Can a thermal paste scratch your CPU die?

I opened up my M17x R4 to see why I was getting 90+ temps on the 3720qm. I had pasted with ICD 24 about 6 months prior, and my computer was getting HOT. Removed the heatsink, cleaned everything with a microfiber cloth and 99% isopropyl alcohol, and much to my dismay, the CPU die did not look great at all. It had scratches/visible deterioration on the die surface in some parts, with other regions of the die appearing perfect. No matter how hard I tried to rub off the marks, I determined them to be scratches. Could this be a result of the thermal paste? Anyhow, repasted with MX-4 now and seeing MUCH MUCH better temps. At 4Ghz won't even break 80 C, and at idle it sits around 38-44 C

 

removing the thermal paste wrongly can cause scratches.

 

Yeah ICD is known to scratch dies if you remove it improperly

 

Wonderful. I usually just use a microfiber cloth + alcohol, but I guess it did some damage to the die. IC should have a warning on the label.

 

Since those scratches have no real impact on performance I can see why they might not.

 

Anybody remember the Innovation Cooling CEO's epic meltdown and attempted coverup when users on some forums accused ICD of scratching/tarnishing their dies (with photographic evidence)? LOL

 

They don't? I thought proper thermal transfer was all about good contact, surely scratches would impede proper contact between die, compound, and heatsink?

 

This is technically what thermal paste is designed to do, fill in the microscopic grooves on the surface of the die so that no air gets trapped. So it would also fill in small scratches as well.

 

Don't get me started

This is one of the reasons why I never have and probably never will buy anything from IC. Besides once you've experienced what Liquid Ultra can do it's really hard to go back lol

 

Yeah I've got enough OCZ Freeze (awesome stuff BTW before it was discontinued) stocked up to last me a lifetime so I have no need to ever go thermal paste shopping again, but even if I did need to, I would never buy from IC again on moral grounds. I have one tube of ICD 7 lying around that I purchased before that whole scandal took place.

 

I had used ICD in the past, until the die of my i7-920XM got scratched. Bye bye ICD, hello MX-4. Cooling performance drop negligible, no need to worry about scratches anymore, plus MX-4 is cheaper.

 

damn, it appears i'm not the only one. My die looks like some of those pictures on that link, not to mention my heatsink too. Guess I know not to use this anymore, wish I'd known this could happen after I've put it on everything I own. No more ICD for me, MX-4 does a wonderful job.

 

I've always wondered why people worry about scratches on the CPU-GPU die but I guess they are as important as scratches on the actual screen

I actually got the best results with ICD stuff but also OZC freeze so at the end of the journal what matters is what works best for you

 

LOLWUT? "I might scratch my die if I'm not careful, so I won't use that any more... but I'll risk bricking my $2000 laptop by using a pure metal compound"... ???

I have been using ICD for years. I have noticed scratches, but it has *never* impeded performance even after a dozen or so repastes. And the best way to remove the compound, honestly, is to soak it with 99.9% alcohol or some other solvent for several minutes then wipe it clean. I usually just soak a paper towel and wipe it off. That's akin to rubbing medium grit sandpaper across the die. I know this, but I'm also cognizant of it, so take due care, in the same way you take due care not to get any liquid ultra on any place other than the CPU die.

I don't doubt that Liquid Ultra does a fine job, just that it seems a bit hypocritical.

MX-4 is not even close to ICD from the testing I've done. Haven't used OCZ Freeze yet though, will have to give that a try too.

 

How the CEO of IC responded irked me a whole lot more than the fact that ICD scratches dies. So I'm avoiding ICD largely on moral grounds as well. And having experienced what Liquid Ultra could do for temps there's even less incentive to try ICD.

Liquid Ultra will also etch your die if you remove it wrong, and if you leave it on for too long you'll have to literally sand it off. But if I'm risking scratching the die anyway, might as well get something that will give the most performance. Yeah there's always the risk of shorting, but from what I've read as long as you don't apply too much the risk is minimal. The real issue is having to repaste every 2-3 months to prevent it from drying up on the die.

 

Yea having to repaste that often is a total deal breaker for me

Sent from my Nexus 5 using Tapatalk

 

With the added benefit of risk of total system failure each time as well

Sent from my Nexus 5 using Tapatalk

 

Once every 3 months really isn't THAT bad. Plus it's supposed to last at least half a year based on what others are saying, I'm doing it every 3 months because I'm paranoid.

