The EMSL Ab Initio Methods Benchmark Report: A Measure of Hardware and Software Performance in the Area of Electronic Structure Methods
Contributors to previous versions: R. A. Kendall and M. J. Brightman
Molecular Science Computing Facility
William R. Wiley Environmental Molecular Sciences Laboratory
Pacific Northwest National Laboratory(a)
Richland, Washington 99352
|1.||Representative Improvements in Floating Point Performance for Workstations|
|2.||Performance of the CRAY T3D Parallel Processor on Direct Hartree-Fock Calculations|
|3.||Representative Gaussian 92 Hartree-Fock CPU Times for a Variety of Workstations and Mainframes|
|4.||Representative MP2 and MP4 CPU Times For a Variety of Workstations and Mainframes; Obtained With Gaussian 92 and MOLPRO 94 Running the Ethylene 6-311++G** Test Case|
|5.||Growth in RHF/STO-3G Times as a Function of the Number of Carbons in a Hydrocarbon Chain|
|6.||Growth in RHF/6-311++G** Times as a Function of the Number of Carbons in a Hydrocarbon Chain|
|7.||Speedup Curve for GAMESS(US) Running on the KSR/2|
|8.||Speedup Curve for GAMESS(US) Running on the Intel Pargon|
|9.||Log-based Timing Comparison for GAMESS(US) Running on Three Systems|
|10.||Wall clock times for direct Hartree-Fock calculations on an SGI PowerChallenge|
|11.||Wall clock times for direct Hartree-Fock + gradient calculations on an SGI PowerChallenge with 75 MHz R8000 processors.|
|1.||Molecules and Basis Sets|
|3.||The List of Ab Initio Programs|
|4.||The List of Computers Included in the Benchmark Study|
|5.||Relative Performance Averaged Over the 16 Ethylene 6-311++G** Tests Using Gaussian and Normalized to the SPARCstation2|
|6-8.||Sun Timings (Postscript) (Acrobat)|
9. IBM PowerPC 250 Timings (Postscript) (Acrobat)
10. IBM RS/6000 340 Timings (Postscript) (Acrobat)
11. IBM RS/6000 370 Timings (Postscript) (Acrobat)
12. IBM RS/6000 390 Timings (Postscript) (Acrobat)
13. IBM RS/6000 550 Timings (Postscript) (Acrobat)
14. IBM RS/6000 580 Timings (Postscript) (Acrobat)
15. IBM RS/6000 590 Timings (Postscript) (Acrobat)
16. IBM RS/6000 591 Timings (Postscript) (Acrobat)
|17-19.||HP Timings (Postscript) (Acrobat)|
|20-27.||SGI Timings (Postscript) (Acrobat)|
|28-32.||DEC Timings (Postscript) (Acrobat)|
|33-37.||Intel Timings (Postscript) (Acrobat)|
|38.||FUJITSU Timings (Postscript) (Acrobat)|
|39-43.||CRAY Timings (Postscript) (Acrobat)|
|44-46.||GAMESS(US) Timings (Postscript) (Acrobat)|
|47.||SUPERMOLECULE 1.08 Timings (Postscript) (Acrobat)|
|48-51.||SGI Multiprocessor Timings (Postscript) (Acrobat)|
In the intervening two years many new workstations and parallel systems have been introduced and software packages continue to be updated. We have, therefore, continued to monitor performance and to extend the tables of timing data found in the first report.
We encouraged contributions from people outside of this laboratory and are grateful to everyone who responded. A reader of the current version will notice many footnoted entries referencing these contributions. The intent of the report remains the same, to assist scientists and administrators who have to make decisions regarding computers and electronic structure packages in judging which hardware/software combinations are the best suited for their needs.
The timing data included in this report are subject to all the normal failures, omissions, and errors that accompany any human activity. In an attempt to mimic the manner in which calculations are typically performed, we have run the calculations with as many defaults as each program would allow and a near minimum amount of memory. This typically may not produce the fastest performance possible. It is not known to what extent improved timings could be obtained for each code by varying the run parameters. If sufficient interest exists, it might be possible to compile a second list of timing data corresponding to the fastest observed performance from each application, using an unrestricted set of input parameters. Improvements in I/O might have been possible by fine tuning the Unix kernel, but we resisted the temptation to make changes to the operating system.
Due to the large number of possible variations in levels of operating system, compilers, speed of disks and memory, versions of applications, etc. readers of this report may not be able to exactly reproduce the times indicated. Copies of the output files from individual runs are available should questions arise about a particular set of timings.
The results reported here should not be misconstrued as an endorsement by Battelle for any particular software package or computer.
We also thank the EMSL computer support staff (B. J. Pitre, K. A. Anderson, N. Nanni, S. Campbell and J. S. Eckert) for their usual professional job of keeping the machines running so that these benchmarks were possible. Special thanks are given to A. K. Lebold and G. R. Elliott for help in setting up the WWW version of the report.
Ms. Betsy Foote and Mr. Alan Riddle of the National Energy Research Supercomputer Center and Dr. Tony Hess of PNNL are thanked for their assistance in scheduling the parallel SGI PowerChallenge benchmarks. Silicon Graphics, Inc. is gratefully acknowledged for making the two machines available.
Digital Equipment Corporation is thanked for providing early access to the DEC AlphaStation 600 5/266.
This research was funded, in part, by the Environmental Molecular Sciences Laboratory construction project and by the Division of Chemical Sciences, Office of Basic Energy Sciences, U.S. Department of Energy under Contract DE-AC06-76RLO 1830 with Battelle Memorial Institute, which operates the Pacific Northwest National Laboratory. We also wish to thank the Scientific Computing Staff, Office of Energy Research, U.S. Department of Energy for a grant of computer time at the National Energy Research Supercomputer Center.
|(a)||The Pacific Northwest National Laboratory (PNNL) is a multiprogram national laboratory operated for the U.S. Department of Energy by Battelle Memorial Institute under contract DE-AC06-76RLO 1830.|