A/UX: System Benchmark Results using "lloops"


Article Change History
----------------------
08/31/92 - REVIEWED
* For technical accuracy.
01/20/93 - UPDATED
* Vendor information.


This article describes the floating-point, computational performance of an
A/UX system. The benchmark test run was "lloops," a suite of benchmarks,
developed at Lawrence Livermore Labs, and written in FORTRAN. In this case,
the suites were written both in NKR Research's (now PenWare, Inc.) Optimizing
FORTRAN and f77 (shipped with A/UX) compilers.

The test system was a Macintosh IIcx with 4MB RAM and a HD80 SC hard drive.
All tests were done in single-user mode with all networking and the toolbox
turned off. All calculations were made on data defined as REAL*8. Because
"lloops" contains a variety of common scientific and math program segments,
floating point performance varies by "sub-test". Overall results (using
the NKR compiler) were as follows:

Mflops range: 0.0364 to 0.3148 Mflops/sec
Harmonic mean: 0.1189 Mflops/sec
Median rate: 0.1319 Mflops/sec +/- 0.0577 Mflops/sec
Average rate: 0.1496 Mflops/sec +/- 0.0668 Mflops/sec

Results using the f77 were lower by 30% to 60%--presumably because of the
lack of an object optimizer with f77.




Published Date: Feb 18, 2012