TIM application test.

bluezero5 · Jun 3, 2012

I got a request to release this.
This was done around 2 months ago in fact, by 4 people, spanning over 3 weekends (5 days). All these people doing tests are in my group of Overclocking friends. (9 of us total, all Uni Grad students/Prof/Lab Techs). So these are all professionals. A special thanks to all those who took part. We have to run a 24 hour lab, so we come in at shifts. (I happen to be the supervisor) and while we do our normal daily work, we do these tests inbetween intervals. You can imagine this is actually quite.. intensive.

The debate was on TIM application. and we decided to just do a test over it. The test was as follow:

We used 1 PC only With Quick Disconnects on the CPU Block, so we can change the CPU block without affect too many variables. The System is water cooled with Distill water, and we purposedly made the Tubes LONGER, so when we mount and dismount the CPU block, it be easier.

We tested 3 blocks.

1, XSPC RayStorm
2, Koolance 370
3, EK Supreme HF
4, -- scrapped-- Corsair H100 was not tested due to time constraints.

Each mount was tested at least twice. (some thrice, when variance was larger than 1'C in tests between same mounts.) so this is a test data of 68 mounts total.

MX4 was used for this test.
CPU is a 3930k at 1.48V at 4.9Ghz.
To make the temperature difference more obvious, we purposely ONLY used 120.1 radiator from Swiftech.

In reality, when 120.2 Rad used, the temp difference will EVEN BE SMALLER.

Definitions:
Loose Mount: Mount just touching CPU. Screwed in Causally.
Normal Mount: Hand tightened til can No longer go any further.
Tight Mount: after hand tighten, we use tools for an extra 1/4 twist per screw.

PEA TIM: just a pea size TIM per mount. Usually means 75% of the Heat dissipater covered when we inspect.
Cross TIM: we use the cross Method, usually means 90% of the Heat dissipater covered when we inspect.
Film TIM: we use a thin fim covering CPU, always 100% covered, usually cause a bit excess squeezed out on the sides, which is cleaned up with a clean Q-tip.

Test Procedure and Temp probing
Then we put the computer on LOAD with P95 for 30 mins each.
due to our sensors, our results are to the nearest 0.5'C only.
and ambient temperature is always 25'C as standard Lab room temperature is controlled.

You will see Normal and Tight mounts here are actually very similar in result.
Loose mount comparison is where it gets interesting.

View attachment TIM application.bmp

Conumdrum · Jun 3, 2012

So tight mouts win. Okay. I can see that on all counts. Propery install cooler, it's good.

bluezero5 · Jun 3, 2012

Conumdrum said:
So tight mouts win. Okay. I can see that on all counts. Propery install cooler, it's good.

For EK and Koolance, Normal Mount and Tight mount basically same result.

I think the main take away here is:

Always try to cover the Heat Dissipator with TIM, more surface area in contact = better
In General, tighter mount = better. (common sense. you have mount, you WONT squash the CPU, don't worry, unless you used wrong mount screws)
The tigher you go, the less the TIM amount matters. (also common sense)

do note, since we only used 120.1 rad, the temperature difference seems over >1'C, however in real life, when you have a min of 120.2 for CPU, we expect the difference to be dampened. So all of the above test should really yield less than 1'C difference in general. Except for the loose mounts, but in reality, you should NEVER have a loose mount anyway.. we tested it just as a sample of 'bad' mount.

wingman99 · Jun 3, 2012

Good test interesting results.

m0r7if3r · Jun 3, 2012

I'd be interested to see the results repeated with one block and more mounts to really single out which tim application is the best, since cross and film were close enough to each other that I'd say they were within the margin of error. Great testing! Certainly gives you something to think about.

bluezero5 · Jun 4, 2012

thanks. Credit goes to my team really. I was personally only responsible for the koolance block testing. But the fact the result so tightly corresponds to each other.. is a sign the experiment was done in a good controlled environment.

I mounted twice for normal and tight, and 3 times for the loose experiement. and yes, the cross/film method shows little to no difference on normal/tight mounts. which means the cross method is good enough. I prefer the thin film method as though I need to clean the sides, it really shows NO downside in applying more over less. and I don't mind the peace of mind.

