Thermal transfer compound comparisonReview date: 13 March 2002.
Last modified 07-Jan-2013.
Overclockers love grease.
Thermal grease, to be exact. "Thermal transfer compound", if you want to be formal. It's the stuff you put between your CPU and your CPU cooler, to aid in the transfer of heat from the former to the latter.
All modern PC CPUs produce enough heat that they need a heat sink. Almost all of them need a heat sink with a fan. Many heat sinks come with some sort of thermal transfer thingy pre-applied - a patch of grease, or a square of chewing-gum-like semi-solid material, or just a rubbery pad for the low performance units. Most CPUs don't produce enough heat that the stuff you put between the chip package and the heat sink matters very much, as long as the computer case has decent ventilation and the ambient temperature isn't sauna-like. There just has to be something between CPU and heat sink.
The reason why there has to be something there is that the two mating surfaces of processor and sink aren't flat. They may look flat. They may have a mirror polish. But, on the microscopic scale, they look like a scale model of the Andes. And the mountains on one item do not match the valleys on the other.
Without thermal transfer compound, everywhere heat sink metal doesn't mate with CPU package material is a teeny-tiny air gap. Air is a good thermal insulator. As long as your heat sink looks flat when you lay a ruler on it then there'll be a decent amount of actual contact, of course, but the amount of heat that'll actually make it around the air gaps may be surprisingly small.
Hence, thermal compound. It's grease with lots of minuscule thermally conductive particles mixed into it, basically. It doesn't conduct heat as well as direct contact, but it's a heck of a lot better than air gaps.
A popular view among those of us who've spent more time cleaning thermal grease off our hands than we'd care to remember is that it doesn't really matter much what kind of thermal grease you use. Plain cheap white zinc-oxide grease, fancy silver grease, ultra-fancy super-exotic better-than-the-stuff-NASA-uses grease; they're all much the same. As long as you apply the stuff reasonably sparingly, you'll be fine.
I'd never actually tested this, though. Perhaps the marketing bumf for the current crop of exotic super-greases was right; perhaps they really are spectacularly better than plain cheap white thermal goop. Perhaps the fancy greases have advantages beyond their thermal performance, too.
Since I'd recently received a few new super-goops for review, it was time for a comparison.
Of which more in a moment.
Arctic Silver are the biggest name in exotic thermal goop, and they're up to the third formulation of their premium silver-loaded product. Silver is a superb thermal conductor, so you'd think it'd be a good choice for use in thermal goop. Arctic Silver 3 is now meant to be 70% silver by weight, and is now based on "a proprietary mixture of advanced polysynthetic oils that work together to provide three distinctive functional phases", rather than the usual silicone oil base. Silicone oil itself is a decent thermal conductor, and so, presumably, is the new Arctic Silver base.
The manufacturers go on, thankfully, to explain that the advanced polysynthetic whatever means the stuff should fill gaps well when freshly applied, yet not go all dry and flaky after a long time. It is, indeed, somewhat runnier than the earlier Arctic Silvers, fresh from the syringe. It also has a faint art-supply smell that it didn't have before.
AS3, like the earlier Arctic Silvers, is only trivially electrically conductive, though it's a bit capacitive. This means that an overlooked splodge of it across a couple of chip pins isn't likely to cause any problems, but painting your CPU with it is still a bad idea.
The special polywhatchamacallit oils in AS3 could be quite useful, if they mean that this stuff really will stay soft as it ages. Cheap thermal compounds can dry out after a few thousand hours, which causes problems for a few reasons.
First, wet compound that's dried out won't have the density, and thus the thermal conductivity, that it had when it was wet. These compounds aren't a solution, they're a mixture; remove the liquid from them and you end up with lots of little spaces where the liquid was, which is bad.
Things can then get worse if you move the CPU cooler - which can happen, if the case gets dropped or bumped. Then you can end up with nasty lumps and ridges of compound in the gap between CPU and cooler, and an even lousier thermal connection. On the smaller scale, the natural vibration of a PC could reasonably be expected to open up larger air gaps in dry and powdery thermal compound. The stuff has to stay at least somewhat pliable in order to keep working well.
At $AU19.80 for a three gram tube (delivered price from Aus PC Market; it's only $AU13.20 if you buy it with some other item), Arctic Silver isn't exactly a huge extra expense compared with the price of the rest of your fancy overclocking kit.
70% silver by weight multiplied by three grams of grease at the current $US4.50 or so price for a troy ounce of silver, by the way, means that a three gram tube of this stuff contains about 30 US cents worth of silver. So I wouldn't be hoarding Arctic Silver 3 in a safe, if I were you.
