thermal paste and overheating, cause and effect?

[micky]

Is it true that: Not only can uneven (missing on both surfaces in
some areas), dried-out white thermal paste after the heat sink is
removed be a symptom of a CPU that overheated, it can also be the
cause of the overheating in the first place instead. ??

If that's true, and if one wasn't watching the cpu temp at the time of
the failure, is there a way to tell which it was?

Not a momentous question, but I'm curious.

Thanks.

 

[Paul]

Is it true that: Not only can uneven (missing on both surfaces in
some areas), dried-out white thermal paste after the heat sink is
removed be a symptom of a CPU that overheated, it can also be the
cause of the overheating in the first place instead. ??

If that's true, and if one wasn't watching the cpu temp at the time of
the failure, is there a way to tell which it was?

Not a momentous question, but I'm curious.

Thanks.

Thermal paste is subject to "pump-out". I'd assume you had a
uniform layer when it was installed, and some of it dried out
or pumped out.

Paul

 

[VanguardLH]

Is it true that: Not only can uneven (missing on both surfaces in
some areas), dried-out white thermal paste after the heat sink is
removed be a symptom of a CPU that overheated, it can also be the
cause of the overheating in the first place instead. ??

If that's true, and if one wasn't watching the cpu temp at the time of
the failure, is there a way to tell which it was?

Not a momentous question, but I'm curious.

Thanks.

Thermal paste has no where near the thermal conductivity of metal. You
want the best metal-to-metal contact you can get. To that end, some
enthusiasts will lap the CPU and especially the heat sink so they mate
as well as the end user can make them (you use a piece of plate glass as
the flat surface on which to grind). However, even that won't give you
perfect contact. The thermal paste is intended to fill in the
microscropic non-contact areas. It is NOT meant to substitute for
metal.

If you glop on thermal paste, you reduce the thermal transfer that
would've been possible with a thin layer. You want to fill in the tiny
gaps between CPU and heat sink, not sandwich in a layer of paste. You
want to start out with an even *thin* layer of paste and then let the
pressure squeeze out the remainder but then you don't want gobs pushed
out around the CPU, especially if the paste is conductive. You want as
much metal-to-metal contact as you can get. The only reason you use
thermal paste is to fill in the tiny (unseen to your eye) gaps between
surfaces because, well, at least thermal paste has more thermal
conductivity than air.

If the thermal paste dried out, it was crap quality. Don't rely on heat
sink makers to include a decent quality paste, if they include it at
all. The little blister of paste that comes with the heat sink is way
more than you need for a single application. Apply and then nearly
scrape off the excess. While the pressure of the heat sink against the
CPU plate will squeeze out the excess paste, don't expect the paste to
travel very far. No, you don't have to go hyper and pay royal for the
best paste since the difference is often minimal, but there are quite a
few good quality pastes available.

That users think they know how to apply thermal paste doesn't mean they
do know the proper procedure or even what the paste is supposed to
accomplish. Only enthusiasts (nuts) go as far as lapping the heat sink
and CPU plate to ensure best mating surfaces (but still have to be
careful not to overly thin the CPU plate). I have yet to see any
pre-built makers lap the two or computer shops do it. It takes time and
skill for perhaps all of a 1-2 C temperature change, but users that
think the paste is the thermal transfer mechanism end up slapping way
too much of it on. Most users don't even bother cleaning the two mating
surfaces using proper low-residue solvents (heptane, mineral spirits,
99% isopropyl alcohol). Fingerprints can be up to 0.005" thick which
prevents the micronized particles in the paste from directly contacting
the metal surfaces.

You want a *little* bit of paste to replace the air in the microscopic
gaps between metal surfaces. You still want as much metal-to-metal
contact as you can get. Sandwiching in a layer of paste means you get
the far reduce thermal conductivity between the surfaces than you would
get with the metal surfaces. Air is bad, paste is better, metal is
best. Also be sure you are using a thermal paste and not a thermal
adhesive as the later could mean you'll never be able to remove the heat
sink.

 

[larry moe 'n curly]

Is it true that: Not only can uneven (missing on both surfaces in
some areas), dried-out white thermal paste after the heat sink is
removed be a symptom of a CPU that overheated, it can also be the
cause of the overheating in the first place instead. ??

