Subject: Re: probing CPU speed?
To: David Brownlee <abs@anim.dreamworks.com>
From: Chris G. Demetriou <cgd@netbsd.org>
List: port-i386
Date: 11/18/1998 18:28:16
David Brownlee <abs@anim.dreamworks.com> writes:
> 	It may be dead weight, but it would be nice to get a rough
> 	rating of CPU 'speed' across ports, something akin to (dare
> 	I mention it) Linux's bogomips. Even if you're not printing
> 	out Mhz on the CPU, a bogomips (or similar) rating would
> 	help you match up CPU to speed.

*sigh*

And for what purpose would it be nice?

Such a number (as with CPU speed) is useful only if you know exactly
what generates that number.  It says nothing useful about any metric
of performance other than the performance of "a certain instruction
sequence, executed in the CPU's on-chip cache (if any)."

It doesn't matter if your CPU is nice, if any of the rest of the
components of the system are in any way sub-par.

Take the Alpha LCA (2106[28]) family...  The CPU, running from the
on-chip cache, would appear to not suck.  do anything else, and you
lose.

Or look at the difference between the high-end TurboChannel Alphas and
the alphastation 200 and 400 systems.  Same or faster CPU speed in the
latter than in the former, but actual overall system performance was
worse in the latter because of worse memory bandwidth.


Who _cares_ what the raw CPU speed is?  It's at best useless, at worst
seriously deceptive...


cgd
-- 
Chris Demetriou - cgd@netbsd.org - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.