Three grams is still rather more than most people are likely to need. I've gone through a few tubes of Arctic Silver 1 and 2 myself, but I'm not exactly a typical user. The manufacturers say that three grams is enough for "at least 20 to 30 small CPU cores, or 7 to 12 large CPU cores, or 3 to 6 heat plates", and I'll go along with that. If you magically manage to avoid wasting any grease then you could make one tube stretch a lot further, but anybody who's used thermal compound knows that more of it will end up distributed approximately evenly around the computer room, and the person applying it, than will end up where it's meant to go.
People whose budget doesn't stretch to Arctic Silver can now buy a cheaper product, Arctic Alumina, based on aluminum oxide instead of silver. It's got the same advanced polysynthetic nuclear fish discotheque technology, but it doesn't work quite as well, or cost as much. A 1.75 gram tube of Alumina - still more than enough for most users - is $AU9.90 delivered from Aus PC Market.
Cooler Master have comprehensively defeated the grease-all-over-your-shopping-bag problem, by blister-packing the syringes that their thermal grease comes in.
There are two varieties of Cooler Master grease, melodically named PTK-001 and HTK-001. The PTK stuff is in the gold tube, and its quoted thermal conductivity is worse than that of the HTK stuff, which is in the silver tube.
The grease itself is just grey and white, respectively, and most of the other specifications are pretty similar. PTK is supposed to be useable at 50 degrees below zero while HTK isn't meant for use below freezing, but only people living in Siberia or playing with liquid nitrogen are likely to care about that. The two Cooler Master compounds do have very different quoted viscosity; HTK is 300 Poise (between chocolate syrup and tomato sauce), while PTK is 1700 Poise (about 0.7 times as viscous as peanut butter).
You get 1.6 grams of PTK, and two grams of HTK.
The extra Cooler Master packaging has a purpose beyond avoiding premature grease ejaculation. Inside, there's a stick-on stencil sheet for small and large contact patch processors, a credit-card-esque piece of plastic, and instructions. The instructions tell you how to stick the appropriate stencil onto the bottom of your heat sink and squeegee a line of grease across the hole in the middle, to get a neat square of grease of sensible thickness onto the bottom of the cooler. Just like a bought one.
If you've got your motherboard in front of you, then using the card to spread a smaller amount of grease on the CPU top itself isn't much harder. But this is still a neat feature for beginners. And it'll save you from getting grease all over your Visa card.
ESG Associates's Nanotherm thermal compounds have "multisynthetic, non-silicone carrier fluid", much the same as the current Arctic goops, but Nanotherm did it first.
The two Nanotherm varieties are Ice (white) and Blue (um, blue). They're the same stuff; the Blue variety is just cosmetically tinted. The tint may make it easier to judge the thickness of the layer of the stuff you're putting on an item. Apart from that, it makes no difference.
Both Nanotherms give you two grams of goop in a syringe with a nifty shrouded tip that ought to reduce the amount of goop you get on your fingers, and then transfer to your nose/forehead/fly zipper. Both Nanotherms also have the usual list of good qualities. The only unusual thing on the list, from the point of view of PC users, is the claim that they allow "Fast, Easy Cleanup - No Mess". That's an attractive quality in a thermal goop, let me tell you; I've gone through quite a lot of naphtha lighter fluid (it's useful for so many things!) cleaning goop off CPU-related paraphernalia, and off bits of myself.
And now, the ostriches running in the horse race.
The toothpaste is... well, it's toothpaste, all right? Requests for me to also test brand name pastes and/or Tartar Control and stripey variants will be treated with all the respect they deserve.
Vegemite, on the other hand, is an Australian thing.
For those who haven't tried it (as food, not as thermal compound), it's not as weird as you might have heard. Even I, an Aussie raised on the stuff who puts enough on his sandwiches that the type of bread being used becomes difficult to determine, can appreciate that even a smear of Vegemite on a cracker may be a pretty intense culinary experience for people unaccustomed to the distinctive flavour of concentrated yeast extract. But there are most definitely worse things.
It tastes as good as it looks, folks!
Buy some today!
To test the different goops, I used my standard CPU-simulating heater-with-a-thermal-probe-in-it, but without the usual copper piece on top of it to simulate the contact patch of a real processor. With no copper in the way, the heater's contact patch is far larger than that of a real CPU, and I only have to use one layer of goop.