Yes the risk is always there, but can be kept to a minimum if you observe all safety precautions. Honestly, it boils down to much risk you're willing to take to get the absolute lowest temps possible. That being said, Liquid Ultra will be very useful for summer gaming in a hot room with no AC (ambient temp consistently above 80F or 27C).

 

You missed the part where I said I avoid them "on moral grounds." I don't care if ICD scratches my die, and the performance is honestly not that great compared to OCZ Freeze and it's way harder to apply. But even if it were the second coming of Jesus I wouldn't buy it because I won't support a company run by an asshat like that.

 

Before we degenerate to flaming and fanboyism let's look at the facts.

In short? Definitely yes.
The reason is because IC Diamond has a much higher diamond particle filler loading than other pastes. Its basically why it is so difficult to apply but is also the secret as to why it performs so well with crap HS contact with incredible stability.

Its basically equivalent to toothpaste (which is what many use to Lap heatsinks) in terms of abrasiveness. To remove it, you liberally apply acetone and very slowly get it off.

In terms of performance
Technically yes many other pastes can vastly outperform IC diamond in terms of raw thermal conductivity but very very few can match its ability to remain practically inert at high sustaibed temperatures and the capability to bridge large gaps (most TIMs fall completely flat at high film thicknesses). I often refer to IC diamond as a liquid thermal pad rather than a true TIM.

 

I agree. IC Diamond technically should never have to be replaced, and there is no real "set time". Although I understand moral reasoning for not wanting to use it based on the CEO's attitude (after actually reading the posts more closely, my bad). I just avoid acetone because that stuff can melt plastic and other things if not careful.

 

Wouldn't isopropyl alcohol (rubbing alcohol) work too? Never used ICD however isopropyl works great one the pastes I've tried. It's friendly to most plastics (all I've tested on).

 

I don't consider scratching the die a result of "improper removal" since its nigh impossible to remove it without any scratches. And ICD will degrade like any other paste, so you CPU won't be looking too great after a few reapplications (doesn't seem to affect temps though).

That said, I still use ICD because it's the best performing TIM out of the few I've used.

 

I use it to remove the thermal paste dried and still crusty thermal and it has worked fine for me so far. And it works good on clean the board to as it remove crud off the board. I use it to clean the clean board and touchpad and palmrest without no problems. I also use it to clean the thermal off the heat sink as well and it does a good job at it. I use the grey thermal paste itself and those have worked fine for me.

 

Overall, acetone is usually considered a stronger solvent. IPA might work for ICD, but it may need more time, it also tends to evaporate somewhat faster than acetone (which also evaporates crazy fast). PCBs should be resistant to acetone, that isn't to say that you should dip them in the stuff, but a little shouldn't hurt them. Avoid stuff like dichloromethane (methylene chloride) which are downright nasty.

 

At work I often clean stuff with denatured alcohol as I find it's the gentlest on materials and fastest to evaporate.


Sent from my iPhone using Tapatalk

 

Denatured alcohol as in a mix of ethanol and methanol/isopropanol?

Anyways, if you wanted to go overkill, you'll be hard pressed to find something purer than that:

 

holy crap tijo are you also a fellow chemist?

but just so you know, Sigma sells 99.999% pure electronic grade 2-propanol, which is totally appropriate for electronic stuff considering it's electronic grade

 

I'm a PhD candidate in chemical engineering. Ironically, your electronics grade IPA is cheaper than my HPLC grade IPA. Not that I would ever use IPA that costs 85 CAD per liter to clean thermal paste.

If I ever get bored, I guess I could do a PCB vs solvent test. I've got a ton of different solvents in the lab.

 

Oh cool, I do synthesis and it's not everyday that someone knows about methylene chloride and its nastiness.

Also HPLC grade stuff is just more expensive for some reason. I'd like to think they carefully filtered the solvents so they don't mess up the pumps but that's probably wishful thinking.

 

I would say that it's so they don't mess up the columns rather than the pumps. I've used methylene chloride when I was polyepichlorohydrin and polyesters with a similar structure since it was one of the rare solvent they were soluble into. Well, that and doing the polymer synthesis in methylene chloride with diacid dichlorides (instead of carboxylic acids) and a dialcohol was way easier at the lab scale. Dealing with the solvent could be a pain though.

 

Bingo! a truly insightful observation Marksman30k.

