Overclocking: Leave voltage as is?

[aether]

I'm running a 3500+ (90nm) processor on an AN8 ('Fatal1ty')
motherboard, and I've overclocked it from 2.2 to 2.5 GHz. However, the
voltage remains at 1.4 despite having multiple options in BIOS. Should
I up the voltage? If so, how much? Could I be damaging the processor by
leaving the voltage as is? I've not encountered any problems, but it
does seem as though it would require more juice.

 

[Conor]

I'm running a 3500+ (90nm) processor on an AN8 ('Fatal1ty')
motherboard, and I've overclocked it from 2.2 to 2.5 GHz. However, the
voltage remains at 1.4 despite having multiple options in BIOS. Should
I up the voltage? If so, how much? Could I be damaging the processor by
leaving the voltage as is? I've not encountered any problems, but it
does seem as though it would require more juice.

If it works, leave it alone. Upping the voltage increases heat and
damages the processor.

 

[Pete D]

Sounds like there is nothing to fix here, move along!!

 

[aether]

Will leave as is. At one point, I did increase it to 1.5. Was that
enough to damage the processor? (the max is 1.8)

 

[Ben Pope]

Will leave as is. At one point, I did increase it to 1.5. Was that
enough to damage the processor? (the max is 1.8)

Something will shut it down before it is damaged.

It probably won't be damaged until past 100°C.

Voltage doesn't damage CPUs, heat does.

If there is not enough voltage, you will get transient errors (but no
damage).

Ben

 

[Paul]

Will leave as is. At one point, I did increase it to 1.5. Was that
enough to damage the processor? (the max is 1.8)

You can judge that from the processor datasheet, not from what
the BIOS is saying. For example, there are tech docs here:

http://www.amd.com/us-en/Processors/TechnicalResources/0,,30_182_739_7203,00.html

Absolute max Vcore is listed on PDF page 51. It is 1.65 volts.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/31411.pdf

If it was my processor, I wouldn't use more than 1.65V. In addition,
due to overvolting by some Vcore circuits, a safer limit would be
1.6V. The higher voltages are fine for veteran overclockers on
unlimited budgets

Next, look at the operating data:

http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30430.pdf

On page 13, I see ADA3500DIK4BI. Maybe that is the 90nm processor,
I'm not really sure. Look at the operating points. An extra 0.050V
buys 200MHz of extra core. That is AMD's estimate. If you set the
voltage to 1.5V, instead of 1.4V, that means you should be able to
go from 2200 to 2600MHz.

To test stability, use something like Prime95 (mersenne.org). First
try your overclock at 1.5V . See if the board runs error free for
at least several hours. Now, drop the voltage 0.050V at a time
(or whatever the smallest step the BIOS allows for voltage change).
Take small steps, so you don't corrupt your operating system.
(I use a Knoppix read-only linux boot disk for this kind of testing.
There is a linux version of Prime95 available from mersenne.org .)
At the point that it starts to error in Prime95, you have reached a
limit. Now, you want to apply a bit more voltage again. Run Prime95
again. You should be able to come up with a set of conditions which
is optimal for your particular processor, and stable for hours on
end.

Electromigration is one failure mechanism, and it is my belief
that the cases you read about, where a processor can no longer
run at its rated speed, is an example of electromigration damage.
Electromigration is related to the amount of current flowing in
the wires that route the logic signals on the silicon die. The
wires are made wide enough, that the chip can operate at frequencies
higher than the nominal operating frequency. You should be able
to operate your processor, at least as fast as the faster speed
bin version of your processor die. There is no way to estimate
how much further you can safely go, as each transition (from 130nm
to 90nm to 65nm and so on) will bring with it, different
electromigration rules. It doesn't sound like your current overclock
is too extreme.

HTH,
Paul

 

[David Schwartz]

Will leave as is. At one point, I did increase it to 1.5. Was that
enough to damage the processor? (the max is 1.8)

So long as you didn't overheat it, the higher voltage won't damage it
(since it's still less than the absolute maximum permitted). You should use
the lowest voltage at which the processor is reliable. Keep a close eye on
the CPU temperature, especially as summer rolls around and as the fan ages
and dust accumulates on the heat sink. If the CPU becomes unreliable or the
temperature gets too high, re-evaluate everything.

DS

 

[aether]

So long as you didn't overheat it, the higher voltage won't damage it
(since it's still less than the absolute maximum permitted). You should use
the lowest voltage at which the processor is reliable. Keep a close eye on
the CPU temperature, especially as summer rolls around and as the fan ages
and dust accumulates on the heat sink. If the CPU becomes unreliable or the
temperature gets too high, re-evaluate everything.