I don't think it means the CPU overheated but that the paste wasn't
applied thoroughly in the first place. However having a gap like that
is probably even worse than running without any paste at all because
paste is only supposed to fill small air gaps and conducts heat about
50x as well as air does. That doesn't mean its useless, because one
website about electronics cooling said it doubled the overall heat
conduction between the chip and heatsink.

If that's true, and if one wasn't watching the cpu temp at the time of
the failure, is there a way to tell which it was?

Measure the temperature difference between the CPU package and
heatsink?

 

[micky]

I don't think it means the CPU overheated but that the paste wasn't
applied thoroughly in the first place.

I've read that the paste should be replaced periodically. Some people
even say once a year. For this computer, I think it was 7 years or
more. So I would think the paste can be bad even if was good at the
start, or why replace it at all?

(Until this July, I didn't know anything about periodically replacing
the paste.)

And on a personal note, I was very careful when I put the paste on and
I'm sure I was throrough. There is absolutely no way that I left 10
or 20% uncovered, or even 1%. And it was the white paste that came
with the heat sink, so I coudl see what I'd done.

I just got the AZ5 in the mail yesterday, and the replacement cpu. I
hope to put it in today if there is time.

However having a gap like that
is probably even worse than running without any paste at all because
paste is only supposed to fill small air gaps and conducts heat about
50x as well as air does. That doesn't mean its useless, because one
website about electronics cooling said it doubled the overall heat
conduction between the chip and heatsink.

Uh huh. (This means nodding in agreement. Once I said this and
someone thought it was meant sarcastically!)

Measure the temperature difference between the CPU package and
heatsink?

I only have one temperature value, the CPU I guess. I did buy a
cheap Harbor Freight, maybe 10 dollars on sale for 5, remote temp
sensor, but I don't think it's "lens" is good enough to distinguiish
the heat sink from anything nearby, and I wouldn't even guaratee the
thing is accurate. Still, if I can find the thing and its battery
isnt' dead I guess I'll try it.


Thanks, and thanks Paul and Van.

 

[Paul]

I just got the AZ5 in the mail yesterday, and the replacement cpu. I
hope to put it in today if there is time.

ArcticSilver provides instructions for the various processor types.

There are two ways to install paste. Spread it with a credit card,
or put a rice sized grain in the center of the target, and
"squish" it with the heatsink. It's hard to say which avoids
air gaps or bubbles the best.

You can do the "squish" as a test. Put a half grain sized dot of
paste, apply the heatsink and fasten the lever. Then, disassemble
and inspect the result. Now, noting the size of the grain
deposited, you can scale up the grain the next time, to a
size estimated to be enough to cover the CPU, plus a bit more.
If properly applied, the joint between surfaces should show a bit of
paste, as proof it flowed out to the edge. If you were
to use too much, it would ooze out. So that's a way of
calibrating how much paste is needed. Use too little to
start, and see how much it spreads.

Your motherboard should at least have a CPU temperature
sensor. On an Athlon board, the sensor can be a thermistor
just below the CPU socket. Or, on newer boards, the
hardware monitor is wired to the thermal diode inside
the CPU, and you get a "die temperature" readout. To
make the socket type reading agree with a die temperature
reading, usually there is a "fudge factor" applied to
the thermistor type readout. Obviously, the thermistor type
isn't very accurate.

None of this matters though, because in the future, you'll
be looking for "relative" shifts in temperature, not an
absolute temperature. So lets say:

1) Install CPU and AS5 today.
2) Wait two days for the AS5 to "bed in" and the lowest
temperature possible to be evident.
3) Now, measure room temperature (with an ordinary thermometer)
and CPU temperature. Ideally, you'd want a case temperature,
but use whatever you've got.
4) Work out a delta (CPU - room_temp) or (CPU - case temp).
5) A year from now, repeat the test. It helps if the
room temperature, the CPU loading, and other conditions
are constant. You wouldn't want to run Prime95 for one test,
and not for the other. Taking a delta, is supposed to subtract
the contribution from an overly hot room. The cooler
performance is actually a measure of (CPU - case temp),
but you may not have a case temp sensor on the motherboard.
Some of my motherboards have a case temp sensor, and it
would be preferred.

If, after all of this, the CPU temp rise is 10C hotter
than at the initial installation, that's telling you
something has changed. Either the heatsink is plugged
with dirt, the fan is slower than it used to be, or...
the paste is bad.

That method doesn't guarantee you'll detect all possible
defects in the paste. I wouldn't expect AS5 to fail in
exactly the same way as your previous product. Any product
can pump out or dry out, but hopefully without leaving
a portion of the die completely devoid of material.

