> Jeff Kell saith:
> I'll leave the extrapolation of a 995 from a 960 to someone else :-)

...so I'll take a stab at it.

Let's see:

                  Benchmark   Relative
Model             CPU secs.   Performance
---------------   ---------   -----------
HP3000/42           313.90            1.3
HP3000/58           248.66            1.7
HP3000/950           22.011           6.5
HP3000/960            9.156          14.7
HP3000/995-400        x             118.0

("Relative Performance" numbers are from HP.)

First, test the "Relative Performance" numbers:

(firstCpuSecs*firstRelPerf)/secondRelPerf = secondCpuSecs

42->58: (313.90*1.3)/1.7 = 240.041 (measured 248.66)
950->960: (22.011*6.5)/14.7 = 9.733 (measured 9.156)

Not too far off -- good enough for our purposes.
Let project the CPU seconds the benchmark would take on the 995-400,
using both the 950 and 960 measurements:

950->995-400: (22.01*6.5)/118 = 1.212
960->995-400: (9.156*14.7)/118 = 1.141

So the 950 benchmark predicts the 995-400 will take 1.212 CPU
seconds, and the 960 benchmark predicts the 995-400 will take 1.141
CPU seconds.

Pretty close to each other.

Finally, the IBM 4381-2 relative performance can be
calculated by: (hpCpuSecs*hpRelPerf)/ibmCpuSecs=ibmRelPerf

Using each of the measured HPs, we get:

950->4381-2: (22.06*6.5)/30.22=4.7
960->4381.2: (9.156*14.7)/30.22=4.5

950->3083-BX: (22.06*6.5)/16.91=8.5
960->3083-BX: (9.156*14.7)/16.91=8.0

950->3090-400: (22.06*6.5)/4.96=28.9
960->3090-400: (9.156*14.7)/4.96=27.1

In conclusion, here is a fairly meaningless relative performance
table:

HP 3000/950        6.5
IBM 4381-2         4.7
IBM 3083-BX        8.5
HP 3000/960       14.7
IBM 3090-400      28.9
HP 3000/995-400  118.0

(I use the numbers most favourable to IBM in all cases.)

Enjoy!
---
Michael D. Hensley             | [log in to unmask] (personal)
Software Development Manager   | [log in to unmask]    (business)
Lund Performance Solutions     | http://www.lund.com