
rungms $INPUT $PBS_NCPUS 02 >& $OUTPUT
Note that GAMESS runs the conventional SCF algorithm by default . For modern day cpu architecture, like on raijin nodes, the direct SCF algorithm is noticeably faster than conventional one. To invoke the direct SCF algorithm, user must specify
$SCF DIRSCF=.TRUE. $END
in the input file.
The VB2000 calculations can run on a single node only.
To run calculation having the NBO part your 'rungms' script call line should be
rungms $INPUT $PBS_NCPUS nbo7 >& $OUTPUT
The NBO part of calculation is not parallelized.