Table of Contents
Gaussian98 Performance Guide on a Heterogeneous Environment
Agenda
Motivation
Gaussian Architecture
Shared and Distributed Memory Parallel Versions
Link 0 Prallel Command
Shared Memory Parallel Input
Distributed Memory Parallel Input
Gaussian Usage
Algorithm
System Resources
Parallelization of two-electron Integrals
Parallelized Links
CRAY SV1 Architectural Features
Memory Allocations for Parallel Runs
Performance
Single CPU performance as a Function of Basis Sets
Single CPU performance as a Function of the Number of Atoms
a-pinene: Scalability as a Function of Basis Sets
a-pinene: Scalability as a Function of Dunning’s Basis Sets
Scalability as a Function of the Size of the System
C20H42: Scalability as a Function of Molecular Symmetry
Differences in Scalability Between HF and B3-LYP
Total versus Individual Links Speedups
C20H42: MP2 Single CPU Performance
MP2 as a Function of Basis Sets
MP4 Calculations: Mflops as a Function of the Basis Sets
Summary
Future Work
Gaussian Information
|
Author: Carlos P. Sosa (CRAY) and Michael Frisch, Gaussian, Inc.
Email: cpsosa@cray.com |