AMD doesn't like Artic Silver (or other greases)

[yakkowarner]

From AMD's Builder's Guide for Desktop/Tower Systems (circa 2002/05 - this is not an old document):

[*]Always use an appropriate amount of an AMD-recommended thermal phase-change
compound (see Table 6).
Note: For production builds, thermal grease is never an appropriate solution.
Thermal grease can be used for short-term testing and validation. When used
for a longer period, thermal grease has a tendency to be pumped out from the
gap between the processor and the heatsink due to the differing thermal
expansion and contraction rates of the aluminum heatsink and the processor.

Table 6 is a list including items like the Bergquist HF225UT and Chromerics T725, pads included on heatsinks, which users of these forums regularly suggest scraping off and replacing with a thermal grease, Artic Silver 3 being the most loved.

People certifying that the greases work better than the pads indeed seem to confirm this with short-term tests. Frankly, as much collective experience as the Forums users have, I'm inclined to trust AMD's engineers on issues of long-term reliability. Can anybody explain to me why the official word from AMD is wrong?

 

[mechBgon]

My theory:

• Longevity: AMD gives a 3-year warranty on a retail CPU. The melt-to-fit PCTC they recommend evidently has the "staying power" to go 3 years with adequate performance, without needing to be refreshed periodically by the user.
• Consistency: Perhaps AMD is worried that a user will pick a lower-quality thermal compound (like that icky new Radio Shack stuff) will dry out or bleed out over time, resulting in warranty claims or a reputation for CPUs that lock up.
• Foolproof installation: AMD can't count on everyone knowing how to apply thermal paste well. The melt-to-fit pad takes care of this and can probably fill gaps to a small extent.
• Cushioning: The PCTC pad is a soft cushion that should help prevent novices from cracking the CPU core on initial installation.

The downsides, of course, are that the AMD-recommended PCTC pads are a one-shot solution (and very hard to find for purchase) and have lower performance than a high-quality thermal compound.

 

[MrThompson]

What I find interesting is if you go back a couple of processor generations and read AMD?s documents, AMD recommends the use of thermal grease instead of pads for the same reasons. You can?t have it both ways.

My thoughts? With the higher heat output, AMD has increased the strength of the clip. This takes its toll on the fragile cores. A thermal pad cushions the core during HS installation. Also, if you notice AMD carefully sizes the HS to the heat output of the CPU. This coupled with a thermal pad serves to limit overclocking

Here?s what Intel has to say about it.

Intel Xeon Thermal Design Guidelines

10.4.3. Interface Material Performance
Two factors impact the performance of the interface material between the processor and the heatsink base:

1. Thermal resistance of the material
2. Wetting/filling characteristics of the material

Thermal resistance is a description of the ability of the thermal interface material to transfer heat from one surface to another. The higher the thermal resistance, the less efficient the interface is at transferring heat. The thermal resistance of the interface material has a significant impact on thermal performance. The higher the thermal resistance, the higher the temperature drop across the interface and the more efficient the thermal solution (i.e. heatsink) must be to achieve the desired cooling.

The wetting or filling characteristic of the thermal interface material is its ability, under the load applied by the heatsink attach clips, to spread and fill the gap between the processor and the heatsink. Since air is an extremely poor thermal conductor, the more completely the interface material fills the gaps, the lower the temperature drop across the interface. In this case, thermal interface material area also becomes significant; the larger the desired thermal interface material area, the higher the force required to spread the thermal interface material.

Intel has determined through thermal characterization that it may be challenging to meet the thermal performance targets with the use of phase change thermal interface materials. The use of thermal grease in conjunction with high performance heatsink technologies (e.g. copper base folded fin) has been demonstrated to meet Intel thermal performance requirements. The use of thermal grease is recommended.

I have noticed that Intel recently edited the two other P4 Thermal Design Guidelines to remove the thermal pad recommendation and replace it with thermal grease.

 

[mechBgon]

Keep in mind that PCTC pads are a relatively new development. Back in the days of the K6's, pads were not phase-change material, and the contact area was large. Likewise, on a P4 or FCPGA2 processor, the interface between the heatspreader slug and the heatsink is very large, so there is low heat density and it's not likely to cause thermal grease to dry out. Just my opinion...

At work, I'm building AthlonXP systems and I do use AMD's suggestion: test with thermal grease to make sure it's a live CPU, then clean it off and use the retail heatsink and PCTC pad. If that's what gets me a 3-year warranty, that's fine with me.

 

[ShinSa]

I do feel that my artic alumina gets pumped out after long use.

MY CPU temp has risen from 45F to 51F over 3 month period.

When i reapplied artic alumina, the temp drops back to 45.

