Solver run commands

reference · Reviewed 2026-09-14

The command line each solver expects on Rescale, plus the license flags and per-solver details that aren't obvious from the software catalogue. Look up your solver; the entries are independent.

The command goes in the Command field on Software Settings. See Create and submit a job for where that sits in job setup, and Command Templates for the suggested forms Rescale offers.

Host and machine files

MPI solvers need the list of nodes in the cluster. Rescale writes it into the $HOME folder of your workspace in the common formats, so reference it by full path:

--hostfile $HOME/machinefile.openmpi

List $HOME to see what's available for your run. The usual set is hosts, mpd.hosts, machinefile, machinefile.gpu, machinefile.openmpi and PCF.xml.

Some solvers want the host list somewhere specific rather than as an argument:

Solver	Where it looks
Abaqus	abaqus_v6.env
ANSYS	the $MACHINES environment variable, pre-populated in the required format

Abaqus

abaqus job=<job> cpus=$RESCALE_CORES_PER_SLOT mp_mode=mpi interactive

<job> is your .inp file without the extension. $RESCALE_CORES_PER_SLOT picks up every core in the hardware you selected, so the same command works on any coretype.

Common optional arguments:

Argument	Use for
scratch=$PWD/tmp	put scratch files on the node's local disk
gpus=<count>	GPU-accelerated solves
memory=<size>	cap or raise the memory Abaqus requests
output_precision=full	full precision in the output database, rather than single
resultsformat=both	write both odb and sim results, rather than one

For the complete list, SSH into the cluster and run abaqus -help.

Global and local models

A two-stage global/local analysis is two commands in one job, not two jobs. Upload both .inp files, then run the global model and pass its name to the local model with globalmodel=. Rescale runs the commands in order:

abaqus job=input_global cpus=$RESCALE_CORES_PER_SLOT mp_mode=mpi \
  output_precision=full double=both parallel=domain interactive

abaqus job=input_local cpus=$RESCALE_CORES_PER_SLOT mp_mode=mpi \
  output_precision=full double=both parallel=domain globalmodel=input_global interactive

Fortran user subroutines

A user subroutine extends Abaqus where the normal input syntax cannot express what you need, most often a custom material model. Pass the source or object file with user=:

abaqus job=<job> user=<user-subroutine> cpus=$RESCALE_CORES_PER_SLOT mp_mode=mpi interactive

gfortran only

Abaqus on Rescale is linked against gfortran. A subroutine that depends on Intel Fortran extensions will not compile.

Using a Dassault Systèmes License Server

To check out from a DSLS server rather than FlexLM, prepend these two lines to your command on the Software step:

export ABQLMIMPL=DSLS
export DSLS_CONFIG=$(IFS=:; DSLS=$HOME/DSLS ; if [ ! -f $DSLS ] ; then for i in $LM_LICENSE_FILE ; do echo ${i#*@}:${i%@*} >> $DSLS;done;fi;echo $DSLS)

Simcenter STAR-CCM+

A typical batch invocation:

starccm+ -power -rsh ssh -np $RESCALE_CORES_PER_SLOT -machinefile $HOME/machinefile -batch macro.java -load input.sim

The -power flag is required when you use the Rescale On-Demand License or your own Power Session license.

• To mesh only, use -batch mesh.
• To mesh and then run, use -batch mesh,run.

For a Power-on-Demand (POD) license, select Provide Existing License on the Software Setup page, enter 1999@flex.cd-adapco.com as the license, and put your product code in the Project Code field.

LS-DYNA

LS-DYNA ships with a Rescale wrapper. The core arguments are -i for the input, -p for precision, -n for DMP ranks, -s for SMP threads, and -a for additional options.

# SMP
ls-dyna -s $RESCALE_CORES_PER_SLOT -i <input-file> -p <precision>

# double precision
ls-dyna -i <input-file> -p double

# hybrid: DMP ranks with SMP threads per rank
ls-dyna -n <dmp-ranks> -s <smp-threads-per-rank> -i <input-file> -p <precision> -a "memory=600m memory2=60m"

Sending a sense switch to a running job

To write a restart or check progress mid-run, SSH into the cluster, change to work/shared, and write the switch to d3kil:

echo sw2 > d3kil

The file is consumed and removed after execution.

Switch	Effect
sw1	Write a restart file and terminate
sw2	Print the current time and cycle
sw3	Write a restart file and continue
sw4	Write a plot state and continue

CONVERGE

Run converge. If you need super licenses, add the flag for your major version:

converge -a super   # CONVERGE 2.x
converge -S         # CONVERGE 3.x

Rescale recommends running CONVERGE on a multiple of the cores-per-node for the coretype you chose. Check the cores per node for your coretype in Coretype Explorer rather than assuming a number.

COMSOL

To reach a running COMSOL server from your own machine:

1. Start a COMSOL Multiphysics Server or COMSOL Server job on Rescale.

2. Note the username and password printed in process_output.log.

3. Open an SSH tunnel to the cluster, with source and destination port 2036:

   ssh -i </path/to/key> <user>@<ip> -L 2036:localhost:2036 -N -v

4. Connect your local COMSOL client to localhost:2036.

From a Rescale Desktop, skip the tunnel and use the hostname printed in the log.

When you're finished, stop the server from the Rescale job page so the cluster shuts down.

LAMMPS

The installed package list depends on the version you select. Recent versions ship with the standard package set plus most USER packages: USER-ATC, USER-CGDNA, USER-DIFFRACTION, USER-INTEL, USER-OMP, USER-PHONON, USER-REAXC and others. Older versions differ. Check the specific version you plan to use before relying on a particular package.

Related

• Browsing the Catalog for finding a package and picking a version
• Software for how software works on Rescale: licensing, versions, publishing your own
• Create and submit a job for where the command goes
• Hardware & Coretypes for matching a coretype to a solver