Using the temperature method, is to avoid having to
take it apart too much.

Paul

 

[larry moe 'n curly]

On Wed, 13 Jul 2011 15:47:03 -0700 (PDT), "larry moe 'n curly"

Is it true that: Not only can uneven (missing on both surfaces in
some areas), dried-out white thermal paste after the heat sink is
removed be a symptom of a CPU that overheated, it can also be the
cause of the overheating in the first place instead. ??


I've read that the paste should be replaced periodically. Some people
even say once a year. For this computer, I think it was 7 years or
more. So I would think the paste can be bad even if was good at the
start, or why replace it at all?

I have a feeling the people who recommend annual changes either sell
thermal paste or unneeded computer maintenance or are obsessed about
cooling just for the sake of cooling -- notice that ARStechnica.com
has a forum called "Case & Cooling Fetish". Or maybe I should have
changed the thermal paste of my old TV 35 times by now (still works
great). I once checked a couple of ancient Pentium II mobos with
fanless CPU heatsinks. Their white opaque thermal paste hadn't dried
out but was much more viscous than anything I had seen (could have
been that way since new). I measured the temperature difference but
don't remember if it improved afte rI cleaned off the old thermal
paste and applied new Radio Shack stuff. OTOH after 4-5 years the
thermal paste in my generic drip coffee maker dried out enough to
become crusty, but the stuff I had scraped off when it was new was
still creamy (I disassembled it when new because it wasn't UL or CSA
approved for safety). Also I found chaulky thermal paste in an old GM
car, where it was used for waterproofing the lightbulb sockets of the
tail lights.

If you measure the temperature difference between the CPU and
heatsink, be careful to place the heatsink temperature probe at the
same location each time, preferrably shielding it from fan air,
especially if the fan speed varies. Also apply a spot of thermal
paste to the probe, but not anything containing silver or aluminum.

Here's what Dow Corning says about its thermal pastes. Unlike Arctic
Silver, they actually manufacture their products, have tested them
thoroughly, and know everything about them. Nor do they act like
weasels and give out contradictory information ("AS5 has become a
standard in the wind power industry for use in IGBTs which may run
fairly high voltage", but we don't know if it's safe to such voltage
because we've never run any formal tests and have never used it with
anything above 12V):

http://www.dowcorning.com/content/publishedlit/11-1712-01.pdf

If a CPU runs cool _enough_ under worst-case conditions, there's no
reason to make it run cooler, and I'm including keeping it below
temperatures that cause old wet-style electrolytic capacitors to age
prematurely (probably mostly from heat conducted through the circuit
board). OTOH it's probably a lot better to handle any capacitor heat
problem with higher quality capacitors.

And on a personal note, I was very careful when I put the paste on and
I'm sure I was throrough. There is absolutely no way that I left 10
or 20% uncovered, or even 1%. And it was the white paste that came
with the heat sink, so I coudl see what I'd done.

It could be that paste pump-out problem Paul mentioned. AMD has
discussed it (why they switched from thermal paste to phase-change
material for their heatsinks), and Dow Corning mentions how their
current pastes don't have the problem (implying their old pastes
did???).

I just got the AZ5 in the mail yesterday, and the replacement cpu. I
hope to put it in today if there is time.

I only have one temperature value, the CPU I guess. I did buy a
cheap Harbor Freight, maybe 10 dollars on sale for 5, remote temp
sensor, but I don't think it's "lens" is good enough to distinguiish
the heat sink from anything nearby, and I wouldn't even guarantee the
thing is accurate. Still, if I can find the thing and its battery
isnt' dead I guess I'll try it.

Most of those infrared thermometers have a spread of 8:1, meaning if
you hold them 8" away, they report the average temperature of a 1"
diameter area. Also only expensive ones are designed to accurately
read surfaces that aren't nearly black, but fortunately about any
surface that's not yellow, white, or silvery is virtually the same as
black.

 

[micky]

I meant AS5.

ArcticSilver provides instructions for the various processor types.

Yes, you posted for me the pdf url. Thanks.

There are two ways to install paste. Spread it with a credit card,
or put a rice sized grain in the center of the target, and
"squish" it with the heatsink. It's hard to say which avoids
air gaps or bubbles the best.