Testing CPU cooling products on an actual CPU is good, because you're, well, testing on an actual CPU. But that's the end of the good news.
CPU temperature probe calibration is unknown. If you've got a probe sitting under the CPU (as most Socket A systems still have) then that probe's reading will probably be related to the CPU's core temperature with a pretty simple curve, but you don't know what that curve is, and you also don't know whether the sensor is actually reporting badly skewed numbers, generally because of a bug in the computer's BIOS.
These problems don't stop you making comparative measurements, of course, but they do stop you from being able to take one person's results and transfer them to another person's computer, even if that other computer has the same motherboard and processor.
It's also difficult to standardise the heat output of a CPU; there's no way to just click it into maximum heat output mode. You can run a standardised benchmark, of course - a 3D game in timedemo mode using software rendering will do - but you still can't tell exactly how much of the CPU's specified maximum power rating is being used by that benchmark.
And then you've got to allow for ambient air temperature - the warmer that gets, the warmer the CPU will be, all things being equal. And for the air flow reaching the CPU, which may stay constant or may not, depending on whether your CPU-fiddling moves cables around that influence the CPU's air supply.
Plus, CPU sockets will only last for so many cooler attachment cycles.
So I use my heater block.
I stuck a Thermaltake Volcano 6Cu+ cooler on the heater - I review this cooler in my monster comparison here. It's got a round copper slug built into its base, and the copper stands slightly proud of the aluminium around it, which made it easy for me to standardise the contact area from test to test. Slight alignment changes from taking the cooler off the heater and putting it back on ought to make little difference to the amount of metal in contact with the heater plate.
The copper slug also allowed me to apply the thermal compound to a neatly defined area of the bottom of the cooler, rather than slathering it all over the heater plate and increasing still further my contribution to the finances of Australia's paper towel manufacturers.
By monitoring the voltage across the heater's 4.7 ohm power resistor I can tell how much heat it's outputting. And with a dual-input thermometer that lets me see the ambient air temperature, the temperature of the probe inside the heater, and the difference between the two (delta-T), I can tell how well the cooler's managing to get rid of the heater's output, without the result being polluted by changes in the ambient temperature.
Divide the delta-T by the heater wattage, and you get a number in degrees Celsius per watt (°C/W). The lower the number, the better.
All of the results are still a bit on the high side, because the temperature probe inside my heater is rather close to the resistor's element. But they're repeatable, and good for comparison purposes.
First up, no goop at all. Bone dry. The result - 0.66°C/W.
Next, some plain white zinc oxide thermal goop, as provided in little plastic sachets with, or pre-applied to the bases of, lots of cheap CPU coolers. I applied the white goop in the time-honoured fashion, with my index finger, and didn't make any particular effort to keep the layer super-thin. Whole copper slug covered deeply enough that you could no longer tell it was copper. Lots of fingerprint marks. Luvverly.
The result was 0.50°C/W.
Next up was the Arctic Silver. I first applied it in the More Is Better style, lathering on a good half-millimetre layer of the stuff. Lots of people over-apply thermal grease, and I wanted to see what difference it might make.
In the real world, over-thick grease layers are thinned out considerably by the pressure of the CPU cooler retention mechanism, at least if you're putting the cooler on a small-contact-area CPU. But there's only so much that even the strongest cooler clips can do with the giant pancake-syrup grease dollop that many people use.
Anyway, the thick AS3 scored... 0.50°C/W.
Next, I scraped most of the grease off with my fancy Kyoto Ceramics Ceraplaner, which has seen a lot of service in my CPU cooler reviews. This scraper isn't any better than a credit card for the first few uses, but it's very, very hard and so can be used many, many times without losing its edge. A sharp edge is a good thing for a grease applying tool to have.
Anyway, I greatly reduced the amount of grease on the bottom of the cooler and tested again, and now Arctic Silver 3 scored 0.48°C/W. This is about as small a performance difference as my test rig can discern. But hey, the thinner layer did work better.
On to Arctic Alumina, again applied in a thin layer. This white grease has the same oily consistency as the silver AS3, and is just as easy to spread. And it scored 0.50°C/W.
I began to see a pattern.
Next, Cooler Master's gold-tube, grey-grease PTK-001.
This stuff is a big contrast to the current Arctic greases; it's really thick and dry, in comparison. It's not easy to spread, as a result; I really don't think the supplied application kit would work properly, as the grease sticks to itself much better than it does to metal, and can't be applied with just one pass of a blade.