 

[yakkowarner]

So ShinSa's experience seems to confirm (though it's only one case) the idea that thermal paste might be better for a short time and just need frequest reapplication.
I wonder if the recent development of PCTC pads (how recent?) mechBgon mentions helps explain the divergence of most Forums users' and AMD's views on the subject.
Of course, that would still leave the question of why Intel would recently (how recent, MrThompson?) change their recommendations from PCTC pads (and they do specifically mention phase-change materials, not just pads) to grease. mechBgon's suggestion that area makes a difference between what's right for Intel chips and what's right for AMD chips sounds like an explanation, but I don't get how heat density plays a role and how grease would dry out any more or less depending on the size. (Heat density is, after all, temperature, and that's the number at issue; to say it's lower in one case or another begs the question.) But maybe the answer is also contained in that Intel doc:

In this case, thermal interface material area also becomes significant; the larger the desired thermal interface material area, the higher the force required to spread the thermal interface material.

This suggests that the point isn't the difference in grease's performance depending on area but the difference in pads' performance: with a larger surface area, it's harder to press the pad into the gaps without applying undue force. (Still, going by the label "phase-change", I don't picture a solid substance that needs pressing once the heat is on, but maybe?)

 

[mechBgon]

The first I can recall seeing the PCTC pads is with the arrival of flip-chip-style CPUs, and I think AMD was the first to get on the bandwagon there. The issue with the heat density is the raw temperature involved: an AMD CPU puts out a lot of heat through such a small area that it might bake the livin' daylights out of el-cheapo stuff like Radio Shack sells these days (I believe compuwiz1 confirmed this). A Pentium4's slug has many times more contact area and wouldn't need to get so hot. That's the degradation of the material itself due to very high temperatures at issue here, a separate issue from pump-out. I think pump-out is probably less of a factor with a large contact area too.

I think AS3 may resist pump-out better due to the simple fact that solid metal particles don't "pump" so easily. Maybe Nevin can drop in here with some insight on this... I'll see if I can send him a PM pointing out the thread.

 

[MrThompson]

I believe phase change pads have been available for over 20 years. One thing that needs to be noted, when phase change pads heat up, they become a liquid and are subject to the same ?pump out? problems thermal grease is proposed to have. Kind of makes you wonder if it?s marketing hype to cover a manufacturer?s derriere for a product that is under so much thermal stress, that the core can?t maintain reasonable dimension changes under normal operating conditions.

The first time I noticed Intel recommending thermal grease for the P4 family was this February, in the Xeon document I referenced. At the time, the socket 423 and 478 docs still recommended phase change pads. A little over a week ago I was doing some research on clock throttling. This is when I noticed the socket 423 and 478 docs had changed to recommend the use of thermal grease.

 

[mechBgon]

Maybe with PCTC, there's not only "pump-out" but an equal amount of "pump-in"?

As for the existence of PCTC, maybe I should have said I don't remember seeing it used on CPUs up 'til flip-chips came out. Have you seen it used on CPUs prior to that? Seems to me like the heatsinks and CPUs up to that point used graphite or rubbery pads...

 

[striker21]

I read MrThompson's link to the Xeon design guide, posted above. It matches his quoted section exactly. However, I was trying to look specifically for the P4 versions of those documents. I found this:
Intel ® Pentium ® 4 Processor in the 478-Pin Package Thermal Design Guidelines (see Appendix A). It looks just like the section quoted by MrThompson above, except that the last paragraph is completely missing from the P4 doc:

Intel has determined through thermal characterization that it may be challenging to meet the thermal performance targets with the use of phase change thermal interface materials. The use of thermal grease in conjunction with high performance heatsink technologies (e.g. copper base folded fin) has been demonstrated to meet Intel thermal performance requirements. The use of thermal grease is recommended.

What do you suppose that means about Xeon vs. P4 478-pin? On one hand, you might say that what's good enough for the Xeon should be good enough for the P4. On the other hand, I don't see where Intel specifically recommends thermal grease for the P4-478 as they do for the Xeon. Maybe I'm missing something...

Furthermore, in the version of this same doc for the 423-pin P4 here, the paragraph above does appear, along with the following:

Intel has determined through mechanical characterization that the use of phase change thermal interface materials
may lead to motherboard, processor, and /or surface mount component damage in mechanical shock or mechanical
drop testing. Phase change thermal interface materials create a strong adhesive bond between the processor package
and heat sink that can lead to large deflections and high stresses. The damage induced may not be readily
detectable.

Didn't remember seeing this in either of the other docs...

