ORCA on AI.Panther

ORCA 6.1.1 is available to all AI.Panther users. It runs as a batch job and has no graphical interface.

Load it with:

module purge
module load orca

Running a calculation

Save your ORCA input as input.inp, then create a submission script orca.sbatch:

#!/bin/bash
#SBATCH --job-name=orca
#SBATCH --partition=med
#SBATCH --nodes=1
#SBATCH --ntasks=4
#SBATCH --hint=nomultithread
#SBATCH --time=04:00:00
#SBATCH --output=orca-%j.out

module purge
module load orca

$(which orca) input.inp > input.out

Submit it:

sbatch orca.sbatch

Four rules for parallel runs

Skipping any of these causes the job to fail, usually with an error that does not point at the real cause.

1. Match %pal to --ntasks. The %pal nprocs N end in your input file must equal the --ntasks=N in your submission script.
2. Include --hint=nomultithread. Without it Slurm hands out hyperthreads, ORCA's parallel layer counts physical cores, and the job stops with "There are not enough slots available in the system".
3. Call ORCA by its full path. Use $(which orca) rather than a bare orca. ORCA locates its own helper programs relative to how it was invoked, so a bare call works for serial jobs and fails as soon as you use %pal.
4. Keep module purge in the script. Your job inherits the environment of the shell you submitted from. Without the purge, a module already active in that shell carries into the job and your module load orca silently does nothing.

A minimal input to test with:

! HF def2-SVP
%pal nprocs 4 end
* xyz 0 1
O   0.0000   0.0000   0.0626
H  -0.7920   0.0000  -0.4973
H   0.7920   0.0000  -0.4973
*

This finishes in seconds and ends with ****ORCA TERMINATED NORMALLY**** in input.out.

Limitations

• CPU only. ORCA 6.1.1 has no GPU support, so submit to short, med or long rather than the GPU partitions.
• Single node. Keep --nodes=1. Multi-node ORCA is not currently supported.
• No DMRG. The DMRG interface is not included in this installation.