18
6
Following the relevant question I was wondering if adding a pea-sized (or half-pea since there's already some) would hurt ? As far as I'm aware, and have heard from various sources, Thermal paste is one of the few things in this line of work, where "The more the merrier" applies in full force, so long as no paste touches anything but the top of the CPU. CPU is i5-7600
Иво Недев
Posted 2017-04-12T08:07:18.460
Reputation: 498
79
No, adding more would be bad. What you want to do is to clean off all existing paste (use isopropyl alcohol if you can) and apply a bit of fresh paste.
If you're talking about the layer that comes with a new cooler, you can usually use it directly - you don't need to use your own at all. Replacing paste is only really worthwhile for old paste.
Also, the correct saying here is "less is more"1 :)
With thermal transfer from the integrated head spreader (IHS, the metal on top of the CPU die) to the heatsink, it vaguely goes:
So your best-case scenario is if you can maximise direct metal contact between the IHS and the heatsink. That means they should be as clean and smooth as possible, and a fair amount of pressure pushing them together.
Now, if direct metal contact is best, why do we have paste? Because it's very difficult to get solid metal smooth enough for perfect contact, so you inevitably end up with lots of tiny air bubbles, resulting in poor transfer. Adding paste fills up these little gaps, but adding too much paste1 will either form a thick layer and prevent direct contact, or will end up getting squished out the side.
Even worse is trying to apply fresh paste on top of existing old/dried paste - that way you have the poor performance of dried paste (which can no longer spread effectively once disturbed) plus an additional layer. It's much better to just clean off the existing gunk first.
1 You'd want enough paste. There's a bit of leeway here, but you also don't want to go squeezing a whole tube in - once you have enough, adding more won't help. Keep in mind that what looks like a tiny bit will actually spread out quite far once pressure is applied - you're squeezing a 3mm-high blob into less than a tenth that height. Optimally, you'd have somewhere maybe a little bit over enough.
For those interested, there's further discussion of specific application techniques and their relative effectiveness here: https://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/
Bob
Posted 2017-04-12T08:07:18.460
Reputation: 51 526
4Also, if the thermal compound comes out of the sides, you can end up with a short on the CPU (depends on the CPU and if the thermal compound is conductive). Or, the best case scenario, next time you need to do some maintenance, the CPU is glued to the heatsink. This happens a lot and is a real pain to clean up properly. You may risk destroying the CPU or dropping it while "unglueing" the CPU. (I've done the last one, and bent 5 pins on a CPU.) Also, it is a real pain to clean layers of thermal compound, or nearly impossible to remove the older layer. – Ismael Miguel – 2017-04-12T10:21:44.700
5In fact the advice used to be to apply a small amount and scrape it level with something like an old credit card, removing much of that small amount at the same time as making the layer even. The thermal conductivity of thermal paste is really low, just much better than the air it replaces. – Chris H – 2017-04-12T10:41:51.300
1Also beware that some coolers have thermal compound pre-applied. – Pieter De Bie – 2017-04-12T10:44:31.980
@ChrisH I find it much better and less risky to introduce dirt/dust is to let cooler pressure distribute the paste. – Иво Недев – 2017-04-12T11:11:34.787
@ИвоНедев it depends on how much pressure you can apply. Modern CPU heatsink clamps probably apply enough even for demanding use if you start with it slightly spread out. I've used these pastes many more times on other devices than CPUs, and the 90s/early 2000s CPU coolers I first used them on had flimsy clips that didn't apply much enough pressure to spread a blob out to the edges, or to squeeze out excess. If you apply it in a normally clean environment and spread it with a clean tool you'll be fine. – Chris H – 2017-04-12T11:15:28.027
1I don't agree with 'less is more'. How much thermal paste you apply makes little to no difference, except when you apply to little. (or way to much) – raznagul – 2017-04-12T14:16:20.713
1
Building off of @raznagul's comment. The 'too much thermal compound' is really fear-mongering. See the LTT video, which actually ran an experiment on thermal performance on various application methods (https://www.youtube.com/watch?v=r2MEAnZ3swQ). Most thermal paste is electrically non-conductive (https://en.wikipedia.org/wiki/Thermal_grease), so that argument doesn't hold up either. In reality, the pressure of the heat sink upon the CPU will squeeze out excess compound, leaving a uniform layer given a sufficient amount of thermal compound.