 

[striker21]

Of course, I guess that last part makes sense, based on what we know about the phase change material. It's more like a wax that melts into a plastic state and re-solidifies, than a grease. Probably not much of an issue unless your machine will be subjected to mechanical stress I guess that's the same reason that Intel can be confident that the phase change material won't run/bleed--at least for three years

Has anybody applied AS3 and actually left it alone? MrThompson, I think you had some of the first samples--you still using it as originally applied in any of your machines? I'd like to use some (and have a new tube here on my desk), but I'm not going into my machine each month to re-apply if it bleeds/runs. I know it's not supposed to, but...

 

[KF]

AMD says the grease pumps out. I'm sure it does a bit. Is that a bad thing? Maybe not, unless the interface is very bad. The best interface is direct contact. Having excess goo pumped out is a good thing. I always see a ridge around the CPU slug when I take the sink off. With carefully machined and flat surfaces, over the major portion of the interface there should be direct contact, with the grease filling the roughness and scratches. The clip should apply enough pressure to ensure that. I don't think what is left will pump out. (Who knows?) But a good surface may not always be the case.

I have always wondered about the deep grooved lettering on AMD slugs. It seems to me the large amount of material needed to fill those would cause the chip to have hot spots. Remember, those transistors are fractions of a micron, and all heat goo has poor thermal conductivity compared to aluminum/copper. Arctic silver is no exception, although it may not be as bad. The silver particles are suspended in grease (by design) and do not make direct contact, as proven by the fact it is a poor electrical conductor.

For any heat sinks I have taken off after a few years, the heat goo looks and feels dry compared to the newly applied stuff. I think the grease migrates to the edges under pressure (where it oxidizes or vaporizes) , and the solid material stays behind. For arctic silver the stuff remaining is so thin I can't tell what it's consistency is. Even on new stuff, the thin film left is hard to get off. (Haven't had any AS that was in use for years. Maybe six months.)

For TIMs I have had on for weeks, the material left is transparent. Most of it squeezes out forming a ridge around the slug. It looks about as thin as heat goo. For things like the K6, the grease did not squeeze out like this. It was not nearly as thin when the sink was removed.

I had a TIM on a 300MHz OEM Celeron that looked almost unused when I took the sink off after a couple of months. It just had dents corresponding to the corners of the slug. I don't think it ever got hot enought to melt it, and the pressure from the clip was pathetic by today's standards. (It OCed a lot better after I honed the slug flat and used Radio Shacke white heat compound.)

 

[striker21]

KF--
I wonder if that goo you saw pumped out was due to the initial setting of the HS on the core--apply a little too much goo, and I think your CPU core would instantly have that ridge of goo around it--even without the pumping effect. In any case, I think it's better to have slightly too much than slightly too little--that ridge won't hurt anything and is better than if there are air pockets between your HS and CPU core.

Thanks for relaying your experiences!

 

[MrThompson]

striker21 - My Xeons have been running undisturbed since late February. Nothing has changed since the initial settling of the compound. I am aware of systems that have been running the inital application of AS3 since January with no problems.

 

[KF]

Originally posted by: striker21
KF--
I wonder if that goo you saw pumped out was due to the initial setting of the HS on the core--apply a little too much goo, and I think your CPU core would instantly have that ridge of goo around it--even without the pumping effect. In any case, I think it's better to have slightly too much than slightly too little--that ridge won't hurt anything and is better than if there are air pockets between your HS and CPU core.

Thanks for relaying your experiences!

It is 99% the initial setting at least. I put on enough AS so it looks even and I can't see throught it. This only take a tiny drop of AS total. If I try to put on less it looks uneven and bare in spots. If I immediately take off the HS, as I have sometimes done (to check the result), the AS remaining is translucent. It looks like a faint gray fog, mostly on the HS. You can read the lettering in reverse as darker areas on the HS. As for how much might pump out, there is no way to tell because of this. The total quantity that is not squeezed out is neglegable from the look of it, so I don't think you could tell if any was pumped out even if you could apply the precise amount and not have any squeeze out initially. On a bady fitting HS it may be different.

 

[tuan121]

who cares !!!! like most ppl on anandtech keep there cpu for more than a year to year & a half!!!

 

[KF]

Quote

--------------------------------------------------------------------------------
Intel has determined through thermal characterization that it may be challenging to meet the thermal performance targets with the use of phase change thermal interface materials. The use of thermal grease in conjunction with high performance heatsink technologies (e.g. copper base folded fin) has been demonstrated to meet Intel thermal performance requirements. The use of thermal grease is recommended.