Shin Etsu fits the mold of what an OEM requires besides high thermal conductivity they also require a high bulk loading for extended life. A high bulk loading is what gives it it high viscosity.

Shin Etsu Viscosity (Pascal * Second): 200
IC Diamond Viscosity (Pascal * Second): 260

Typical retail compounds are usually in the 50 - 80 Viscosity (Pascal * Second) Range which is why they are vulnerable to pump and bake out type early failures.

Fine surface finishes are more delicate than the base material itself lacking peripheral support for the finish structure and can be easily marred by much softer material than the sapphire/ruby/corundum/ aluminium oxide or diamond you find in thermal compounds which can only polish.

Both can polish to a mirror shine by intent in less than 20 seconds as noted in the demo below with either compound.

Noticeable scratches are generally wear and tear due to the delicate nature of fine finishes


I would note that in the following 1-1 comparison with IC Diamond and Shin Etsu and in point of fact our experience is that there is no intrinsic reason for it to be any different than any other compound, similar particle sizes will have similar polishing characteristics.


In the polishing test set up below IC Diamond is compared to ShinEtsu. Copper plate was prepped with .001 steel wool to establish a uniform texture grain from left to right, paste in then rubbed top to bottom against the texture grain.




Application pressure was approx 8-10 PSI measured on a shipping scale with about about a ¼ of an inch contact area this works out to be 30-40 PSI. This is 10 times the pressure you would use if you were lapping a CPU and 200 times our recommended removal pressures and so is a fairly extreme test.






Final result after 10 to 15 seconds application rubbing. The result shows no difference between the two, obviously all thermal compounds are polishing compounds. Shin Etsu contains aluminium oxide (same as most compounds MX4, Arctic Silver etc.) which is the most commonly used abrasive and is the main component in most sand papers



Below note the highlighted area which was enlarged and pasted in the upper left. These scratches were incurred with a plain soft paper shop towel purchased from Lowes hardware. Paper towels contain wood fibers and Numasists recommend not using paper envelopes as just removing and replacing coins the paper will scratch them. The soft paper towel scratches are deeper and more prominent than either of the compounds.




You do not have to be aluminum oxide or diamond hard to make scratches on many surfaces as textured surfaces are often weaker than the underlying material. The ugly scratches below were made with just a fingernail (Fingernail MOHS 2.75, Copper MOHS 3).




If you have a concern test it yourself take some IC Diamond and Shin Etsu and polish up a heat sink base and you will get a mirror finish in about 10 to 15 seconds with either compound, with no deep scratches.

AS5 will take 40 seconds MX4 about 30 seconds, all compounds polish about to the same shine as the particles are the same size. The difference in the time it takes is due to the amount and kind of filler used.

For example AS5 contains silver, aluminium oxide, zinc oxide and so with less abrasive filler you have a kind of emery cloth dilution so polishing is slower. MX4 with higher aluminium oxide content will be somewhat faster etc.


Misconceptions Regarding Diamond Hardness

In the nomenclature of the abrasion trade when they say “diamond cuts better than anything else” what is really meant is that it lasts longer. In the abrasion literature nearly all specifications relate to movement, RPM, tool feed, heat the particle material can endure under friction etc. So the assertion that diamond cuts better than anything else in what is essentially a static non cutting application one would be attributing characteristics that do not apply to the application and removal of thermal compound.

Example

Take two sheets of sand paper, one aluminium oxide the other diamond each with a 800 grit, approx the particle size of most compounds. Now scuff a couple CPU's with the sandpaper and you note no difference between the two and the result would be the same as our example above, Why? Because they are the same particle size and density as well as both being appreciably harder than the material the material being abraded.
The only time that hardness comes into play is after some number of cycles the aluminium oxide(MOHS 9) will begin to wear sooner that the diamond (MOHS 10) and that's it, the only advantage is that diamond has a longer working life due to hardness, there is no “magic” to hardness as implied, no increased probability of it scratching a component over another compound, simple.

Of course one could always use the product as directed by the manufacturer.

Removal

On the W/mK, there is no set standard as everybody tests and calculates independently/differently so unless you know how the number was calculated and the parameters of the test comparisons are meaningless. for example a reported 10W/mK on our test set up we have tested at 3.8W/mK. - number is good for marketing bragging rights I suppose.

