Re: Broken Athlon XP CPU diode?

In article <1147973426.729964.215840@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>, "BL"
<bl_bl@xxxxxxxxxxx> wrote:

Paul!

I appreciate your answer but aren't you talking (mainly) about the CPU
sensor on the motherboard? What I think I have a problem with is the
CPU diode which is supposed to be built-in inside the processor. When I
use Everest Home Edition I get a reading called CPU Diode which I think
is the same as the CPU Temperature (or whatever it is called) in the
ROM Setup (Hardware monitor)).

Besides, I get the problem in the ROM Setup (BIOS Setup) and it's there
where the Hardware Monitor is available. When the processor is
installed and I boot into the ROM Setup and watch the hardware monitor,
the temperature rises without any stress to finally shut down the
computer. So if there is a possibility to modify the values with some
software (to modify some sort of ROM or similar that is) or to modify
alt. disable the CPU diode temperature sensor that could be a solution
if nothing can be done with some "super-duper" cooler kit.


TIA!

Sorry I didn't get back to you sooner. I've been off experimenting.

First of all, there are a number of A7N8X family motherboards. In
fact, it turns out they all don't have the same equipment on board
for temperature. So, we'll start with an overview.

There are two ways to sense temperature. A thermistor can be placed
in the motherboard socket, underneath the processor. This sensor
responds slowly, and the temperature seen will be less than the
real temperature. The thermistor is connected to the main monitoring
chip. In this case, it could be a ITE8705 Super I/O, for which a
data*** is not available (the 8708 is similar, and what I use
as a reference for the 8705). The thermistor is available, to provide
a temperature readout for the user, but its main purpose is not to
safeguard the processor. The monitor in a Super I/O chip can typically
measure 7 voltages, 3 temperatures, and 3 fan speeds.

AthlonXP processors did not have overheat protection designed into
the chip. The Intel Pentium processors have a feature like that.
(called THERMTRIP). This missing feature did not become obvious,
until the Tomshardware video was made, of an Athlon burning up
when the heatsink went missing.

To fix this problem, the designers looked at what resources were
available. The thermistor in the socket certainly measures temperature,
but the processor can be well and truly fried, before the socket
temperature gets high enough to be an issue. There are some
motherboards that do use the thermistor for overheat detection -
they can detect if the fan has failed (as the temp rises slowly
then), but if the heatsink falls off, the thermistor based protection
schemes aren't fast enough to guarantee the processor will be
saved.

The designers also realized that there is a thermal diode available
on the processor. This is not documented on all models of Athlons,
but is available in the later models. The diode is part of the silicon
die, so it gets temperature information instantly. (The only time
it gets fooled, is when a user puts the heatsink on rotated 180 degrees
from the correct position. Only half the silicon die gets cooled,
and if the thermal diode is on the "cool" side of the silicon die,
it may miss the overheat happening on the other half of the chip.)

To measure the diode temperature, the designers needed a chip that
does not rely on the BIOS or the processor or software, as part of
the detection scheme. They found a couple of chips that work as
shutdown devices, and the chips don't need any help from the processor.

The Winbond W83L785TS-S chip doesn't need any help to do its job.
It has some resistor straps, that are used by the motherboard
designer, to tell the chip what to do. In addition, the clever
Winbond guys put an I2C interface on the W83L785TS-S, and that
can be connected to the SMBUS (serial management bus). That is the
same bus that is used to read the memory timings out of the SPD chip
on the DIMM(s). The Winbond chip apparently is used on A7N8X and
A7N8X Deluxe, revisions <2.0 and revision 2.0. The assignment
of temperature measurements differs between the boards, as
seen in this Nforcershq posting.

http://www.nforcershq.com/forum/viewtopic.php?t=16364

For the A7N8X-E Deluxe (the board I had the misfortune to
experiment with), Asus decided to use a different chip. It
took me the longest while to tweak onto this, as the computer
case on my A7N8X-E is a pain to open up, and I didn't start by
visually inspecting the thing first.

There is an Attansic chip in there, an ATTP3. Attansic is a fabless
semiconductor company, of which Asus is an investor. Attansic
modified a ATTP1 and made a custom version with the ATTP3 part
number. There is no data*** (typical Asus style). The ATTP3
does the same job as the Winbond chip, except it has no SMBUS
interface. That means, on the A7N8X-E Deluxe, there is no option
to turn off the overheat detection, and no way for MBM5 to
display a diode temperature (only socket is available).

Now, back to your problem.

