ORCA
ORCA is an ab initio quantum chemistry program package that contains modern electronic structure methods including density functional theory, many-body perturbation, coupled cluster, multireference methods, and semi-empirical quantum chemistry methods. Its main field of application is larger molecules, transition metal complexes, and their spectroscopic properties. ORCA is developed in the research group of Frank Neese. The free version is available only for academic use at academic institutions.
Available
- Roihu-CPU: 6.1.1
Note that due to licensing issues every user has to install their own copy of the program.
License
ORCA users should register, agree to the EULA, download and install a private copy of the program (via the ORCA forum website). The free version is available only for academic use at academic institutions.
Usage
- Download ORCA 6.1.1, Linux, x86-64, shared-version,
orca_6_1_1_linux_x86-64_shared_openmpi418_avx2.tar.xz - Move the downloaded file to your computing project's application area
(
/projappl/<proj>/<user>/orca/) - Unpack the package:
mkdir -p /projappl/<project>/<user>/orca
cd /projappl/<project>/<user>/orca
tar xf orca_6_1_1_linux_x86-64_shared_openmpi418_avx2.tar.xz
On Roihu, set ORCADIR before loading the module. The module then
loads the required MPI libraries (gcc/15.2.0, openmpi/5.0.10)
automatically — no manual module loads are needed.
Jobs must be submitted from the CPU login node (roihu-cpu).
Example batch scripts
Note
Wave function-based correlation methods, both single and multireference, often create a substantial amount of disk I/O. In order to achieve maximal performance for the job and to avoid excess load on the Lustre parallel file system it is advisable to use the local disk where available.
On Roihu, the module loads MPI automatically. No mpirun wrapper
script is needed. Set ORCADIR before loading the module. For a
full list of available partitions see the
Roihu batch job partitions
page.
Note
Local NVMe disks on the compute nodes can be accessed through $TMPDIR. Jobs in the "small" partition have 20 GiB quota available, while "medium" nodes have 600 GiB available. $TMPDIR storage does not need to be separately reserved in Slurm to be accessible.
#!/bin/bash
#SBATCH --partition=small # see batch-job-partitions for all options
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=12
#SBATCH --mem-per-cpu=2667 # MB; total memory = nprocs * maxcore
#SBATCH --account=yourproject
#SBATCH --time=01:00:00
export ORCADIR=/projappl/<project>/<user>/orca/orca_6_1_1_linux_x86-64_shared_openmpi418_avx2
module purge
module load orca/6.1.1
${ORCADIR}/orca orca6.inp > orca6.out
Memory — ORCA's %maxcore directive sets memory per MPI process in
MB. The corresponding --mem-per-cpu value should match:
For example, with 12 MPI processes and 32 GB total memory:
%maxcore 2667 and --mem-per-cpu=2667.
After the job completes, check actual memory usage with:
Note
Please remember to adjust %pal nprocs in your input file according to
the total number of requested MPI tasks (--nodes × --ntasks-per-node).
Example of scalability
The following results were obtained on Roihu for a naphthalene
DLPNO-CCSD(T)/aug-cc-pVTZ single-point energy calculation using
%maxcore 5000 (6 tasks: %maxcore 8000) on an M-node with Lustre scratch.
Jobs were run sequentially. Wall times on shared nodes may vary due to
I/O and memory bandwidth contention from co-located jobs.
| MPI tasks | Wall time | Peak scratch | MaxRSS total | %maxcore |
|---|---|---|---|---|
| 6 | 56 min | ~32 GB | 47 GiB | 8000 MB |
| 12 | 48 min | ~34 GB | 59 GiB | 5000 MB |
| 24 | 38 min | ~34 GB | 117 GiB | 5000 MB |
| 48 | 31 min | ~34 GB | 231 GiB | 5000 MB |
The input file is available at naphthalene_dlpno.inp.
DLPNO-CCSD(T) shows good parallel scaling for this system. Peak scratch disk (~34 GB) is essentially independent of core count — the total integral work is determined by the system size, not the number of processes. Memory per task decreases with more processes, reflecting DLPNO's domain decomposition.
Estimating memory requirements
ORCA's %maxcore sets memory per MPI process in MB. Set
--mem-per-cpu to match:
After a completed run, check actual memory usage with:
Divide MaxRSS by the number of MPI tasks to get actual peak
per process and adjust %maxcore and --mem-per-cpu accordingly.
Note that ORCA may require more memory per process with fewer tasks
— if you reduce %pal nprocs, increase %maxcore proportionally.
References
The generic references for ORCA are:
- Frank Neese. The ORCA program system. Wiley Interdiscip. Rev. Comput. Mol. Sci., 2(1):73–78, 2012. doi:https://doi.wiley.com/10.1002/wcms.81.
- Frank Neese. Software update: the ORCA program system, version 4.0. Wiley Interdiscip. Rev. Comput. Mol. Sci., 8(1):e1327, 2018. doi:https://doi.wiley.com/10.1002/wcms.1327.
- Frank Neese, Frank Wennmohs, Ute Becker, and Christoph Riplinger. The ORCA quantum chemistry program package. J. Chem. Phys., 152(22):224108, 2020. doi:https://aip.scitation.org/doi/10.1063/5.0004608.
Please do not only cite the above generic references, but also cite in addition the original papers that report the development and ORCA implementation of the methods you have used in your studies!