--------------------------------------------------------------------------------
Unquote


Then the description of the boxed Celeron 1.7G:

Included with the Boxed Intel® Celeron® Processor in the 478-pin Package

Intel® Celeron® Processor in the 478-pin package
Intel Designed Thermal Solution (includes high quality variable speed fan heatsink and clip assembly)
--->Thermal interface material (attached to the heatsink)
Installation Instructions and Certificate of Authenticity
Intel® Inside logo label

 

[FenrisUlf]

Also of note here is that AMD is using aluminum heatsinks rather than copper. Aluminum will expand and contract more with heat. Using a copper heatsink will reduce thermal expansion and contraction. Older heat sink greases had a problem with drying out after a while (I had some Rat Shack white goo turn into a solid after 2 years on my 486). Most of the newer high quality greases are supposed to be non-evaporative. Also, these TIM pads are intended for neophytes and assembly lines - places where there is either little experience or little time to get a good application of grease. They also help to protect the CPU core. A serious drawback of a TIM is that it can only be used once - a real pain if you have to take the heatsink off any time in the future. Probably not an issue for the Gateway/Dell crowd, but for someone who pokes in thier case alot, changes things, etc., it's an issue. TIMs work better with the small contact area and high pressure of the AMD or FCPGA contact patch than with the older, bigger cores. When AMD goes to a heatspreader like the P4, they'll probably revert to the grease. Look anywhere that engineers in other fields deal with heat transfer and heatsinks - paste is used almost exclusively over TIMs. Personally, I've had bad experiences with improperly installed TIMs - you have to have them heat up to a certain temperature before they will "melt" and conform to shape. This should be done by running the CPU but stopping the fan until it reaches a certain temperature - many times the installer will fail to do this and there's just enough heat transfer to keep this from happening, but not enough to keep the CPU happy.

 

[paralazarguer]

A way to prevent the pumping out of thermal compound from expanding and contracting is to using a distributed computing client like ECCP which uses all your spare cycles but doesn't slow down the computer when you need it. This keeps the processor under a constant load 24/7 so there is no expanding or contracting. It also helps other people out.

 

[Barnaby W. Füi]

Originally posted by: FlorianRed
A way to prevent the pumping out of thermal compound from expanding and contracting is to using a distributed computing client like ECCP which uses all your spare cycles but doesn't slow down the computer when you need it. This keeps the processor under a constant load 24/7 so there is no expanding or contracting. It also helps other people out.

thats part of it, but i think the larger issue is turning the computer on/off repeatedly. of course you could just tell people to leave the thing on all the time, but some people arent down with that, for heat/noise/paranoia/electricity/other reasons.

the issue seems pretty straightforward to me. use what comes with your cpu. (if you arent a techie)

if you are often removing the heatsink and/or want better temperatures, use (good) thermal grease instead of the TIM (if you use an amd cpu).

 

[wurmyhi]

This might sound stupid, but there's probably a couple of things at work...

The sticky stuff is sticky and keeps the heat sink from sliding around...

The stuff that is more thermally conductive is slippery, so the HSF slides around more and allows more material to be pushed away from the important contact area.

The things that make it worse are:
The mobo is typically mounted vertically and the hard drive(s)' movement is on a perpendicular plane.
High-pressure between the core and the heatsink squeezes more stuff out.


So, in the long-run, you'll trade better cooling for maintainability.

 

[FenrisUlf]

Originally posted by: wurmyhi
This might sound stupid, but there's probably a couple of things at work...

The sticky stuff is sticky and keeps the heat sink from sliding around...

The stuff that is more thermally conductive is slippery, so the HSF slides around more and allows more material to be pushed away from the important contact area.

The things that make it worse are:
The mobo is typically mounted vertically and the hard drive(s)' movement is on a perpendicular plane.
High-pressure between the core and the heatsink squeezes more stuff out.


So, in the long-run, you'll trade better cooling for maintainability.

Heatsinks don't slide around, at least not decent ones. Every P4 heatsink is bolted to the motherboard. They don't move. Some Socket A heatsinks like the Alpha 8042 are bolted to the MB. For the rest, the pressure and small contact area along with clips and indexing notches prevent heatsink movement assuming you're using a quality heatsink, which you probably are if you're even worried about this issue. Unless you're using your computer in some kind of circus act or let the Three Stooges movers get ahold of it, the heatsink isn't going to move around. Hard drive movement? Think about it - the heads are less than a gram and the case is thirty pounds. There might be a tiny vibration, but there's probably more vibration from the fans, especially the one on the heatsink. This isn't necessarily a bad thing - you want as little thermal grease between the heatsink and CPU as possible while still filling in all the tiny surface imperfections. Pumping out the excess can help.

 

[Futher]

I've found the solution.... we should all move from thermal grease to...... THERMAL PASTE!@# What now =)