"Besides, I get the problem in the ROM Setup".

Well, bingo. What you are seeing is the "socket" temperature
rising. The BIOS does not display the diode temperature. Only
MBM5, with a lot of coaxing with the right .ini file, will
display the diode temperature as measured by the W83L785TS-S.

If the socket temperature is rising, that suggests the chip is
really getting hot. Maybe your heatsink is not making contact
with the die. Some people use rubber bumpers or a shim, to
prevent the silicon die from getting chipped, and if you do
that, more compression force is required to ensure the heatsink
is actually pressing against the die. You can test for this, by
placing a half-rice-grain sized dot of AS5 on the die, plop the
heatsink on it, fasten the clips, disassemble, and observe the
spreading pattern of the thermal compound. If the compound does
not get spread out, then you know the problem is a mechanical
mounting issue.

If the W83L785TS-S diode temperature by itself was rising (as
can only be observed in MBM5), you might have a point to trying
to disable the W83L785TS-S. On the other hand, if the thermistor
temperature is rising (measured by ITE8705, observable in the BIOS
or with MBM5), it would be a bit stupid to turn off the
W83L785TS-S, as it is measuring something entirely imdependent.
If the chip temperature really is rising, there is no sense
turning off the only defenses that the chip has.

Now, in my travels, I did read something interesting. In fact,
MBM5 does have the ability to write to SMBUS connected devices.
You can put write commands in one of the .ini files. (I found
this on the livwiredev.com site, but right now, for some reason,
DNS translation is not working for that domain. Almost as if
it is dyn-dns or something.) If I remember, the next time I
can reach the site, I'll post a URL.

HTH<
Paul



Paul wrote:
In article <1147951452.103026.317860@xxxxxxxxxxxxxxxxxxxxxxxxxxxx>, "BL"
<bl_bl@xxxxxxxxxxx> wrote:

Hi!

I've had some problems with my 3200+ 400FSB processor right from the
beginning. I bought the processor last summer, had problems with the
new cooler right at installation and tried my old cooler which I used
for my 2600+. I managed to get it working by having the box constantly
open and getting the new cooler to be attached "perfectly" (after some
5-6 attempts). The reading I got from the CPU diode when looking in ROM
was some 80C+ but it worked for normal computer use (mostly) except
hardcore gaming when the box was opened :-(

Anyway, I thought I'd deal with this problem now and get it properly
installed. I bought some new fans for the box and installed these and
closed the box. It worked for some 10+ minutes before the lovely A7N8X
Deluxe just shut off the system. After that I've tried multiple tests,
taking the CPU off the motherboard, cleaning it from old cooler paste
and adding new again (doing the same to the heatsink), re-oriented the
cooler heatsink, trying my old heatsink etc. etc. Now I cannot come to
a stable setup with my box opened any longer (it's even worse)! When it
runs at 3200+ speed (200MHz) it takes some 1-2 minutes sitting and
watching the hardware monitor in ROM before it shuts down. All other
temperatures except the CPU Diode is correct (MB CPU is about 45-55).
As said I'm using an ASUS A7N8X Deluxe (2.0) with latest official ROM
(1008).

What can I do?

Is there some super-duper state of the art cooler that will guarantee
that I can run it properly (and I mean guaranteed)? Which can lower the
temperature some 15-20 degrees C.

Is it possible to disable the CPU Diode reading and dependency? I mean
the heatsink is not hot at all so I think the reading is incorrect and
I'm prepared to take the risk if necessary.

Is there any guide with pictures of how to correctly apply cooler paste
to an Athlon XP / heatsink for optimum performance?

Is arctic silver (or other products possibly) that much better that
standard cooler paste (used for transistors etc.)?

Right now I'm using my old 2600+ and it's ok but the 3200+ was
definately faster.

TIA!

If the diode is not working properly, or the monitor chip is
not working properly, adding cooling may not help.

The device used to sense the processor temperature is likely this
one:


http://www.winbond-usa.com/products/winbond_products/pdfs/PCIC/W86L785TS-S.pdf

One way to modify the behavior of the part, would be
writing to the control register via the SMBUS. Programs like
Speedfan or MBM5, may know how to probe for the device, but
I doubt those programs will allow you to do some arbitrary
writes to the device. (Linux code, like lmsensors or mbmon,
may show you a few tricks for accessing the SMBUS, but don't
ask me how you would write equivalent Windows code.)

There are some things you could do with a soldering iron, to
stop it, but that would void your motherboard warranty.

Paul
.