DS

I'm sure it didn't overheat, as I was monitoring it and the highest I
saw it reach was 53c. At that point, I lowered the voltage back to 1.4.
The members of this newsgroup have been a big help to me. I appreciate
it.

 

[Trent]

Voltage doesn't damage CPUs, heat does.

You are an utter ****ing imbecile.

 

[NuTCrAcKeR]

Please visit alt.comp.hardware.overclocking.amd for better information than
is being presented here.

- NuTs

 

[Mercury]

Thank you for the clarity and pointedness of your communications. However,
although I can guess at the intended purpose of your statement I prefer not
to. In future, perhaps you may wish to put more effort into not just what
you are trying to communicate, but also how.

 

[- HAL9000]

Trent, what do you mean? Ben was correct in a common sense or
practical way. Were you thinking of super cooling the cpu and then
applying extreme voltages to break down it's barriers, or ?

Forrest

Motherboard Help By HAL web site:
http://home.comcast.net/~mobo.help/

 

[Stephen]

Voltage doesn't damage CPUs, heat does.

Voltage sure can damage a CPU, just hit it with a nice static
electrical charge and see if it still works.

Stephen


--

 

[Ben Pope]

Voltage sure can damage a CPU, just hit it with a nice static
electrical charge and see if it still works.

And what exactly do you think a few thousand volts discharge through the
chip *actually* does?

The voltage potential causes a current to flow. That current flows
through "thin wires". The "thin wires" have a high resistance. That
high resistance causes the "thin wires" to heat. The heat damages the
"thin wires".

So my point was that heat damages CPUs, not voltage. The voltage causes
a current, which causes the heat, but the voltage itself didn't cause
the damage, the heat did. If the source is current limited, you could
apply a high voltage and not damage the chip.

Besides, I was assuming that the CPU was in the motherboard when the
user was adjusting the voltage from 1.4V to 1.5V, not running aimlessly
around the coffee as fast as he can in his carpeted lounge wearing
rubber shoes, holding the CPU in his hands, shouting "DIE, DIE".

He might have been, but it seems unlikely.

Ben

 

[aberger]

And what exactly do you think a few thousand volts discharge through the
chip *actually* does?

The voltage potential causes a current to flow. That current flows
through "thin wires". The "thin wires" have a high resistance. That

high resistance causes the "thin wires" to heat. The heat damages the
"thin wires".

So my point was that heat damages CPUs, not voltage. The voltage causes
a current, which causes the heat, but the voltage itself didn't cause

the damage, the heat did. If the source is current limited, you could
apply a high voltage and not damage the chip.

Besides, I was assuming that the CPU was in the motherboard when the
user was adjusting the voltage from 1.4V to 1.5V, not running aimlessly
around the coffee as fast as he can in his carpeted lounge wearing
rubber shoes, holding the CPU in his hands, shouting "DIE, DIE".

He might have been, but it seems unlikely.

Ben

Ben,
While you've given many people good advice, I have to disagree with
you on this one. You can certainly damage an integrated circuit through
overvoltage. I'm not talking about a 2 KV zap from a carpet, I'm
talking about running a chip over its maximum voltage rating. You are
correct in saying that additional heat will be generated by upping the
voltage, but you also risk parts failure through exceeding the maximum
voltage.

Arnie Berger

 

[Ben Pope]

Ben,
While you've given many people good advice, I have to disagree with
you on this one. You can certainly damage an integrated circuit through
overvoltage. I'm not talking about a 2 KV zap from a carpet, I'm
talking about running a chip over its maximum voltage rating. You are
correct in saying that additional heat will be generated by upping the
voltage, but you also risk parts failure through exceeding the maximum
voltage.

What is the failure mode in that case?

Ben

 

[David Schwartz]

What is the failure mode in that case?

Overvoltages can cause damage by causing excessive temperatures;
however, excessive voltages can also cause damage directly. See, for
example:

http://www.findarticles.com/p/articles/mi_m0HPJ/is_n5_v45/ai_16182389

"The physical effects of ESD and EOS [electrical overstress] on ICs can be
categorized as thermally induced or electric field induced failures. Among
the thermally induced failure mechanisms are drain junction damage with
melted filaments, polysilicon gate filaments, contact metal burnout, and
fused metallization. Typical field induced ESD-related failure mechanisms
are dielectric breakdown (gate oxide rupture) and latent hotcarrier damage."

Dielectric breakdown is caused by the excessive voltage itself, not any
heat created by the greater voltage. The electrons almost literally punch
holes.

http://www.semiconfareast.com/oxidebreakdown.htm

DS

 

[aberger]

What is the failure mode in that case?