You can do the "squish" as a test. Put a half grain sized dot of
paste, apply the heatsink and fasten the lever. Then, disassemble
and inspect the result. Now, noting the size of the grain
deposited, you can scale up the grain the next time, to a
size estimated to be enough to cover the CPU, plus a bit more.

I watched a youtube video about this and he started with a whole rice
grain size, but his surface seemed a lot bigger than mine. Mine is
about 1/2" square, 1/4 sqare inch. Is that a just a small fraction of
what they are now?

If properly applied, the joint between surfaces should show a bit of
paste, as proof it flowed out to the edge. If you were
to use too much, it would ooze out. So that's a way of
calibrating how much paste is needed. Use too little to
start, and see how much it spreads.

Okay.

The rest will be more imprortant when attaching it is done.

Your motherboard should at least have a CPU temperature
sensor. On an Athlon board, the sensor can be a thermistor
just below the CPU socket. Or, on newer boards, the
hardware monitor is wired to the thermal diode inside
the CPU, and you get a "die temperature" readout. To
make the socket type reading agree with a die temperature
reading, usually there is a "fudge factor" applied to
the thermistor type readout. Obviously, the thermistor type
isn't very accurate.

None of this matters though, because in the future, you'll
be looking for "relative" shifts in temperature, not an
absolute temperature. So lets say:

1) Install CPU and AS5 today.
2) Wait two days for the AS5 to "bed in" and the lowest
temperature possible to be evident.
3) Now, measure room temperature (with an ordinary thermometer)
and CPU temperature. Ideally, you'd want a case temperature,
but use whatever you've got.
4) Work out a delta (CPU - room_temp) or (CPU - case temp).
5) A year from now, repeat the test. It helps if the
room temperature, the CPU loading, and other conditions
are constant. You wouldn't want to run Prime95 for one test,
and not for the other. Taking a delta, is supposed to subtract
the contribution from an overly hot room. The cooler
performance is actually a measure of (CPU - case temp),
but you may not have a case temp sensor on the motherboard.
Some of my motherboards have a case temp sensor, and it
would be preferred.

If, after all of this, the CPU temp rise is 10C hotter
than at the initial installation, that's telling you
something has changed. Either the heatsink is plugged
with dirt, the fan is slower than it used to be, or...
the paste is bad.

That method doesn't guarantee you'll detect all possible
defects in the paste. I wouldn't expect AS5 to fail in
exactly the same way as your previous product. Any product
can pump out or dry out, but hopefully without leaving
a portion of the die completely devoid of material.

Using the temperature method, is to avoid having to
take it apart too much.

That's good. I don't like taking it apart. Especially since there is
so little space left on my desk.

 

[micky]

ArcticSilver provides instructions for the various processor types.

The instructions also say one can clean the cpu and the heat sink with
xylene (such as in Goof-Off) and high-purity isopropyl alcohol. All I
have is the isopropyl alcohol they sell at the drug store,

If I don't use that, beause drug store alochol is worse than nothing,
I assume I shouldn't use the xylene?

Is drug store alcohol worse than using nothing?

 

[micky]

I have a feeling the people who recommend annual changes either sell
thermal paste or unneeded computer maintenance or are obsessed about
cooling just for the sake of cooling -- notice that ARStechnica.com

I've thought about that, but ultimately I'm gullible and I assume they
are correct, even if I don't do it even every two years.

has a forum called "Case & Cooling Fetish". Or maybe I should have
changed the thermal paste of my old TV 35 times by now (still works
great). I once checked a couple of ancient Pentium II mobos with
fanless CPU heatsinks. Their white opaque thermal paste hadn't dried
out but was much more viscous than anything I had seen (could have
been that way since new). I measured the temperature difference but
don't remember if it improved afte rI cleaned off the old thermal
paste and applied new Radio Shack stuff. OTOH after 4-5 years the
thermal paste in my generic drip coffee maker dried out enough to
become crusty, but the stuff I had scraped off when it was new was
still creamy (I disassembled it when new because it wasn't UL or CSA
approved for safety). Also I found chaulky thermal paste in an old GM
car, where it was used for waterproofing the lightbulb sockets of the
tail lights.

The tail lights etc. don't use thermal grease, do they? They use
dielectric grease, I think.

If you measure the temperature difference between the CPU and
heatsink, be careful to place the heatsink temperature probe at the
same location each time, preferrably shielding it from fan air,
especially if the fan speed varies. Also apply a spot of thermal
paste to the probe, but not anything containing silver or aluminum.