The stiffness of this goop also means that an over-thick layer wouldn't squish out the sides of the contact patch nearly as well as thinner grease would. The stuff behaved like a cross between ordinary goop and chewing-gum-consistency pre-applied thermal interface material.
The contact patch of my un-spacered test rig is bigger than the top of any CPU, so I didn't use the Cooler Master application kit anyway, but just applied a layer to the bottom of the 6Cu+, as before. Despite the grease's stiffness, it wasn't very hard to get the bottom of the 6Cu+ covered with a decently thin layer of PTK-001 grease. And it performed just fine - another 0.48°C/W result. It wasn't any harder to clean away than the other greases, either.
The Cooler Master HTK-001 compound looks like ordinary cheap white thermal goop. Much wetter than the PTK-001 compound, not as wet as the new-fangled non-silicone compounds. It scored 0.50°C/W, just like the ordinary compound it resembles.
The Nanotherm compounds are less fluid than the average, though not as dry as Cooler Master's PTK-001. Unlike the PTK-001, the Nanotherm compound sticks to metal well, and is easy to spread very thinly and quite evenly. If you're a fetishist about using the absolute minimum amount of thermal goop necessary, then Nanotherm should suit you very well.
I tried an ultra-thin layer of Nanotherm Blue, which is what you'd want to use on really mirror-finished processors and coolers. The 6Cu+ is decidedly not a mirror-finished cooler, though; the smear of grease wasn't enough to completely fill the ridges on the bottom of the copper slug, and the score was only 0.56°C/W.
So I added more Nanotherm - and noted that the blue dye in the stuff wasn't evenly distributed, and the result was some rather fetching abstract stripes as I spread the stuff over the bottom of the cooler. Oddly, though, more Nanotherm didn't improve the 0.56°C/W result.
The reason for this seems to be that Nanotherm is meant to bed in better after it's had a couple of days of use. Some of it squishes out of the sides of the contact patch; what remains gets compressed.
Fair enough, then. I turned off the 6Cu+'s fan, turned the heater down to six and a half watts to keep the junction warm and encourage the goop to settle, and left it overnight. The next day, it managed 0.51°C/W.
I then turned the fan off again and left the heater running at full power for a few minutes, until the heater temperature probe read 100°C. Then I turned the fan on again, and now the reading settled to 0.50°C/W.
There's no reason to believe that it might not do a bit better again, if given more time. The large contact patch of my spacer-less heater rig means the oozing happens slower than it would with a real processor.
So this goop also works pretty much as well as everybody else's for plain heat transfer; you just have to give it a moment to bed in, or else magically get just the right amount in there in the first place, which is easier said than done.
As regards ESG's claims about the ease of cleaning Nanotherm up - well, it was a bit easier to clean up than the runnier pastes, but the difference wasn't anything to write home about. Naphtha Zippo fluid still did the job just fine.
Ok, time to send in the clowns.
Depending on the labelling laws where you live, the only ingredient listed on the side of a tube of fluoride toothpaste may be sodium monofluorophosphate. Sodium monofluorophosphate or sodium fluoride are the only active ingredient, in the medicinal sense, in normal fluoride toothpaste. There's also stuff to give the paste the right consistency, stuff to help it foam up, a mildly abrasive polishing ingredient, probably dicalcium phosphate dihydrate, and stuff to make it taste less like chalk jelly. It's water based, which makes it inherently useless as a real thermal transfer agent, because it'll dry out quite quickly. It may also cause corrosion.
There was nothing besides common sense stopping me from doing a quick test with the stuff, though, so I spread some on the 6Cu+'s base. The toothpaste I chose turned out to be slightly runnier than Arctic Silver 3, in case you care; it's pretty easy to spread it very thin. I put on a thicker layer, though; "optimising heat sink toothpaste application" is not something I intend to put on my curriculum vitae.
Firing up my Minty Fresh Thermal Transfer Experiment produced, rather hilariously, an excellent score - 0.47°C/W.
That's right - Toothpaste Superior To Arctic Silver 3! Film At Eleven!
This was only over half an hour, though. When I removed the cooler from the heater, the toothpaste that'd been squished out around the edge of the contact patch was already dry and crusty, and the rest of the paste would follow soon enough. It'd be just minty white dust inside a week, and probably wouldn't work much better than no thermal compound at all. Maybe worse.
But in the short term, and assuming you don't bridge any contacts with it (it's sufficiently conductive to cause problems), it would seem that toothpaste works just fine as thermal compound. If you're almost out of genuine thermal goop, can't get any more right away, and know you're going to be reseating a processor cooler umpteen times as you fool around with whatever it is with which you are fooling around - well, you can probably make do with toothpaste.
