P4Pe Core Voltage problems :-(

[Guest]

Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

Hi

I have PeP4 mobo. It has been fine for about a year.
Now I am overcklocing P4 3.06 processor. It is stable up to 155 FSB @ 1.575V
Core. I can clock it higher but it will get unstable. MBM5 Log tells that V
Core is variating a lot.
I set it to 1.6V:
On idle VCore goes up to 1.66V
Under stress it drops to around 1.56V and every now and then down to 1.52V.
(I assume it crashes when voltage drops too low)

The MoBo Core voltage requlators are not good! If only the Vcore was more
stable around 1.57V I could go up to 160FPB or even higher as cooling is OK.
(Temperature under 50)..

MBM5 log tells allso that PSU is good, all voltages within spec and no
voltage drop at the same time as VCore drop occures.

What to do? I have found several Voltage mods for newer boards, but not for
this particular board. Anoyone know a good site?

BR

Risto

 

[Paul]

Archived from groups: alt.comp.periphs.mainboard.asus (More info?)

In article <cmoo5k$bme$1@phys-news1.kolumbus.fi>, "Risto Pirhonen"
<ripir@koti.soon.fi.nospam> wrote:

> Hi
>
> I have PeP4 mobo. It has been fine for about a year.
> Now I am overcklocing P4 3.06 processor. It is stable up to 155 FSB @ 1.575V
> Core. I can clock it higher but it will get unstable. MBM5 Log tells that V
> Core is variating a lot.
> I set it to 1.6V:
> On idle VCore goes up to 1.66V
> Under stress it drops to around 1.56V and every now and then down to 1.52V.
> (I assume it crashes when voltage drops too low)
>
> The MoBo Core voltage requlators are not good! If only the Vcore was more
> stable around 1.57V I could go up to 160FPB or even higher as cooling is OK.
> (Temperature under 50)..
>
> MBM5 log tells allso that PSU is good, all voltages within spec and no
> voltage drop at the same time as VCore drop occures.
>
> What to do? I have found several Voltage mods for newer boards, but not for
> this particular board. Anoyone know a good site?
>
> BR
>
> Risto

Have you ever heard of a load line ? See the figure on page 29:

ftp://download.intel.com/design/Pentium4/datashts/29864312.pdf

The upper and lower lines on the plot are the max and min.
If the regulator circuit held the voltage steady at "VID" volts,
then the regulator would be out of specification, because its
voltage would be higher than the upper line.

In other words, droop is acceptable and is expected and
specified as such. As long as the droop is not too much.
For your processor, at 65 amps, a nominal drop of 160mV is
acceptable. At 1.52V, the droop is 140mV and still in spec.

160MHz is a typical overclock for processors of that vintage.
See http://www.cpudatabase.com/CPUdb and look for P4 3.06.
There is a P4PE running a clock of 160MHz, which results
in Fcore of 3.7GHz.

You can increase Vcore, to try to make the processor go faster.
In Table 2-5, look for "Absolute Max". In this case, it is
listed as 1.75V, and you don't want to set VID to that voltage,
but to some lesser voltage. (Absolute max is defined as the
voltage at which damage could happen to the processor, over
the long term. For example, a Tualatin had a rating of 1.75V
and would last for about four weeks at 1.8V.) But as you can
see, there probably isn't much more to be harvested there
anyway. You'll need to run Prime95 torture test (mersenne.org)
for several hours, to test for error free stability, and for
that to work, you have to back off a few megahertz or add a
bit more Vcore.

In this case, since a setting of 1.7V will give 1.76V at idle,
that is as high as you should go. If it was my processor, I
would stop at 1.65V. (The design margin at the factory is a
big unknown. Occasionally, there are papers published when a
new processor is introduced, and in one of those, I saw that
the margin was 0.2V. But each process is different, and you
really shouldn't go higher than Abs max if possible.)

I have noticed that some web sites talk about "droop mods",
but I've never examined their results in detail, to see
whether what they were doing was justified. If a circuit
has more droop than the load line spec, then by all means,
do a droop mod. If the regulator performance falls between
the two lines, then it doesn't need a mod.

HTH,
Paul