Date: Tue, 24 Sep 1996 00:18:59 +0200 (MET DST) From: "Julian H. Stacey" <jhs@freebsd.org> To: hardware@freebsd.org Cc: gj@freebsd.org Subject: cache size in cmos Message-ID: <199609232218.AAA11775@vector.jhs.no_domain>
next in thread | raw e-mail | index | archive | help
This is sent:
To: hardware@freebsd.org
bcc: doc@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@ , but suggest follow up on just hardware@
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@vector:/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@freebsd.org http://www.freebsd.org/~jhs/
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199609232218.AAA11775>