John writes:

> Believe it or not, there is some misinformation in this thread! You see,
>  as far as OFFICIAL HP numbers, there are two scales that have been
>  published over the years. One is RELATIVE PERFORMANCE which compares all
>  models to a base model. Another is PERFORMANCE UNITS, which puts all
>  systems on a relative scale. Yes, Virginia, there is a difference. In
>  1998, HP started publishing Performance units for all models. The
>  official numbers DO differ on system models where both scales had been
>  used. Particularly in the 9x9 area. The 987/200 you ask about was never
>  rated on the performance unit scale by HP. The Relative performance,
>  however, was 6.0.
>
>  However, some third party vendors have taken to publishing their own
>  estimates of performance based on their own experience.

While there is a difference, as John says, mostly the difference is one of
remeasurement or remarketing. While some third party vendors (and we being
one of them) are renormalizing the new HP performance so as to match the old
numbers, there isn't really very much magic in doing that. But the benefit
you get by doing that is that you can compare in one swell foop the
performances of any two HP3000s ever manufactured, Classic or RISC.

The conversion process no more complicated than converting kilometers to
miles. You still have walk the same distance, regardless of how you measure
it. And CPUs run just a fast (or as slow) in the new numbers as they do in
the old.

If there is any confusion in the normalizing process, it is that the
conversion has to be accomplished using a semi-metric in a not-quite
completely Euclidean hyperspace. In English, all that means is that there is
some inconsistency in HP's published numbers. When you compare a system to
one specific series, you get one performance ratio. When you compare the same
system to another series, you get a slightly different set of numbers. What
you do at that point is call in James Baker, who will demand that you stop
thinking about what you're doing and just pick a number that seems about
right :-).

Nonetheless, there isn't all that much inconsistency and the numbers do tend
to generally agree fairly well. Not that much interpretation in the
normalization process is required to truly determine the relative performance
ratios of any two systems, old to new or new to new.

Wirt Atmar