John Schick wrote:
> Can anyone provide me with a rough estimate comparing the system
> performance of an HP3000 995/400 with an IBM 3090-500J?
I attempted this many moons ago (1988) and just dug this out of my
dusty archives:
----------------------------------------------------------------------
Tests were run using the Fortran Whetstone benchmark source from the
National Bureau of Standards library (NBSLIB). Constants were set to
invoke 100 million Whetstones rather than the supplied 1 million so
that timings on faster machines would be less prone to variance. The
IBM timings were obtained from total time reported on the CMS prompt
at completion. HP timings were obtained from PROCTIME intrinsic calls
on either end of the benchmark.
------------------------------------------------------------------------
| CPU tested | Total CPU sec. | Compiler and commentary |
------------------------------------------------------------------------
| HP3000 Series III | 429.19 | FORTRAN/66 (FTN77 unavailable) |
| | | |
| HP3000 Series 42 | 313.09 | FORTRAN/77 |
| | | |
| HP3000 Series 58 | 248.66 | FORTRAN/66 (for comparison) |
| | 222.08 | FORTRAN/77 |
| | | |
| IBM 4381-2 | 40.26 | FORT-G |
| | 37.11 | FORTVS opt level 0 (lowest) |
| | 30.22 | FORTVS opt level 3 (highest) |
| | | |
| HP3000 Series 950 | 28.32 | FTNXL opt level 0 (lowest) |
| | 22.06 | FTNXL opt level 2 (highest) |
| | | |
| IBM 3083-BX | 16.91 | FORTVS opt level 0 (lowest) |
| | 12.99 | FORTVS opt level 2 (best time) |
| | | |
| IBM 3090-400 | 4.96 | FORTVS opt level 3 |
|----------------------------------------------------------------------|
Last minute addendum - I found the benchmark source code and re-ran it
on our 960. CPU time was 9.156 seconds. And for kicks I moved the
resulting object code to our 950 and validated the 950 figure above
(actually I got 22.011 secs).
I'll leave the extrapolation of a 995 from a 960 to someone else :-)
Jeff Kell <[log in to unmask]>