If you think I'm making some sort of guarantee that it won't cause unspeakably awful consequences, though, you're out of your mind. All care, no responsibility.
Actually, not even very much care.
On to the Vegemite.
Vegemite's runny compared with thermal goop, as well; nearly all of the Vegemite I applied squished out around the edges of the contact patch once the heat was on, and it then started drying out rapidly, just like the toothpaste. But the stuff still delivered a solid 0.48°C/W result.
Vegemite is, of course, an even less sensible choice for real purposes than toothpaste. Since it's salty, it's an obvious corrosion risk. But since the whole concept of Vegemite as thermal goop is inherently grounds for committal to a special place where the staff are very calm and the dinner forks have corks on them, I don't think corrosion problems are the single factor that rules it out.
By the way - after testing toothpaste and Vegemite back to back, and heating them up in the process, it strikes me that a toasted toothpaste and Vegemite sandwich would be an excellent thing to dare someone to eat.
Anyway, it's clear that inappropriate substances like the above pair work fine for the simple task of passing heat from one thing to another, as long as the junction temperature isn't high enough to boil them. I wondered why this was.
To see whether these substances work so well simply because they're wet, I tried some plain tapwater between the heater and the heat sink. And lo, it scored a magnificent 0.41°C/W - possibly at least partly because, without any gelling material, it was free to move heat by convection as well as conduction. In such a thin film, though, convection is probably negligible; it's the water's own rather good thermal conductivity, and the thinness of the layer, that are more likely to be the significant factors.
Of course, plain water's even less practical than toothpaste as a thermal material; once the water evaporates, there'll be nothing left in the gaps.
It does, however, bring home the fact that thermal compound is not actually a really good thermal conductor. That's why you're not meant to put it on with a trowel, people.
It would also appear that loading thermal compound with silver, or exotic oxides, or whatever the heck the thermal goop formulators come up with next, really doesn't seem to help much.
Exotic thermal goop can easily be justified in situations where a lot more heat needs to be transferred across the junction than any PC requires. But computer CPUs that draw more than 150-odd watts are a rarity, and most CPUs draw a lot less than that most of the time. If you're heat-sinking a big amplifier, or the motor in a high-performance radio-controlled car, exotic goop could help a lot. If you aren't, not so much.
There are lots of computer-tweaking situations in which the difference a given tweak makes is small. Small enough that it doesn't matter. Thermal grease seems to be one of those situations. Other good examples are tweaking up processor or RAM clock speeds on motherboards or graphics cards, when the things you can't run any faster are what's actually limiting the system performance.
This creates a problem for people who want to make such underwhelming tweaks sound exciting.
As a general rule of thumb, anything that makes a less than 10% difference to system performance doesn't matter. Not a 10% difference to CPU performance, or disk performance, or any other individual thing - a 10% difference to the actual speed of the computer for a real task. Since one subsystem is not often the whole limiting factor for a given task, you're generally lucky if you see a 5% real-world speed gain from increasing the speed of some component by 10%. A 5% gain isn't likely to be noticeable.
Because this fact doesn't make for a very good story, and because many people don't have much of a head for figures, it's easy to find situations where someone's found a small difference between things, but called it a big one. And made a graph to back up that claim.
It's easy enough to do that. Take, for instance, this honest graph of three values:
The values in this graph are 1023, 1060, and 5000. First two very much the same, last one much bigger.
Let's say that the first two numbers represent the StudlyMark2002 scores of Chip the Junior Overclocker's begging-for-death Duron box with a GeForce2 MX, before and after he's changed the epsilon margin interleave tolerance that's been added to the hacked BIOS he just flashed. And let's say that the last number represents the score of the computer that belongs to Susan across the street, which is a brand-name boring-box, but a new, shiny, really expensive one. And let's say that Chip would like his mad tweaking skillz to look more impressive than they are, and Susan's computer to look as if it's only barely better.
Here the same data are, after being put through the Mislead-O-Tron. Now the first two numbers look as if they differ by almost a factor of three; the last one still looks bigger again, but not by nearly as much as the real difference.
That's because now the minimum Y axis value is just below the lowest value, and the maximum Y axis value is not that far above the middle value. So Chip's two numbers look different, and Susan's number shoots off the top of the graph and gives the viewer no real idea of how different it is from the others.
Here's a similarly misleading view of the thermal paste results (click it for the legible full size image).
