Thermal Paste Comparison, Part One: Applying Grease And More

[ihog]

"Since GPUs don’t have a heat spreader, but allow the cooler’s sink to directly sit on the die..."

Not all GPUs don't have heat spreaders. My MSI Twin Frozr II GTX 570 has a heat spreader.

 

[Calculatron]

Using the spread method won't cause overheating, unless it is a pretty botched job in a bad situation (high ambient, hot processor, etcetera). It just has a higher likely-hood of capturing air bubbles, especially since people tend to apply way too much thermal compound. It is just easier and more foolproof to use a type of dot-or-line method to achieve maximum heat transference. (There is nothing saying you can not achieve the same performance with the spread method, there are just more factors at play.) As a general rule, the actual difference is going to be measured in the single-digits, if there is a difference. Even then, we have to remember that a lot of people use 2C as a margin of error.

HardwareSecrets has done a number of thermal compound round-ups, and only chocolate has done worse than no thermal compound at all. Butter, lipstick, diaper ointment, and toothpaste have all been used to varying degrees of success. To speak facetiously, as long as you have *something* between there, it is going to work.

 

[butremor]

Whoever's responsible for Charts should put more work into them. For example i'm reading Viscosity chart which is measures "score (in Score)" from 1 to 10, which tells me nothing. Should I interpret 1 as most liquid or solid? Next example, CPU Air Cooling which is in Kelvins? Which means first paste in chart is https://www.google.com/#q=32K%3DC ? Forgive me if i'm being ignorant but that's sloppy. What I'm saying is, you need to add title to charts which would say: "← Less is better" or "Less is more viscous"

 

[slomo4sho]

Those benchmarks suggest that CL Ultra and Pro are still the best solution but it should be noted that they do corrode aluminum. However, was there a reasoning why only CL thermal pads were used for the GPU benchmarks?

Also, thanks for including the denture cream in the benchmarks. I assume that was interesting to take off after the tests

 

[NightLight]

more articles like this one please!

 

[jay2tall]

How the heck did Innovation Cooling Diamond (IC Diamond) thermal paste not make it on your list of test products? This stuff is literally the gem of thermal pastea... Literally, it is a GEM. I have used many pastes and this is the first one that has really impressed me and isn't some outrageous price.

 

[ojas]

Really nice read!

Just a quick question, though, aren't the Ivy Bridge thermal issues less due to paste being used than the fact that an air gap exists between some parts of the die and the IHS?

For Haswell of course, there's an additional factor: FIVR.

 

[smeezekitty]

For Haswell of course, there's an additional factor: FIVR.

Yes. Putting a heat producing voltage regulator in package wasn't necessarily such a good idea.

 

[wysir]

I am currently running an H100i and IC Diamond paste on a Phenom II 1100T overclocked by base Turbo socket overclock (3.84Ghz) with idle temps at 19-21c. Will you'll be covering IC Diamond and would you'll recommend using Contact Clearner (non-conductive) for cleaning off old paste applications?

 

[InvalidError]

How the heck did Innovation Cooling Diamond (IC Diamond) thermal paste not make it on your list of test products?

From what I understand, the review process usually works something like this:
1- the (would-be) reviewer shops around for products to test
2- the reviewer then sends a form letter to all manufacturers to request test samples by a certain date
3- the reviewer receives samples
4- the reviewer may (or not) purchase additional products at his own expense
5- the reviewer tests them, writes the article and submits it
6- the reviewer's paycheck gets adjusted to account for reasonable expenses

So if the product is not in the list, the manufacturer either did not send a sample or the reviewer did not know or forgot about it.

 

[g-unit1111]

Awesome, glad we finally have a guide for this! Can't wait to see part 2!

 

[bigshootr8]

@The Von Matrices I believe you are referring to deliding a cpu where you remove the gunk on the actual cpu portion and apply new thermal. kind of like this http://www.youtube.com/watch?v=FirsCzTnbdQ

 

[InvalidError]

Just a quick question, though, aren't the Ivy Bridge thermal issues less due to paste being used than the fact that an air gap exists between some parts of the die and the IHS?

I read an Ivy Bridge delidding article from someone who used caliper and paper shims to find out exactly how thick the gap between the CPU die and IHS was.

He measured the thickness from CPU substrate bottom to IHS top, de-lidded, removed the epoxy, measured the IHS thickness and thickness from CPU substrate bottom to CPU die top. From that, he calculated the gap from CPU die top to IHS bottom and this thickness was a few paper shims thick - I do not remember actual numbers... lets say 0.1mm - I remember it was surprisingly thick.

The guy then replaced Intel's paste with popular off-the-shelf pastes using shims to reproduce the gap on Intel's stock IHS and found out that none of the commercially available thermal pastes outperformed Intel's paste with that much of a gap, proving that Intel's "crappy paste" is actually very high performance stuff that gets ruined by excessively thick gap between the IHS and CPU die.

He then re-tested popular pastes with fewer shims to reduce the gap's thickness and that produced the expected thermal improvements.

So from his testing, it seems pretty clear that the problem is not the quality of Intel's TIM but the mechanical gap between the CPU die and IHS.

