"Oldish sod" said:
I've looked at doing this for some time on an A7N8X del. 2.0, and I've
come to the conclusion that it's not possible to undervolt Vcore on this
mobo, whatever revision/version.
I've not even seen any 'wire mods' (anyone know of one?)
I'm not really complaining though as I bought it for overclocking, and
ran an XP2500 as a 3200 for a couple of years, so..
There is another tool that can change VID setting. It is called
CPUMSR, but it doesn't seem to work with Nforce2 boards. The
board will freeze if you try to change the voltage (and the
processor used, must be a XP-M mobile, or be an ordinary chip
modded to become a mobile, to work with CPUMSR). I'm not at all
certain how this is supposed to work at the hardware level, as
the description on the CPUMSR web page suggests there are "softVID"
pins as well as the ordinary VID pins on a Mobile processor.
Since I cannot download a Mobile datasheet, this is hard to
verify.
This page shows the wire mods graphically:
http://www.ocinside.de/go_e.html?/html/workshop/pinmod/amd_pinmod.html
If you select "Socket view", and set Vcore to "1.85 volts",
you'll see an example of what area of the processor socket
gets the wires added.
PDF page 66 of this document, shows the five VID socket holes on
the left hand side (see the pins on row "L"). Connecting a pin to
VCC is a logic "1" and alternately connecting to VSS is a logic "0".
You'll notice that VID4 has easy access to VSS, but not VCC, so
forcing a logic "1" on there could be a bit dangerous, if a long
bare wire is required. That would be a PITA if you need to set VID4
to logic 1. Insulating the wire could cause the processor to not
sit flat, causing poor CPU cooling.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26237.PDF
In the same document, the "desktop" voltage table is shown on
PDF pg.90.
If you are using an AthlonXP-M processor, the VID voltage table for
that can be found on fab51:
http://fab51.com/cpu/barton/athlon-e23.html#L11
The L11 table shows the correspondence between mobile and desktop
processors. My "Q" for example, would draw 1.450V if inserted in
a mobile motherboard, and instead it shows as 1.575V on my A7N8X-E.
That means VID4..VID0 values are 01011, as page 90 of the 26237.PDF
document shows 1.575V being 01011 binary value.
Looking at the table, the nearest convenient voltage value with
respect to 1.575V (01011) would be 1.475V (01111). I picked that
case because only one piece of wire would be needed to change
VID2 from logic "0" to logic "1". A wire inserted in socket hole
"L5" joined to "M4", would connect VID2 to VCC ("1"). The BIOS
Vcore setting _must_ be set to "Auto" when you do a wire mod,
as you do not want the BIOS fighting with the electrical short
caused by the wire. If you select a non-Auto Vcore value in
the BIOS, there have to be some GPIO pins somewhere that are being
used to drive the VID pins, and you have no way of knowing
exactly how Asus is doing it. A selection of "Auto" _should_
cause the BIOS to tristate the GPIO pins, so there is no
contention between wires and GPIOs.
Another thing to consider, is the electrical spec of the processor
itself. A careful experimenter will also want to check that the
wires will not damage the processor.
Table 10. VID[4:0] DC Characteristics
Parameter Description Min Max
IOL Output Current Low 6 mA ---
VOH Output High Voltage 5.25 V *
What this means is, when the processor VID pin makes a "0",
it can sink up to 6 milliamps without damage. Connecting a wire
to VCC (1.5 volts) to force a "1", pumps current into that
signal, potentially violating the spec. If the VID pin is
trying to make a "1", it is effectively an open circuit, and
a voltage of up to 5.25 volts can be applied to it. (The Vcore
regulator has pullup resistors connected to some voltage, and
that is the purpose of specifying the 5.25 volts, so if the
Vcore regulator connects its pullup resistors to +5V, there
won't be breakdown damage to the VID pins.)
What does this say about our proposed VID2 wire mod above ?
It says we could potentially damage the VID2 signal, but I suppose
people do this all the time. I would say according to the Athlon
spec, forcing VID signals to "0" is safe, while forcing them
to "1" we really cannot be sure what will happen to the processor
in the long term. (There is no way to know whether the processor
VID pin is simply a bond wire to ground inside the package, or
is an open drain transistor connected to a lasered fuse - to assess
the situation properly, you would want to do an I versus V curve
for the pin, to get some idea of what structure drives it inside
the package).
To answer this question, we can consult a venerable source:
http://web.archive.org/web/20040412055223/www.beachlink.com/candjac/slotAocg.htm
http://web.archive.org/web/20031208175846/www.beachlink.com/candjac/gfds.htm
The claim here is, that resistors are being used by the processor
for setting VID. This means you can use VCC or VSS to program
the value on a VID pin, with less to worry about. My worry was
that these pins were driven by transistors. The Athlon Model10
datasheet mentions "strapped" (which informally means resistors
in engineering-speak), and if you believe the info on PDF page
50 of the datasheet, the resistors inside the processor are
150 ohms (whereas the candjac site states they are 100 ohms).
The candjac "gfds.htm" page mentions another issue, and that is
whether a wire mod of a VID pin to VCC, gives enough voltage
to make a valid logic "1". According to my notes, the A7N8X
uses a L6917B voltage regulator (a 28 pin chip to the left of
the processor socket). The datasheet for that part says the
VID pins are "TTL compatible", which, one would hope, means
anything more than 0.8V would give a logic 1. Since the wire
mod is actually using the Vcore voltage to make the logic 1,
the lower you set Vcore, the less adequate the logic 1 becomes,
that you are sending to the VID pins on the L6917B. Since
Vcore won't be set lower than about 1.3V or so, this shouldn't
be a problem (and you'll be able to see in Asus Probe, whether
the wire mod is working anyway).
So, it looks like the wire mod should work OK
Why did I go through all this ? To show that wire modding is
not a trivial art
There are some Intel processors, where
this same technique would not be appropriate, so the ideas
don't necessarily transfer to other situations. A good mod
should be reversible and not leave traces of what you've
done
