Does CPU voltage matter regardless of core temp?

Do you have a question? Post it now! No Registration Necessary.  Now with pictures!

Threaded View
If you can keep the CPU cool, does it matter how high the voltage gets? Is the
potential damage to a CPU from OC'ing solely due to heat or are there other

Re: Does CPU voltage matter regardless of core temp?

Quoted text here. Click to load it

Every 6 degree Celcius halves the live of a chip.
Increased voltage increases disssipation, and educes livetime.
BUT... If you have a deep pocket ready to buy new chips,
you have no problem.

Re: Does CPU voltage matter regardless of core temp?

Doc wrote:
Quoted text here. Click to load it
potential damage to a CPU from OC'ing solely due to heat or are there other
Quoted text here. Click to load it

If you look at the datasheets for silicon chips, there is a
section called "Absolute Max". Once the voltage gets past
a certain point, there are breakdown mechanisms to deal with.

Processors vary, as to how the Absolute Max, meshes with
the range of VID codes. On some processors, it's all too easy
to go over a safe value. On Core2 45nm processors, about 1.4V
is a safe long term overvoltage. And that was experimentally
determined by participants on forums.

When it came to ruining certain processors, it wasn't the
absolute voltage which hurt, but a voltage difference that
mattered. On the AMD FX processors a while back, there was
a "VCore" voltage, and also a voltage used to feed the
memory interface. It turned out, again, after users experimented
with it, that if you increased VCore, you were also supposed
to increase the memory interface voltage at the same time.
It seemed, if there was a large delta between the two,
some currents would flow where they should not, inside
the processor, ruining it. Normally, a datasheet will
state in unequivocal terms, what the max difference can be,
such as

    Absolute Max

    VCore < 1.6V
    VCore - VDimm < 0.6V

and that would tell you, if you used 1.6V for the VCore, you
should use a bit more than 1.0V for the VDimm. That's in order to
keep the delta below the stated value. Chips can be quite
complicated when it comes to stating these voltage
relationships, and especially when they have as many as
five input voltages. In the case of the AMD FX processor family,
the dataaheet only addressed VCore, and did not mention
there was a delta spec to be met. And again, the experimenters
on the enthusiast forums, ruined FX processors until they
figured it out. (That dropping VDimm between experiments,
with a constant VCore, resulted in a very expensive chip
being ruined.)

So this is how it works.

1) Processor has hardware limits on the VID code that can
    be emitted. Normally, that would provide some level of
    protection against end-user abuse. But there were also
    chips (like my P4 Northwood), where it's all too easy to
    set VID to a deadly value (so-called instant death
    syndrome for that processor). Some processors "wear out"
    at the higher voltage, rather than dying instantly. A
    Tualatin processor at the wrong voltage, lasted an
    average of three weeks before dying from it. (Using
    a Slot 1 motherboard which lacked the appropriate, lower
    Tualatin core voltage to run the slocket.)
2) Motherboard manufacturer bypasses this protection and
    adds "boost" features to give you more voltage.
3) Processor datasheets gives some idea on Absolute Max.
4) Enthusiast experimenters determine what the "real"
    safe value is, ruining multiple expensive processors to
    get the value.

If you wait long enough, all the necessary information is
out there, to be located by your search engine. That's how
I could find a value for the limits on my 45nm Core2.
Someone else ruined their 45nm Core2, so I could
know what that value is.

And don't trust my values mentioned above. Do your own
data mining. You spent good money for the processor, and
the information is out there. Take responsibility for
your own experiments! I've never actually abused
my 45nm Core2 (it's never been overclocked). I did
try it on my 65nm Core2, because those are known to be
a bit more tolerant on voltage. It's not worth the
trouble on my 45nm one. I just leave that one at
nominal (so I can type this message, of course :-) ).


Site Timeline