– Tyzoid – 2017-04-13T01:31:55.6332
@Tyzoid You seem to be missing the context here, specifically the intention to add more paste on top of existing paste. I stayed away from recommending a specific amount and linked to a very good article for a reason - the intention of this answer was to explain why paste is used and why more paste does not automatically translate to better cooling. If you really want my opinion on how much to apply, I favour the X pattern, which is a fair bit more than absolutely necessary. Doesn't really matter much.
– Bob – 2017-04-13T03:02:48.153I'd also note that it's common for beginners to apply far too much, since they tend to forget about the spreading behaviour. "less is more" was a counter to "the more the merrier" and not originally intended to be an absolute instruction... but apparently you took it that way, so *shrug* – Bob – 2017-04-13T03:05:16.877
2I'd note puget systems builds systems commercially and thermal paste is meant to fill gaps, not act as a thermal interface on its own. A lot of builders literally just squeegee a thin coat atop their processors. Quite a lot of thermal paste is conductive, especially on the high end. Adding enough to squeeze out excess is not recommended by many makers of paste. I'd rather trust Puget than Linus personally. – Aibobot – 2017-04-13T03:25:02.207
@Tyzoid less is more is a good general rule with thermal paste. Only in the case of good heatsink clamps is it not relevant. Even then the pressure required to get a thin layer even in the centre of a CPU is enormous. Probably enough to bend the heatsink. – Chris H – 2017-04-13T08:41:40.853
14
Absolutely not. Thermal paste should be just enough to fill any gaps. A thicker than required layer of thermal paste reduces the efficiency of the paste. It is also not a good idea to mix different thermal pastes unless you know they are chemically compatible. Additives in one paste may break down additives in the other, producing compounds that may degrade the paste.
David Schwartz
Posted 2017-04-12T08:07:18.460
Reputation: 58 310
Do you have an example of such chemically active additives? All thermal pastes I have seen use some sort of oil/grease as a base and non-conductive powder as filling (the cheapest kinds sometimes use metal powder). I've never seen any warnings about chemical compatibility issues on thermal paste. – Dmitry Grigoryev – 2017-04-12T15:53:08.000
5@DmitryGrigoryev Many manufacturers warn about this and caution users to thoroughly clean off old thermal compound before adding new. The only incompatibility I've specifically heard of is thermal compounds containing halogens being incompatible with thermal compounds containing micronized carbon. – David Schwartz – 2017-04-12T17:37:15.597
6
There's a reason heatsinks aren't made from enormous formed blobs of thermal compound. The BEST thermal paste is about 8W/m^2*K. Even steel is about 6 times better at conducting heat, at 50W/m^2*K. Aluminum is 205. They're not even close -- use the least amount of paste you can to fill the air gaps (air being 0.024W/m^2*K). You can actually skip thermal compound entirely if you lap the heatsink and CPU to a mirror polish -- high end overclockers do this occasionally.
