Re: CCL:SUN vs HP workstations



 >
 > >    First of all you may take into account usual benchmarks data as
 > > SPECint, SPECfp and LINPACK MFLOPS. There are:
 > >    workstation         SPECint92               SPECfp92
 > > HP9000 712/60          58                         79
 > > SUNstation 20 mdl 50   69,2                       78,3
 > >
 > It might be worth noting that the 712/80 (80MHz PA-1.1) is
 > rated by HP at the same SPECfp92 as the 712/60 (60MHz PA-1.1).
    The data for HP 712/60 was taken from *independed* sources,
 from DEC on-line service in partition containing  competitive
 comparison of DEC3000 AXP Alpha mdl 300LX with HP 712/60 and
 SGI INDY R4000PC.
    Yes, I found that the data for HP 712/80 taken from the same
  source (comparison of DEC3000 AXP mdl 300X with HP712/80 and
 SGI R4000SC) are : SPECfp92 = 79 (! - the same as for 712/60)
 and SPECint92 = 84.
    Therefore I think that this data contain mistake.
 > There has been evidence that this (quite improbable) result
 > was achieved by intentionally slowing the benchmarks for the
 > faster machine, in order to protect the more expensive 7?5
 > line of technical workstations. Indeed, when we run MD on our
 > 712/80, we find that its performance is more consistent with a
 > SPECfp92 rating of above 100, just what you would expect by
 > simple clock rate extrapolation of either the 735 or the 712/60.
    Unfortunally I don't have HP712 and can't measure  SPECfp :-(
  but I may agree with this estimation if I'll think that SPECint
  data are correct. If it's so then SPECfp92 for HP712/80 is about
  SPECint92 * 1.36 (specfp/specint ratio for HP712/60) what give
  the value much more high than 100.
     Clock rate extrapolation may give bad approximation if the
 cache size for HP712/60 and /80 is different (I don't know is't
 so or no).
 > From my experience, HP is faster than anything else
 > on the market, and they are pretty nice machines, too.
    I agree that HP is excellent family. The main todays advantage
 of SUN station 20 is multiprocessing, not CPU itself.
 Mikhail Kuzminsky