 

Agreed here, also as my pictures have proven, it is easy to not leave scratches if removed properly. I even had 3 soft copper shims that were not overly harmed in multiple removals. One of the worst practices is using QTips. TIM can clog up and or build up in clumps. Strands are just about everywhere. Just horrible to use to actually scrape off the TIM.

I use QTips but to only dab on the cleaner. I then use dry PecPads, camera lens cleaning pads, to again dab off the TIM. It is a slow and time involved process but effective. Now fortunately for me I have not had a dried out ICD application. This even after a 12 month install with the proof pictures provided to ICD(Note: 12 month installation without performance degradation).

Now for a disclaimer; my install had 3 contact patches and copper shims were used to be sure all 3 patches were 100% level and flat. The installation then of ICD on all three areas essentially sealed well enough to keep the contact area moist and intact with no attempt of squeezing out the TIM on any one side of any die. This makes a great argument for proper contact area and EVEN pressure across the die. as an aside note the take down of the install after 12 months was done strictly to show the installations endurance after that time frame for the makers of ICD. My doing this was not only to prove the TIM can hold over time but to show what needs to be done (conditions) to make it do so.

Edit; it should be noted as well ICD was used on both sides of each shim. This means it had to transfer heat through two layers of TIM not just one. It still worked better than direct contact with any other TIM ( an aside for Tuniq TX-4. Extremely effective but thin and under extremes will boil out. (initial application performed slightly better than ICD))

Edit 2; I just realized the questions will be asked. TX-4 performed better for the first week, after the 3rd week it was worse than when I had used MX-4 for the same time frame. by 60 days it just had to be replaced, and I mean had too as the application totally failed. When I pulled it apart you could see the star pattern of TIM boil out from the center to the outer edges of each die. That is both CPU dies and the GPU. This was way worse boil out than ever seen with MX-4 or AS5 or even cheap compounds.

 

What few people fully appreciate is that if you are chasing optimal thermal conductivity, contact intimacy is king. Under tight contact conditions, plain water is actually one of the very best TIMs to bridge a gap with toothpaste being not very far behind. It has almost 100% wetting and a large thermal capacity per volume. The only reason it isn't a good idea to use water or toothpaste is the poor stability leading to curing or boil-out.
Designing a TIM is a delicate balance between wetting, stability and conductivity with many tradeoffs to be made.

 

More important than the thermal conductivity is the thickness of the TIM at the heatsink-die interface. My memory is a bit fuzzy but I distinctly remember reading about a threshold thickness below which the thermal conductivity was no longer the bottleneck and became an irrelevant parameter.

 

Is it this one?
Thermal Grease – Hitting The Wall? | Overclockers

Basically, he found that with a contact of 0.001 film thickness ( i assume its inches). Even a theoretical TIM with the conductivity of silver would only yield a benefit of 1 degree.

 

Took a quick look at the article and their heat transfer equations are sound. I'll probably have a bit of spare time this afternoon, so I could graph temperature drop vs thermal conductivity/thickness.

 

Small world, the author of that article was my business partner up until last year in the interest of disclosure.

Might find this one interesting

Diamond Thermal Grease Testing | Overclockers

 

Well, since I was bored, here's basically what the article Marksman linked in graph format.

Decrease in thermal resistance as a function fo thickness and thermal conductivity.


Temperature difference as a function of thickness for various thermal conductivities. Note that the scales are logarithmic for both axis, they make the data easier to see at lower thicknesses.


It's easy to see that the benefits of higher thermal conductivity decrease rapidly with increasing conductivity (resistance inversely proportional to the conductivity). The benefits of lower thickness remain linear though (directly proportional).

Another thing to note is that the lower the thickness gets, the lower the benefits in terms of temperature difference will be, so better thermal paste application of any good paste will work wonders compared to a poor application even if the thermal conductivity is higher. The data also makes a strong case to avoid thermal pads.

If anyone's wondering, the calculations were performed assuming 160 mm ² die area (Ivy Bridge) and 45 W of thermal dissipation.

 

Another issue not covered with the thickness of the TIM application is that then the TIM actually can act like a thermal mass. One that requires more heat from the chip side in order to get the TIM hot enough on the HS side to start the cooling process.

 

Excellent, yea this proves something I've suspected for a long time. You get a massive performance improvement with 1mm thermal pads vs 1.5mm thermal pads, yet bumping up the conductivity (with expensive Fujiploy pads) yields much less in terms of cost.