When you bring two surfaces to physical contact with paste in-between, the distance between the two near-perfectly flat surfaces is dictated by the most coarse particle size in the paste which is usually less than 10 microns or ~10X thinner than the standard gap under the IB IHS.

 

[nukemaster]

Would a very thin copper shim between the heat spreader and the cpu die fix it up?

 

[InvalidError]

Would a very thin copper shim between the heat spreader and the cpu die fix it up?

The simplest fix is simply to remove the silicone/whatever glue that props the IHS off the CPU.

But the reasons why the IHS is propped up in the first place are likely at least twofold: keep the IHS's edges off the CPU substrate so it does not grind through the top insulation layer due to vibrations leading to an eventual short and prevent the CPU die from getting crushed/cracked by the combined forces of the socket clamp and HSF: the bulk of Intel-based systems are shipped in prebuilt boxes so I would not be surprised if part of Intel's decision to go paste for mainstream CPUs had to do with soldered CPUs dies getting damaged by shipping rough-handling.

 

[nukemaster]

easy enough

 

[tomfreak]

It would be much more helpful if Tom tell is each and diff die positions for each diff CPU type 775 to 2011 platform or AMD solution as well.

 

[nukemaster]

You can find that exact info on the Arctic Silver website

 

[jimmysmitty]

I do on new coolers(I am betting the microscopic stuff does not come out when you clean a cooler after the first use), but it has never seemed to make a difference.

That's when you use something like Big Bath ISO or Jiffy Bath.

I had plenty of CPU sockets that got covered in TIM (Intel LGA and AMD PGA) and that stuff melts most TIMS off the CPU and sockets.

It would be much more helpful if Tom tell is each and diff die positions for each diff CPU type 775 to 2011 platform or AMD solution as well

I think Intel has been pretty consistent since LGA1156/1366 quad cores in die location and size. Before that (775) they had 2 dies next to each other.

Not sure on AMD.

Still if you put a peas sized dot in the middle and give it proper pressure, it will hit all but the corners of the CPU normally.

 

[FannBlade]

Great job on the review! Looking forward to part 2. It's time to put all the obsolete methods out to pasture.

 

[ojas]

Just a quick question, though, aren't the Ivy Bridge thermal issues less due to paste being used than the fact that an air gap exists between some parts of the die and the IHS?

I read an Ivy Bridge delidding article from someone who used caliper and paper shims to find out exactly how thick the gap between the CPU die and IHS was.

He measured the thickness from CPU substrate bottom to IHS top, de-lidded, removed the epoxy, measured the IHS thickness and thickness from CPU substrate bottom to CPU die top. From that, he calculated the gap from CPU die top to IHS bottom and this thickness was a few paper shims thick - I do not remember actual numbers... lets say 0.1mm - I remember it was surprisingly thick.

The guy then replaced Intel's paste with popular off-the-shelf pastes using shims to reproduce the gap on Intel's stock IHS and found out that none of the commercially available thermal pastes outperformed Intel's paste with that much of a gap, proving that Intel's "crappy paste" is actually very high performance stuff that gets ruined by excessively thick gap between the IHS and CPU die.

He then re-tested popular pastes with fewer shims to reduce the gap's thickness and that produced the expected thermal improvements.

So from his testing, it seems pretty clear that the problem is not the quality of Intel's TIM but the mechanical gap between the CPU die and IHS.

When you bring two surfaces to physical contact with paste in-between, the distance between the two near-perfectly flat surfaces is dictated by the most coarse particle size in the paste which is usually less than 10 microns or ~10X thinner than the standard gap under the IB IHS.

Yup that's what i read too. I'm not really sure how well a known fact it is at the non-manufacturing side of Intel, at least Intel India apparently isn't aware (manufacturing or design isn't done here).

 

[Dextron]

Holy crap, I almost had a heart attack after reading this article and thinking of my GTX Titan, GPU painted with liquid metal (CLU) - RISK OF ELECTRICAL DAMAGE!!? - Please hurry up with the second part.
And thanks for a really helpful article; I always thought that the spreading methods you presented were better, but everyone seemed to support using the "spread all-over" technique; that didn't really feel right.

 

[Lutfij]

Best read I've had in quiet a while - since the Sandy Bridge review. Good work Igor!

 

[damian86]

That is a very good study Igor, ps I am looking forward to know more about liquid metal solutions since they claim high conductivity, such as the Coolaboratory Pro/Ultra. I was very tempted to buy one of them to fix my ps3 overheating issues but I don't think anything will fix it.

 

[Traciatim]

Holy crap, I almost had a heart attack after reading this article and thinking of my GTX Titan, GPU painted with liquid metal (CLU) - RISK OF ELECTRICAL DAMAGE!!? - Please hurry up with the second part.
And thanks for a really helpful article; I always thought that the spreading methods you presented were better, but everyone seemed to support using the "spread all-over" technique; that didn't really feel right.

Did you read the article from Puget that jtd871 linked? http://www.tomshardware.com/forum/id-1834941/thermal-paste-comparison-part-applying-grease.html#11697549

Their comparison the pre-spread method performed very well. Everyone uses the 'air bubbles' reason as why you don't want to, but in almost every spread method they have bubbles. All the methods they picked also performed fairly well. As long as you don't use too much or too little, that seems far more important than your spread method.