To: hardware@_ERASE_freebsd.org
bcc: doc@_ERASE_freebsd.org
cc: gj@_ERASE_freebsd.org
Subject: cache size in cmos 
From: Julian H. Stacey <jhs@

This is sent:
	To: hardware@_ERASE_freebsd.org
	bcc: doc@_ERASE_freebsd.org

I had an experience with an old AMD BIOS that seems worth telling...

The lesson I learned:
	If a machine seems really slow, try reloading cmos defaults,
	(after taking careful note of current settings  :-)
Someone might want to put that tip in the currently empty (as of Sep 1)
section 10.3.4 of the handbook. 
Hence I bcc'd this to doc@_ERASE_ ,  but suggest follow up on just hardware@_ERASE_


Why the CMOS Reset ? Well ...

I had a hardware problem on a newish (to me) system built from old
motherboard, old discs etc, & with loose cache chips,
I took cache chips out & I guess I ran that system without to deduce
if it was the chips causing my previous problem, anyway I cured the 
cache chip loose socket problem,

But I think the cmos resized itself as having no cache (it doesnt show
cache chip availability on my old AMD BIOS, I just have the M/board DIPS
to set), & when I put the cache chips back in, the 
system worked, but was slow as treacle, (but I didnt notice as I was chasing
other problems on the system, (like synch/asynch on old HP drives,
& data corruption).

It also, (after the last surgery on board, & until today) used to ignore
the turbo switch (which actually acts not as a frequency controller,
but as a cache enable/disable switch), & the `spinner' at kernel load time
didnt spin, it trudged along slowly.

I decided to track down the `treacle' today, 
I finally (at power time) told it (the AMD bios from DEL) to reload power
(slow) defaults & then bios (faster) defaults,

It used to take 120 secs to do 
	cd usr/src/*/ls ; make
it now takes 
	turbo on        19.20 real        16.66 user         1.71 sys 
	turbo off       51.48 real        45.97 user         4.51 sys 
the above with sh & then time, & 2nd compile, 'cos 1st loads stuff into
freebsd kernel cache (ie disc to ram) so 1st is always slower
than #2 & #3 make.

Oh, BTW I had to disable F000 64K cache else reboot ceased to work.
(The reload with BIOS defaults had enabled it)

The environment
	33MHz Intel 486 Motherboard `Gigabyte' `GA-486US' 256K cache.
	UM82C481A, USA, 9140KV016
        UM82C482A, USA, 9138KV001
        UM82C206F, 9142-C9, C82093
	Copyright American Megatrends Inc., 
                40-0500-D91199-00101111-050591-UMCWB-F
	various CLK/2 & CLK/5 madse no real difference,
	A half week old current that had just finished make world
	(took about half a week or so (Really ! that's why I investigated))
	& kernel that reports
	FreeBSD 2.2-CURRENT #0: Fri Sep 13 22:38:55 MSZ 1996
    	jhs@:/usr2/src/cur-960901/sys/compile/GATE
	& with an /etc/make.conf 	CFLAGS= -O2 -m486 -pipe
	& cc --version	2.6.3
	System idle other than test running.

Well, that system is 6 times faster now :-)
If this tale has merely amused you, it's been worth it,
the more so if it digs someone out of a similar hole.

Julian
---
Julian H. Stacey	jhs@