Here's what Dow Corning says about its thermal pastes. Unlike Arctic
Silver, they actually manufacture their products, have tested them
thoroughly, and know everything about them. Nor do they act like
weasels and give out contradictory information ("AS5 has become a
standard in the wind power industry for use in IGBTs which may run
fairly high voltage", but we don't know if it's safe to such voltage
because we've never run any formal tests and have never used it with
anything above 12V):

http://www.dowcorning.com/content/publishedlit/11-1712-01.pdf

To be read later, along with the rest of your post.

Thanks a lot.

 

[Paul]

The instructions also say one can clean the cpu and the heat sink with
xylene (such as in Goof-Off) and high-purity isopropyl alcohol. All I
have is the isopropyl alcohol they sell at the drug store,

If I don't use that, beause drug store alochol is worse than nothing,
I assume I shouldn't use the xylene?

Is drug store alcohol worse than using nothing?

I just use isopropyl for cleaning.

And if I was using something more harsh, it would only be on the
heatsink bottom. I'd be a bit more suspect about what the
bare silicon die with coating can handle. Presumably there
are some organic solvents which are bad for it.

ArcticSilver makes a cleaning kit, which consists of two solvent
bottles. So that is another option. But I just use isopropyl,
which tends to smear around the stuff. It isn't the right
solvent, but does give you some fluid to work with (keeps
the cleaning cloth damp).

Paul

 

[larry moe 'n curly]

The tail lights etc. don't use thermal grease, do they? They use
dielectric grease, I think.

The dielectric grease used for them is thermal grease. Until Arctic
Silver came around, it wasn't common for thermal grease to be
electrically conductive, with some aluminum-based stuff being an
exception.

 

[VanguardLH]

When I assembled my CPU the instructions said to put "all the
contents" of a small syringe of paste on to the middle of the CPU and
then lever down into place. I remember the great deal of force
required to do this, and no paste was observed oozing from anywhere.
Still, it's working OK.

Some pre-supplied thermal paste syringes are pre-measured to apply just
a dot of paste. This is to help prevent ignorants that gob on the stuff
thinking it's supposed to be the thermal transfer material rather than
the far superior metal-to-metal contact between CPU plate and heatsink.

http://tinyurl.com/6g369na

Although the article claims this pre-measured amount of paste in the
one-use syringe is the proper dose, that's WAY too much paste. If you
used a thin plastic squeegee to spread it out evenly and at a proper
thickness, you would end up with a lot of paste/grease left on the
squeegee.

http://www.diy-gaming-computers.com/applying-thermal-paste.html

This is a better description of how much paste to use. The reference to
pea-sized is not for volume you apply but the diameter of the dot you
dab onto the CPU plate. As the article mentions, you should spread it
out so it is translucent, not so much that it is opaque.

However, I would say with the pre-measured syringes that you apply the
dot of paste, place the heatsink on the CPU, and then *twist* the
heatsink on the CPU to spread out the paste BEFORE clamping down the
heatsink to the CPU. Unless you lap both heatsink and CPU plate, you
won't be able to tell in which direction the two surfaces misalign and
the paste could ooze out in one direction and not get applied in the
other directions.

Because so many users screwup the application of thermal paste, many
CPUs come pre-supplied with pre-measured thermal tape. It liquifies
under pressure when heated. Yet I don't recall thermal tape ever having
the same thermal conductivity as paste/grease - but a better applied
thermal tape is better solution than poorly applied paste.

Alas the one crucial step that is missing is telling the user that they
should FIRST clean both surfaces (heatsink and CPU plate) *before*
applying the paste. A fingerprint is thick enough to prevent the paste
from contacting the surface at that spot. Don't ever expect the
heatsink and CPU to be clean out-of-the-box. You'll notice they never
include a low-residue lintless cleaning wipe with the included paste.
Sometimes you see instructions about using a paper towel to burnish the
mating surfaces.

Regarding "pump out" of thermal paste over time due to the pressure of
the heatsink against the CPU plate, I beg to differ that the loss will
result in heat problems. What pumps out shouldn't have been there in
the first place. If the paste is doing a proper job of filling only the
microscopic gaps between the mating surfaces, what fills the gap won't
get pumped out. What gets pumped out is the excess that should not have
been there in the first place. Now drying out is another matter and
that means shrinkage which means the paste isn't completely filling the
microscopic gap and is no longer plastic to accomodate movement between
the mating surfaces.