Ronald Khoo wrote:
> 
> > As best as I can tell, you have to get a 400Mhz PII to match
> > a 200Mhz PentPro.

My measure was "off the top of my head with usage of what I do
which is a lot of compiles (which mixes short bursty I/O with
CPU intensive) plus occasional CPU soaker programs".

..
> 
> So either:
> 
>         a) You're talking about a measure of performance that SPECint95
>            fails to measure
> 
>         b) You're talking about the 1Mb cache PPros that I can't find
>            on intel's web site (and are extremely expensive)
> 
>         c) Intel's compilers are supergood or cheating :-)
> 
> Which is it ?  If (a) does that then apply to mouse's application mix,
> and if (b) then does that then applly to mouse's budget ?

#b since you get pretty reasonable bcache's with most EV5
(alpha) implementations- that's factored into the cost notions
I have.

Also, I didn't hear der Mouse saying cost was *the* factor. I
took his mail to be "rough equivalents". If cost was *the* factor,
there are a number of other possibilities- including accepting
somewhat but not a lot less performance (the performance curve
is not linear with cost). I heard him (loose translation) ask for
"The best mostly CPU performance with non-flakey h/w for the
best cost". The latter clause implies that cost alone isn't the
criteria, therefore a fuzz factor of, oh, say 20% is not
unreasonable in saying "Damnit- they're both lemons, but the
alpha makes decent lemonade at least".

kapeesh?