Subject: Re: probing CPU speed?
To: Peter Bentley <peter.bentley@nomura.co.uk>
From: Andrew Gillham <gillhaa@ghost.whirlpool.com>
List: port-i386
Date: 11/20/1998 11:04:05
Peter Bentley writes:
> 
> As cgd said, why would you care about raw CPU MHz speed for that?  You
> wanna know which boxes crack RC5 fastest? Crack some keys on 'em and use
> *that* as your relative performance benchmark.  It'll be a sight more
> representative than raw CPU speed.  

Uhmm, there are two different requests in this thread.  One for some
kind of "BogoMIPS" type rating, and one for the apparent clock speed
of the CPU.  I don't care particularly for the BogoMIPS rating, but
I definitely would like to know what kind of CPU is in the box.
Why should a sparc print this info?
...
mainbus0 (root): SUNW,SPARCstation-10
cpu0 at mainbus0: TMS390Z50 v1 @ 36 MHz, on-chip FPU
cpu0: physical 20K instruction (64 b/l), 16K data (32 b/l): cache enabled
...

Isn't this "useless" to know the MHz of the CPU?  Why does the Alpha
port print out the CPU speed?  (and I'm sure several other ports I don't
own would)  Because it is *easily* determined for these ports, and is
somewhat integral to the platform.  The i386 port prints out the family
of the CPU, and the text name, and that is it.  Why should I have to
open the case, pull the CPU, remove the fan and heatsink, clean off the
heatsink compound just to read the speed of the CPU?  Hell, why should
I have to reboot my box to see what the BIOS thinks?  What's wrong with
printing the info out?  Put a big "THIS IS A WILD ASS GUESS, DON'T YOU
DARE EVEN THINK ABOUT THIS NUMBER MEANING ANYTHING" before it.  That might
eliminate the "Hey, my 386SX is running at 1666MHz, cool!" problem someone
mentioned.  (which is actually a problem on a friends Alpha, it is a 
166MHz part, put NetBSD reports it as 500MHz)

> The deceptiveness has nothing to do with the difficulty of obtaining
> accurate MHz numbers, it's because you're measuring something totally
> different.  For load balancing calculations, even a trivial userland
> benchmark like Dhrystone is probably better than CPU speed.  Some sample
> figures from the actual application you want to run are always going to
> be better though...

I think you're addressing the "BogoMIPS" problem, not the "What the
hell kind of CPU do I have?"

-Andrew
-- 
-----------------------------------------------------------------
Andrew Gillham                            | This space left blank
gillham@whirlpool.com                     | inadvertently.
I speak for myself, not for my employer.  | Contact the publisher.