From owner-freebsd-hackers  Wed Jul  3 08:38:18 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id IAA17584
          for hackers-outgoing; Wed, 3 Jul 1996 08:38:18 -0700 (PDT)
Received: from p54c.spnet.com (p54c.spnet.com [204.156.130.11])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id IAA17579
          for <hackers@freebsd.org>; Wed, 3 Jul 1996 08:38:13 -0700 (PDT)
Received: from localhost (localhost [127.0.0.1]) by p54c.spnet.com (8.7.5/8.6.9) with SMTP id IAA08333; Wed, 3 Jul 1996 08:38:10 -0700 (PDT)
Message-Id: <199607031538.IAA08333@p54c.spnet.com>
X-Authentication-Warning: p54c.spnet.com: Host localhost [127.0.0.1] didn't use HELO protocol
To: hackers@freebsd.org
Cc: elh_fbsd@spnet.com
Subject: Re: SIG's 11 and 6... 
In-reply-to: Your message of "Tue, 02 Jul 1996 20:09:27 CDT."
             <m0ubGRb-000VyVC@eagle.ais.net> 
Date: Wed, 03 Jul 1996 08:38:10 -0700
From: Ed Hudson <elh_fbsd@spnet.com>
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk


[...]
> HEAT! My pentium 120 was having the above problems.  Switching out memory
> would solve them for a day or two, but the problems would then start to
> build up again.  FINALLY I replaced the CPU fan and added a big waffle fan
> in the front of the case and VOILA...  No more SIG's for the last week.
> Chances are that in the past the time involved in cracking the case and


howdy.

	simple cmos transistor circuits have 'speed' variation of
	order t**-1.5 (t absolute).  'weak' circuits (eg rams) can
	have much larger variation with temperature, and two chips
	interacting with each other at a fixed cycle time can see
	their variations add.

	if you have a hardware bios or jumpers with tuneable cache/memory/bus
	speeds and you 'tune' them on a cold day you can easily get
	into trouble on a nice warm summer day - a 20deg f (11deg k)
	ambient temperature variation alone can affect peak operating
	speed by 10-20% easily.  backing off on speed (this is not
	always cpu hz, but sometimes memory/cache clock cycles) is
	a very rational check for hardware timing problems - establish
	functionality first, then optimize speed.

	if you try BLOCKING your input air ducts for an hour or two
	(a less traumatic way of making things warmer than removing the
	cpu fan), you may see the problems get worse.  (of course,
	don't do this on a system with ir-recoverable data!  back
	it up first. beware: i've had disks go into thermal shutdown
	and destroy data, and have seen power supplies die and kill
	hardware when thermally overloaded).

	complex, multi-tasking operating systems can stress many paths
	in a chip/system not exercised by dos systems (less true with
	windows 95).  in the old days unix used to break many machines
	that ran dos happily.  an analogous circumstance perhaps exists
	today with speed paths (i never run dos or windows, thanks to
	you folks, so i'm not certain about their crash rates).

		-elh (former cpu/ram designer).