Ben

Several of the other posts answered it. Usually it is either
punch-through, breaking down the dielectric, or electromigration. I
suppose that there can also be some local heating effects (ie, the
thermal conductivity of the substrate is not sufficient for the amount
of heat generated, but I wasn't referring to heat.

I'm not an expert on IC breakdown mechanisms but I know that you can
get a bipolar part to go into latch-up by exceeding its breakdown
voltage. In effect, it becomes a zener diode, and without a current
limiting resisitor, it self-destructs.

arnie

 

[H.W. Stockman]

On Sun, 27 Mar 2005 23:55:41 +0100, Ben Pope
<benpope81@_REMOVE_gmail.com> had a flock of green cheek conures
squawk out:

[...]
The voltage potential causes a current to flow. That current flows
through "thin wires". The "thin wires" have a high resistance. That
high resistance causes the "thin wires" to heat. The heat damages the
"thin wires".

So my point was that heat damages CPUs, not voltage. The voltage causes
a current, which causes the heat, but the voltage itself didn't cause
the damage, the heat did. If the source is current limited, you could
apply a high voltage and not damage the chip.

Metal melts do happen, but junction breakdown and oxide failure are possibly
more important with ESD. Might a really high voltage cause migration across
the n and p layers, making the semiconductor diodes non-functional, without
melting?

There may be some difference between the operative mechanism for damage from
the 1000 to 40,000 volts one may produce walking across the carpet
(dissipated in a millisecond), and damage from supplying an extra 0.5 volts
(over months). Long-term damage occurs at well below the melting points of
many of the metals in semiconductors (though perhaps not below the
temperatures for fast mixing of oxide-metal "alloys"). Latent and/or
long-term failures are probably by very different mechanisms than those that
are caused by ESD.

 

[Paul]

On Sun, 27 Mar 2005 23:55:41 +0100, Ben Pope
<benpope81@_REMOVE_gmail.com> had a flock of green cheek conures
squawk out:

[...]
The voltage potential causes a current to flow. That current flows
through "thin wires". The "thin wires" have a high resistance. That
high resistance causes the "thin wires" to heat. The heat damages the
"thin wires".

So my point was that heat damages CPUs, not voltage. The voltage causes
a current, which causes the heat, but the voltage itself didn't cause
the damage, the heat did. If the source is current limited, you could
apply a high voltage and not damage the chip.

Metal melts do happen, but junction breakdown and oxide failure are possibly
more important with ESD. Might a really high voltage cause migration across
the n and p layers, making the semiconductor diodes non-functional, without
melting?

There may be some difference between the operative mechanism for damage from
the 1000 to 40,000 volts one may produce walking across the carpet
(dissipated in a millisecond), and damage from supplying an extra 0.5 volts
(over months). Long-term damage occurs at well below the melting points of
many of the metals in semiconductors (though perhaps not below the
temperatures for fast mixing of oxide-metal "alloys"). Latent and/or
long-term failures are probably by very different mechanisms than those that
are caused by ESD.

And the mechanism in electromigration, is spelled out in its name.
There is actual material transport, so as time passes, the composition
of the "wires" in the IC changes. If a wire becomes thinner, the
propagation delay of a signal on the wire becomes longer. That
represents a timing change, and might be compensated for by turning
down the clock frequency in the affected part of the circuit.

In the semiconductor fab, great care is taken in choosing the
dimensions and composition of the "wires". If the circuit operates
at 3GHz, someone will try to make sure the wires are good to 3.5 or
4GHz. Since the wire dimensions affect the outside dimensions of
the silicon die, and the manufacturing economics, there are
incentives not to overdo it. From an overclockers perspective,
it means if you put some effort into your overclock, you could
enter the realm of accelerated life testing.

My info is years old, and was from a conversation with someone
at our fab (the head of the cell library development department).
Since materials and methods have changed a lot since
then, the design rules and margins used could be very different.
My info is only intended to illustrate a failure mechanism
which is different from the breakdown phenomenon the other
posters have mentioned.

There is a fine article here, full of wizzy words:
http://en.wikipedia.org/wiki/Electromigration

With regard to breakdown, there was one ISSCC paper presented by
Motorola, where the breakdown voltage was only listed as 0.2V
greater than absolute max for Vcore. Again, many overclockers will
choose to ignore Vcore_max, and it is anyone's guess as to when
they will be hit. Since there are devices like the OCZ DIMM
booster, and the DFI board that can run RAM at 4 volts, I guess
all this knowledge about absolute max is irrelevant ))

Presented by someone with only a casual interest in the subject,
Paul