Hi everyone,
The latest BOINC UK newsletter
(
http://www.boincuk.com/newsLetter/bunc-03.pdf 212 K) has an interesting
article in the 7th page. It's about keeping the CPU cool, and there's a
section about lowering the CPU voltage in the BIOS.
First, are you having trouble keeping it cool? What is the
purpose for this change, the realized benefit you expect
from lowering the voltage? I'm not claiming there aren't
any potential benefits, but a bit more detail about your
specific goal might help.
The most thorough way of doing it is to plot a graph, of
voltage vs. CPU speed. Through trial and error you'd find
the max speed the CPU can run stabily (Prime95's Torture
Test, Large In-Place FFTs setting ran for at least several
hours or terminated for readjustment of settings immediately
if an error is found) at each voltage setting available.
Having this data you can decide what tradeoff to make. Keep
in mind that your board's KT333 chipset doesn't have an AGP
(and PCI) lock on the FSB, so you need to stay fairly close
to the 100/133/166 FSB speeds your board supports (though
166 is unofficial, at 166 and especially going towards
175MHz FSB you may find your AGP video cards start acting
flaky). Point being, it would be somewhat arbitrary to only
think about keeping the CPU at it's present speed and
lowering the voltage as low as it can go while remaining
stable. It might even be possible to raise the CPU speed,
overclock slightly, while still lowering the voltage and
even have lower heat as a result.
On the other hand, for even more heat reduction it might
help to decide on your target voltage, and THEN find the
peak CPU speed that can run stable at that voltage (be sure
to test the final config overnight running Prime95 Test as
mentioned above, before ever booting windows/other OS).
The author says he
managed to lower the CPU temperature without any loss of performance on
his computer.
Sure, if you keep the clock speeds and multiplier the same
and only lower voltage, that will retain performance- so
long as it stays stable which is the crux of the problem,
that if ALL of your model of CPU stayed stable at lower
voltage than AMD had spec'd, they might've just spec'd them
all for lower voltage instead of that they did use. Each
specimen of CPU may respond a little differently, and while
AMD tried to make it easier for users or motherboard
manufacturers by keeping same voltage for multiple speeds in
same CPU family, each may have a little bit different
voltage:frequency curve, also depending on how cool your
heatsink/case/ambient temp, keep it as well.
As lowering the temperature makes your CPU last longer,
I'm interested in doing the same,
Unless your CPU is constantly overheating (say past 65-70C
which is even then more of a stability concern than a
lifespan concern), it's lifespan should be fine without any
attempt to lower the voltage-> heat.
but I wonder if this is feasible on my
hardware (Asus A7V333 motherboard, AMD Athlon XP 2000+ running at 1.1
Ghz - voluntary BOIS setting -, the whole thing under Ubuntu GNU/Linux),
and also if it's reasonable to do so.
Yes, your board is one of the easiest ever made to undervolt
Athlons, at least if you don't mind having the case open and
manipulating jumpers. Use the voltage jumpers as I'd
mentioned in reply to Paul's post, and see this diagram.
http://69.36.189.159/usr_1034/A7V333_Jumpers1.gif
It's been awhile since I fooled around with one of those
A7V333 boards, but I vaguely recall that I did test the
pre-1.5V jumper settings, but on that board with it's
through-hole heatsink mounting system, I had an Alpha
Pal8045 heatsink which was pretty good, it cooled
sufficiently at very low noise so I had no need to keep it
undervolted, think it ended up runnning a pre-release
Thornton @ 1.65V/2.2Ghz/172MHz FSB on it but it was too
picky about memory, wouldn't take several pairs of 512MB
DIMMs I tried at that (172MHz) synchronous memory bus speed
and I reserved all my better memory for nForce2 board uses
since they could exploit memory so much more at the time.
Anyway, what voltage you can hit at 1.1GHz may depend on
which (CPU) core you have, Palomino might still need
1.3-1.4V or so but T-Bred -A, especially -B or Thorton core
might get down pretty close to 1.1V... but frankly unless
you're trying to passively cool the CPU, I don't see the
need to underclock and undervolt it that much unless it's
operating in a quite hostile (high temp) ambient
environment.