I had a Corsair H90 watercooler, it has excellent contact with my AMD 290 GPU die via the NZXT G10 bracket (over 80% of the die is covered), there is absolutely no thermal transfer difference between the generic white ceramic TIMs and the uber expensive He Nanogrease. However with the stock Tri X cooler, the contact was much poorer thus there was a difference of well over 10% between the best and worst thermal pastes.

 

So this may lead to a conclusion, that a cheap Arctic MX-4 may perform just as well as expensive IC Diamond, providing proper TIM application and proper die-heatsink contact. No need to spend more for an illusionary increase in cooling performance!

 

It was an academic paper, but the website may well have cited their results. I remember the threshold thickness being described as 0.1mm in the paper.

As long as 0.1mm (or whatever threshold thickness) of the TIM is enough to fill in all the voids, then that holds true.

I'm starting to think that the reason why Liquid Ultra performs so well (besides the obvious thermal conductivity) might be in part due to how viscous it is and how thinly it can be spread out. Application of Liquid Ultra involves taking a brush and literally painting the die with a thin film of it, and because of its viscosity it stays put and doesn't bleed outside of the die at all.

 

Agreed, I think it could also be because Liquid Ultra has really good wetting properties, akin to plain H2O thus completely filling up the surface.

Unfortunately, gauging good contact is difficult, pretty much all laptop cooling systems have poor contact. That H90 cooler actually had a flat machined base with a slight convexity towards the edge. Basically, if you tightened the G10 screws too much you could actually crack the GPU die. It is very very difficult to reproduce that kind of quality contact (especially in a laptop) thus there will always be a market and place for high conductivity TIMs.
I've used an addition silicone material to determine the pressure of a contact, most cooling systems cannot get the 0.1mm of contact.
I would more advise that every TIM has a place and purpose.

 

Thanks for this info, but not sure I quite follow. If the lower the thickness gets, the benefit is also lower? Wouldn't it be the opposite? The thicker it gets the the lower the benefit?

 

What I mean is that at lower thickness, the resistance to heat transfer is so low that even bumping the thermal conductivity of the TIM is if of little concern. Hence, the benefits of higher conductivity are lower with thinner paste layers. That is of course assuming perfect contact.

 

The problem is self defeating with the thin TIM's. The property that allows them to easily squeeze out is also the same property that allows them to boil out easily. Going back to an earlier test , this is what happens with Tuniq TX-4. It works great at first as it has high heat transfer properties in a super thin TIM. It then does loose transfer properties quickly and the worse the heat starts getting on the dies the faster the TIM boils out to the point of a total failure.

Now with those thin TIMs if you can, or already in the habit of a, repaste every week or so then the extra 1-2c at load may be worth it.

 

What we really need is TIM that has the consistency of honey, sticks like super glue, conducts heat like metal, and washes off like acetone on permanent marker. Liquid Ultra fills 3 of the 4 criteria, but is electrically conductive and a pain in the butt to clean when dried out.

 

Good observation, as a liquid it hits it's bond line thickness (BLT) at practically zero application pressure. The flip side is being a liquid it is a poor gap filler so it's advantages become muted under conditions.

Exactly right, but the real world is always less than perfect particularly with notebooks pressure is typically light and and the pressure required as noted to hit those bond lines would snap the MB and crack the chips.



Contact as well is especially poor in notebooks which is why bulk conductivity becomes important. The often reported w/m-K generally does not mean anything as you do not know how it was tested - at 100 micron or 30 micron thickness? In addition you do not know the calculation and as every manufacturer calculates their number differently so unless you know the method of test and calculation there is no basis for comparison.

So one company's 8.5 W/(mK) in another lab would perhaps test @ 3.5 W-mK.- mostly bragging rights for marketing, not much else.

below are user test results from a giveaway OCUK we did a couple of years ago compared to MX3



We are running another giveaway currently on OCUK and see at least initially no difference between MX3 and MX4 but would like to note that when making comparisons That you compare on similar platforms.

A notebook CPU or GPU while using less much less power actually has 2X the the heat flux due to the reduced area so you see delta T's that are double desktop numbers as a rule of thumb.

Desktop GPU can have an increased surface area over a CPU but can run up to 300W so even with the increased area heat flux is also higher a shown below.

 

Now imagine taking an entire heatsink from a 1mm gap to a 0.5mm (VRMs, VRAM and GPU core) gap overall. I keep telling MSI owners what a difference that makes but no one believes me lol.