Gaussian 09

by Espen Tangen — last modified Jan 13, 2012 03:12 PM

Information on use of Gaussian 09 software on stallo

This page describes how to submit serial and parallel Gaussian 09 calculations on stallo.

Gaussian 09 on Stallo

Job Submission Guidelines

Links

For help with inputfiles, see: Gaussian 09 Inputfiles

For information on Gaussview for g09, see: Gaussview for g09

GAUSSIAN 09 ON STALLO

The default version of Gaussian 09 is the minor revision b.01, which is currently the default g09 version both on Stallo and Titan (Oslo).

Most of the features regarding running Gaussian are analogous for g09 and g03, with some rather important exceptions:

Headings in inputfiles for parallel runs.
Content in the .tsnet.config file.
Content in the g09parallel.job script.
Advised number of nodes and cpus in parallel runs.

Please do read these sections carefully. For a thorough update on differences between g09 and g03, please look here.

Before you do start using Gaussian 09, we encourage you to take a look at these pages at the Gaussian web site:

Information on program limits for Gaussian 09: http://www.gaussian.com/g_tech/g_ur/b_proglimits.htm

Information on efficiency considerations for Gaussian 09:http://www.gaussian.com/g_tech/g_ur/m_eff.htm

To run Gaussian on stallo, the module has to be loaded:

$ module load g09

If you download a run script (see below), this will be taken care of by the script.

JOB SUBMISSION GUIDELINES

Download the run script you need and move the file to your $HOME/bin directory (if necessary create directory first).

For serial jobs: Script file for Gaussian 09 serial jobs
For parallel jobs:Script file for Gaussian 09 parallel jobs

Manually change EMAIL and ACCOUNT to something more relevant before job submission.
Submit job (s) from any directory: (see below)

Running serial jobs:

Job submission (example with water.inp and 15 minutes of runtime):

$ g09serial.job water 15

Running parallel jobs:

Job submission (example with file waterdimer.inp, 8 processors per node, 2 nodes, 20 hours of runtime):

$ g09parallel.job water 8 2 20:00

See also http://www.gaussian.com/g_tech/g_ur/m_linda.htm for info on parallel running of g09.

Restarting a g09 job from a g03 checkpoint-file:

To do this, you need to convert the g03 .chk file to g09 .chk file using a script called c8609 in the g09 folder.

Using global reference, it would look like this on Stallo: /global/apps/gaussian/g09.b01/g09/c8609 water.chk.

NOTE:

For g09 on Stallo, parallel calculations does not require additional keywords in the input file - this is taken care of by the runscript file. This means that input of serial and parallel runs are similar in their header.
If you plan to run on >1 node (using %NProcLinda), make a new file in your $HOME directory called .tsnet.config containing only the line: Tsnet.Node.lindarsharg:/global/apps/bin/pbsdshwrapper.py

Note: if you run >1 node (using %NProcLinda), you have to use >1 processor per node

RECOMMENDATIONS:

If you are running geometry optimizations using DFT methods, we recommends using hybrid functionals. For pure functionals (no HF exchange), unless you have tested it (and see a benefit), it is recommended to run Gaussian on not more than 3 nodes only, with between 6 and 8 processors on Stallo. For hybrid functionals, the time gain is present up to 12 nodes, running 7 cpus on each node. This holds for the latest revision, g09.c01. If you are using the default version (g09.b01) this is valid up to 8 nodes. The default version is recommend up to and including 8 nodes unless you plan do run features that does not come with the b01 revision. Above 8 nodes revision c01 is recommended both regarding speed gain and new features. The pure functionals has not been tested, but there is probably an increase in scaling here also. New info will soon replace this!
If your molecule is >60 atoms, Gaussian uses FMM by default (which is not parallelized), so do not run parallel in this case, unless you turn off FMM (NoFMM keyword). FMM is enabled for non-symmetric molecules with 60 atoms or more. For molecules with high symmetry this limit is 240 atoms for hybrid DFT and 360 atoms for pure DFT.
(The text below is copied from http://www.gaussian.com/g_tech/g_ur/m_linda.htm):

HF, CIS=Direct, and DFT calculations on molecules are Linda parallel, including energies, optimizations and frequencies. TDDFT energies and gradients and MP2 energies and gradients are also Linda parallel. Portions of MP2 frequency and CCSD calculations are Linda parallel, but others are only SMP-parallel, so they see some speedup from using a few nodes but no further improvement from larger numbers of nodes.

It is always best to use SMP-parallelism within nodes and Linda only between nodes. For example on a cluster of 4 nodes, each with a dual quad-core EM64T, one should use
```
%NProcShared=8
%LindaWorkers=node1,node2,node3,node4
```
rather than using more than one Linda worker per node. (Do not run more than one thread/cpu. It works very poorly on Stallo!)

For the SMP-parallel jobs, you are advised to typically fill or almost fill the node (for Stallo we would at the moment advise 7 of 8 cpu/node utilizied for gaussian jobs unless there are extreme memory demands).

Additional comments:

Output (.out) and checkpointfiles (.chk) are transported to $HOME upon job completion.
All temporary directories are removed upon job completion. If jobs abort prematurely, temporary directories need to be removed manually.
To submit a .chk file along with an .inp file and/or retrieve a .chk upon job completion, make sure the .chk has the same name as the input file (e.g. filenames waterdimer.inp and water.chk, with keyword added to .inp: %chk=water).
RESTART OF JOBS: Retrieve the .chk file from the temporary directory and add the restart command to the input (opt=restart or scf=restart, depending on job). Make sure .chk and .inp have the same name. Submit as usual.

Sections

Personal tools

Gaussian 09

GAUSSIAN 09 ON STALLO

JOB SUBMISSION GUIDELINES

LINKS

Document Actions