Bryan Boettcher
Posted 2017-04-12T08:07:18.460
Reputation: 250
3Logically, those overclockers should also mount their heatsinks in clean room conditions, because a few dust particles will prevent good contact even if the surfaces are perfectly polished. – Dmitry Grigoryev – 2017-04-12T15:59:27.993
2@DmitryGrigoryev, it's not a clean room, but taking steps to minimize any chance of dust is part of a typical overclocker's heatsink-mounting procedure. – Mark – 2017-04-12T23:10:36.590
1@Mark Does the quality of the thermal junction last for long? It'd seem like thermal expansion cycling (the CPU expanding/contracting as it warms/cools due to load/idling) plus vibrations (such as from system fans, especially fans/liquid-circulation on the CPU cooler) would eventually introduce particulates, moisture, or some other type of imperfection. Not that that's a problem for extreme overclockers who're going for short-term extreme performance numbers rather than long-term performance, but just curious how stable that solution might be. – Nat – 2017-04-13T05:14:57.727
1@Mark What steps are these? When you have 10 dust particles per cubic cm, you'll get some on your cooler no matter how hard you blow. – Dmitry Grigoryev – 2017-04-13T07:06:30.153
@DmitryGrigoryev: Presumably he meant to say "minimize the amount of dust", and/or reduce the chance of any large concentrations of dust. – Peter Cordes – 2017-04-13T07:21:13.120
@DmitryGrigoryev, "not blowing" is one of those steps. Anything that stirs up the air will also increase the dust concentration. Other steps are things like cleaning with a lint-free cloth and quick-evaporating solvent just before applying the thermal compound to minimize the time available for dust to settle. – Mark – 2017-04-13T08:33:11.613
1@Mark It's funny how the discussion went from skip thermal compound entirely to applying the thermal compound. – Dmitry Grigoryev – 2017-04-13T11:08:39.820
1Manually lapping the CPU and heatsink isn't going to give you any benefit - it'll still be rough at the microscopic level - which is why we have the paste in the first place. – Baldrickk – 2017-04-13T12:59:14.933
@DmitryGrigoryev: Clean polished metal to metal contact have been known to weld parts together. This happens most often in space where there is no air between the parts in the first place but it also sometimes happen in regular workshops and garages. The trick to getting rid of the air gap is to slide the parts together not letting air enter the joint instead of pushing the parts together trying to squeeze air away from the joint. – slebetman – 2017-04-13T15:52:50.587
I gotta say that this whole no-thermal-paste thing sounds really dubious to me. I mean you can probably get away with it with significant care, but it shouldn't be able to outperform a thermal-paste-based solution unless you have advanced industrial tools to finely craft a perfect fit. This shouldn't be reasonably possible to do at home. – Nat – 2017-04-13T19:37:55.123
If anyone can find a link demonstrating a no-thermal-paste demonstration that's arguably competitive with a good thermal paste, that'd be very interesting to see. Presumably should be easy to find it if exists, because the entire point of extreme overclocking is bragging about it online. – Nat – 2017-04-13T19:40:47.067
@Nat: I've done it, it worked well enough, although I don't have the equipment to tell you if it outperformed a thermal-paste-based solution. Figure-8 wet-lapping-on-glass from 400 to 3000-grit sandpaper with both the heatsink and CPU was enough. It took about 15 minutes. – Bryan Boettcher – 2017-04-13T19:41:28.153
Finally, someone who backs their claims with numbers. – ivan_pozdeev – 2017-04-13T23:41:28.473
1
You should apply enough paste so that when you put the cooler on, a small amount of paste appears on the sides. Putting less paste means you're still having air gaps which were not completely filled. Putting more means you're wasting paste, and if you apply too much it may spill on the board and you'll have to clean that.
Of course, you should completely remove the old paste before you apply the new one. The old paste has probably dried up compared to a fresh one, and probably accumulated some dirt and dust. This will prevent it from flowing well under pressure, and you will end up with more air cavities as a result.
Dmitry Grigoryev
Posted 2017-04-12T08:07:18.460
Reputation: 7 505
4Here's a thought: How do you want to apply a thicker layer of thermal paste, if the distance between heatsink and mainboard is pretty much fixed under pressure thanks to the locking mechanism? – Ian – 2017-04-12T11:02:34.493
11
And certainly, not like this: https://gfycat.com/GraciousActiveCoral
– Dai – 2017-04-13T00:51:41.8803
@Dai I told ya guys, the more the merrier: https://img-9gag-fun.9cache.com/photo/abpBb2L_700b.jpg
– Иво Недев – 2017-04-13T15:50:00.470note: Puget Systems posted an excellent article on thermal paste application techniques. tl;dr: best coverage was from an X shape, diagonally from corner to corner atop the chip.
– matt lohkamp – 2017-04-13T23:48:53.263@Dai I, too, like to use mayonnaise as thermal paste. – Derek 朕會功夫 – 2017-04-14T18:27:53.037
1
@mattlohkamp LTT also did thermal-paste application testing, and consistent with Puget's results, it doesn't really matter. Unfortunately, none of them are really scientific, as they only have one trial per condition, so random variations could easily swamp out the 0.25 ℃ difference in measurements. Also note that Puget isn't reporting ambient temperatures, so if they drifted more than a degree or two, all their results are pointless.
– Nick T – 2017-04-14T22:33:58.957absolutely. this shouldn't be taken gospel truth - but if you're wondering where to start, I think it's a fair guide. – matt lohkamp – 2017-04-14T23:46:21.990