The CSF2 has been replaced by the CSF3 - please use that system! This documentation may be out of date. Please read the CSF3 documentation instead. To display this old CSF2 page click here. |
CASTEP
Overview
CASTEP is a leading code for calculating the properties of materials from first principles. Using density functional theory, it can simulate a wide range of properties of materials proprieties including energetics, structure at the atomic level, vibrational properties, electronic response properties etc. In particular it has a wide range of spectroscopic features that link directly to experiment, such as infra-red and Raman spectroscopies, NMR, and core level spectra.
Versions 6.1.1 and 8.0 and 16.1.1 are available on the CSF.
6.1.1 was compiled with Intel 12.0.5, MKL 10.3u5 (for FFT and MATHLIBS) and the ‘BUILD=fast’ option. No Intel SandyBridge specific options were included as this would make it incompatible with Intel Westmere and AMD nodes.
8.0 was compiled with Intel 14.0.3, MKL 11.1u3 (for FFT and MATHLIBS) and the ‘BUILD=fast’ option. A version has also been compiled to run only on Intel CPUs with possible optimization when running on Sandybridge and Ivybridge (AVX) and Haswell (AVX2). The minimum architecture is SSE4.2.
16.1.1 was compiled with Intel 14.0.3, MKL 11.1u3 (for FFT and MATHLIBS) and the ‘BUILD=fast’ option with possible optimization when running on Sandybridge and Ivybridge (AVX) and Haswell (AVX2). The minimum architecture is SSE4.2 and it cannot be run on AMD nodes.
17.2 was compiled with Intel 14.0.3, MKL 11.1u3 (for FFT and MATHLIBS) and the ‘BUILD=fast’ option with possible optimization when running on Sandybridge and Ivybridge (AVX) and Haswell (AVX2). The minimum architecture is SSE4.2 and it cannot be run on AMD nodes.
Restrictions on use
CASTEP is protected by a unix group. Only users who belong to a research group which has been granted a license by STFC may be added to the unix group.
Licenses are normally in the name of the Project Investigator or Head of a research group. Every user of the research group who wishes to use the software must be registered against their groups license in the STFC database.
It is not necessary for users to sign the license, but they must abide by all 3 associated license documents when using the software. Copies are available on the CASTEP website and the CSF under $CASTEP_HOME/licenses
.
To get access to the software on the CSF we will need proof that your group has a license and that you are registered against that license at STFC. Please email its-ri-team@manchester.ac.uk with your details and we will confirm when access has been granted.
Set up procedure
To access the software you must first load the appropriate modulefile which is dependant on whether you are using ethernet or infinband connected compute nodes. See the examples below for details.
Running the application
Please do not run CASTEP on the login node. Jobs should be submitted to the compute nodes via batch.
The examples below are all based Intel nodes which is the most commonly used and recommended type for this application.
Serial batch job submission
Load one of the modulefiles:
module load apps/intel-14.0/castep/17.2-avx2 # Intel CPUs only module load apps/intel-14.0/castep/16.1.1-avx2 # Intel CPUs only module load apps/intel-14.0/castep/8.0-avx2 # Intel CPUs only module load apps/intel-14.0/castep/8.0 module load apps/intel-12.0/castep/6.1.1
Note that the -avx2
versions will allow CASTEP to run on any Intel node type but not the AMD nodes. The software will use an appropriate level of optimization for the node it is running on. For example, if the CSF places your job on one of the newer Haswell or Broadwell nodes then up to AVX2 optimization can be used. If it lands on a Sandybridge node then up to AVX optimization will be used. Otherwise default SSE4.2 optimzation will be used.
Write a job submission script, for example:
#!/bin/bash #$ -S /bin/bash #$ -cwd #$ -V castep.serial myinput
Submit the jobscript using:
qsub scriptname
Parallel batch job submission
SMP (single node) – maximum of 24 cores
Load one of the modulefiles:
module load apps/intel-14.0/castep/17.2-avx2 # Intel CPUs only module load apps/intel-14.0/castep/16.1.1-avx2 # Intel CPUs only module load apps/intel-14.0/castep/8.0-avx2 # Intel CPUs only module load apps/intel-14.0/castep/8.0 module load apps/intel-12.0/castep/6.1.1
Write a job submission script, for example:
#!/bin/bash #$ -S /bin/bash #$ -cwd #$ -V ### Request the SMP pe and number of cores #$ -pe smp.pe 12 mpirun -n $NSLOTS castep.mpi myinput
Submit the jobscript using:
qsub scriptname
More than 48 cores, but a multiple of 24
Load one of the modulefiles, note the ‘-ib’ which is required to run on the Inifinband nodes:
module load apps/intel-14.0/castep/17.2-avx2-ib # Intel CPUs only module load apps/intel-14.0/castep/16.1.1-avx2-ib # Intel CPUs only module load apps/intel-14.0/castep/8.0-avx2-ib # Intel CPUs only module load apps/intel-14.0/castep/8.0-ib module load apps/intel-12.0/castep/6.1.1-ib
Write a job submission script, for example:
#!/bin/bash #$ -S /bin/bash #$ -cwd #$ -V ### Request the Infinband connected PE and number of cores #$ -pe orte-24-ib.pe 48 mpirun -n $NSLOTS castep.mpi myinput
Submit the jobscript using:
qsub scriptname
Pseudopotentials
The pseudopotentials supplied with castep can be found in:
$CASTEP_HOME/Pseudopotentials
Further info
- The CASTEP website
- The CASTEP mailing list, and it’s archive, is a very useful source of information.
- Dr Matt Probert at The University of York has some useful resources on First Principles Materials Modelling using CASTEP.
has documentation, tutorials and information on workshops.
Updates
None.