Product Information
Last update: 12 April 2007

 

Computers Working  


Run Gaussian Calculations in Parallel
on Windows, Linux, Unix and Mac OS X Systems.

Get Your Gaussian Results Sooner

No matter how much we speed up the performance of Gaussian, users always want to get their results faster, or run bigger jobs, or both. Chemists running electronic structure calculations seem to have an insatiable need for CPU resources.

If you're already running the latest version of Gaussian, there may still be ways to get your jobs done more quickly, without having to purchase faster hardware. As of revision D, Gaussian can run many kinds of jobs in parallel on both multiprocessor computer systems and across networks of separate workstations (and the two approaches can also be combined in most environments). Network parallel execution allows you to harness untapped CPU power you already have (for example, computer labs after hours).

In general, parallel execution speeds up calculation by applying the resources of more than one CPU to the job:

  • Multiprocessor versions of Gaussian run across the multiple CPUs (or virtual CPUs in dual core systems) within a single computer system.

  • Network parallel versions of Gaussian run across multiple computers connected in a local area network (LAN). The additional software package TCP Linda is required for this type of parallel computing (available from Gaussian at an additional charge).

Sample Performance Results

The following chart shows the speedups obtained for a calculation on Alpha Pinene on successively more processors of a multiprocessor system:

Multiprocessor Performance

The following chart shows the speedups for a VCD calculation on BiPhenyl Tetramer for increasing numbers of systems in a network parallel calculation (using Gigabit Ethernet networking):

Network Parallel Results

Achieved speedups vary by the job size, job type and specific characteristics of the hardware environment.

What Methods Run in Parallel?

  • HF and DFT Energies, Gradients and Frequencies
  • MP2 Energies and Gradients
  • MP2 Frequencies in SMP environments only
  • Coupled Cluster Energies and Gradients in SMP environments only
  • CIS and Energies, Gradients and Frequences
  • TD-DFT Energies

Available Parallel Versions of Gaussian

The following table indicates what parallel environments are supported by Gaussian 03 Revision D.01 and later:

Environment Multiprocessor Version Available? Network Parallel Available?
Windows Yes (limited to 4 processors/cores) Yes
Linux/UNIX Yes Yes
Mac OS X No, due to limitations inherent in current versions of Mac OS X. However, multiprocessor systems can participate as nodes in network parallel calculations. Yes

Additional Notes

  • On Linux and Unix systems, there is no distinction between the single processor and multiprocessor versions of Gaussian. The same software can be used for both environments. For Windows systems, multiprocessor support is provided by a different software version, available for an additional license fee.

  • You must have a site license for the corresponding version of Gaussian 03 (or G03M or G03W) in order to license TCP-Linda for that platform.

Supported Platforms

The supported workstations for TCP Linda network parallel execution are:

  • AMD Opteron / Intel EM64T running Linux
  • HP Alpha/Tru64
  • HP Alpha/Linux
  • IBM RS/6000 Power 3, 4 or 5 (incl. the SP series), running AIX 5.1 or 5.2
  • Intel Pentium II, III and IV and AMD Athlon/Duron (IA32) running Linux
  • Intel Itanium2 (IA64) running Linux
  • Sun Ultrasparc running Solaris 9/10
  • Macintosh G3, G4 or G5 running Mac OS X version 10.2 or later
  • Microsoft Windows 2000, XP or Server 2003

    See the Gaussian 03 platform requirements for details on the supported Linux versions.

Pricing Information

Linux/Unix pricing information:  Academic     Commercial      U.S. Government
Mac OS X pricing information:  Academic     Commercial     U.S. Government

Windows pricing information: Academic    Commercial    U.S. Government
Maintenance is available for Gaussian and for TCP Linda.