Hey! What a big range of values!
Here's the honest version.
If all this is new to you, then I strongly recommend you get yourself a copy of How to Lie with Statistics.
It's informative. It's useful. It teaches you exactly what the title implies, and along the way also teaches you how to use stats to tell the truth, should you for some reason wish to.
While I'm on a roll, allow me to also recommend The Visual Display of Quantitative Information , which not only makes you look smarter than Yoda if you leave the hardcover lying on your coffee table, but is also approximately one million times as interesting than the title might suggest.
There are reasons to buy fancy thermal compound. Pose value isn't really one of them; unless you're one of the "There's still some left in the tube! And I can still almost see the RAM slots! Squeeze harder!" kinds of people, nobody will be able to tell what thermal goop you've used once you've put the cooler on the CPU.
It is, however, good if thermal compound doesn't dry out. Ordinary cheap white grease will, in time. It's also good to get compound that comes in a neat little syringe instead of a hard-to-meter tube (which typically leaves you putting far too much of the stuff on and then having to clean it up again) or a treacherous little bag (which can explode inside the CPU cooler box before you even open it, as has happened with a couple of the coolers I've had for review).
It's also good to get compound that's easy to apply, but they're all pretty much as bad as each other, really. The PTK-001 Cooler Master grease is about the worst I've seen in this regard, but even it isn't very bad.
(It occurs to me that you may be able to make stiff, hard-to-work-with goop much more manageable if you add a couple of drops of solvent to it - naphtha lighter fluid, meths or what have you. If my results in these tests hold, diluted goop should if anything work better than straight goop, while that extra solvent is still evaporating.)
If you know you're going to go through a lot of thermal goop, then you might as well buy the plain stuff in bulk. Here in Australia, Jaycar will sell you a 10-gram tube of plain white goop for $AU3.95, or a 50-gram syringe of the stuff for $AU9.95, or a monster 150-gram tube for $AU19.95. Electronics stores the world over stock similar stuff in similar quantities for much the same small price, and it's all over eBay too. The drying-out problem isn't, generally, a real issue - especially for overclockers, who probably aren't going to keep the same processor for a whole year anyway.
Most people, however, only need a gram or two of goop. Any more than that will remain unused until, with any luck, technology advances sufficiently that computer assemblers don't need thermal grease at all any more.
In a real-world situation, exotic compounds may well work better than basic goop. Smaller contact area, more pressure, more chance that peculiar formulations will do something remarkable in this somewhat unusual situation in which the thing being cooled must never get hotter than about the boiling point of water.
And, compared with the performance of the solid rubbery thermal pads that still come on basic CPU coolers [note that I wrote this back in 2002; I don't think stock CPUs have come with that stuff on them for a while now], any grease will perform very well. Pre-applied chewing-gum thermal compound can work about as well as grease, but solid pads aren't nearly as good.
If you think any particular grease is going to cause your CPU to run more than a few degrees Celsius cooler than any other grease, though, then assuming the grease is properly applied, you're dreaming.
If your PC's so marginal that a CPU running three or four degrees Celsius warmer will crash it [or, for modern CPUs, cause the processor to auto-throttle itself and substantially reduce system performance], the solution is not to try to edge away from the precipice with better thermal compound. It's to make a big change to the cooling system, or just lower the darn clock speed.
Such a computer will crash when the room ambient temperature rises a lousy four degrees. It will crash when its intake fan filter gets clogged with dust. If it has no intake filter, it'll slowly fill its innards with dust instead. That may not crash things outright, but it can, if a dust blanket insulates some other component and causes it to overheat. And, furthermore, if you know your computer will crash at CPU temperature probe reading X°C, then allow me to assure you that running the CPU at temperature probe reading X-4°C is not likely to give you a rock-solid stable machine at the best of times.
Is fancy thermal goop as good as the manufacturers would like you to believe, and as various Junior Overclockers do believe? Nope.
Is it, therefore, a complete waste of money? Absolutely not.
Just be aware that it's not magic, and don't expect the most expensive stuff to work very much better than generic grease.
Buy Arctic compounds!
Readers from Australia or New Zealand can purchase Arctic Silver 3 and Arctic Alumina from Aus PC Market.
(if you're NOT from Australia or New Zealand, Aus PC Market won't deliver to you. If you're in the USA, try a price search at DealTime!)
Cooler Master grease kindly provided by Rectron Australia.
ESG Associates' site (makers of Nanotherm compounds)
Nanotherm compound kindly provided by Technomancer.