But yea, the thin film usually squeezes out a little on the sides.. take note.

m0r7if3r · Jun 4, 2012

bluezero5 said:
thanks. Credit goes to my team really. I was personally only responsible for the koolance block testing. But the fact the result so tightly corresponds to each other.. is a sign the experiment was done in a good controlled environment.

I mounted twice for normal and tight, and 3 times for the loose experiement. and yes, the cross/film method shows little to no difference on normal/tight mounts. which means the cross method is good enough. I prefer the thin film method as though I need to clean the sides, it really shows NO downside in applying more over less. and I don't mind the peace of mind. But yea, the thin film usually squeezes out a little on the sides.. take note.

You've seen Vapor's TIM testing, right? This reminds me of that...

bluezero5 · Jun 4, 2012

I have seen it yeah.
but we decided to test it anyway.

I think between the two test pieces, there can be some additional conclusions drawn.
So all in the name of experimental science.

Our idea was to test the relation between TIM and the Waterblock.

As you know, TIM is required cause neither the CPU or Water block is perfectly flat OR smooth. and without TIM, as tight as you can get, there will still be 'air gaps'. Theese air gaps can be cause of the shape not matching. (slightly convex surface) or because of Microscopic pores. The idea of TIM is to fill these gaps. Of course you want the waterblock to touch the CPU 100%, but that can never be achieved. As such, TIM which acts like a liquid metal can seeps into these gaps and improve contact. (written by Captain Obvious)

The best thermal conductor is actually DIAMOND, next to Silver, then Copper.
However, it is also noted that Diamond though can conduct heat well, behaves less well in most TIMs, so we decided to try MX4 then.
(right now, we trying Arctic Silver 5.. which seems.. 0.5'C worse maybe?) but doesn't matter. I will be changing it enough to switch to another in 2-3 weeks. lol.

RGE · Jun 4, 2012

What results you find regarding tim application, often depends on what tim you test, ie thick vs thin, etc. Also spreading concern is more over long term voids from trapped air bubbles expanding and coalescing, not as much immediate effect.

Vapor has done a ton of testing mounts, and got better results using pea in middle, he stated that more than once, and he wrote that in his testing.
http://skinneelabs.com/deep-cool-z3-z5-z9/

Spreading TIM isn’t the most effective way of applying it. It readily introduces air bubbles into the TIM joint–air is what TIM is trying to displace–and also requires applying more to get ‘full’ coverage of the IHS.

Back when I was testing heatkiller vs ek sup hf, and testing apogee xt vs hk, doing many mounts of each, first thing I tried was which tim application gave the best and most reproducible method, and that was a measured pea in the middle, but was partly tim variable. And always, first several mounts were not temp tested, but done until I could measure pea size of tim + consistent mounting pressure, and know the footprint covered the entire IHS. But each type of tim spreads differently. Mx2 or mx4 whether 4mm pea size or 8mm pea size, tested very little difference, and line vs spreading was small difference (short term testing), though I was more consistent with measured pea in middle.

Some thick, drier shin itsu tims, size of pea mattered, as even with fairly high and consistent measured mounting pressure, you could end up with thicker bond line with larger amount of tim, ie it doesnt squeeze out the sides as readily as mx2, mx4, pk1, AS5. Also manually spreading thick tims gave worse temps compared to pea size, thin tims very little difference over short term.

But the bigger issue is not the immediate effect of manually spreading, but that manually spreading traps tiny air bubbles which expand from heat and coalescing over time and may create voids which degrade performance over time.

Whether you conclude rice,pea,x,line or manual spread and whether amount matters, will vary some based on characteristics of tim you are using, and likely shape of IHS/block you are testing. Also will depend on whether you test after weeks when voids can form from trapped air bubbles, or whether testing an immediate effect. And some tims, ie grease based, manufacturers including their testing depts like dow corning claim do not suffer voids, even with manually spreading. So again tim specific.

Though I agree, mounting pressure is most important variable.

bluezero5 · Jun 4, 2012

rge said:
What results you find regarding tim application, often depends on what tim you test, ie thick vs thin, etc. Also spreading concern is more over long term voids from trapped air bubbles expanding and coalescing, not as much immediate effect.

Vapor has done a ton of testing mounts, and got better results using pea in middle, he stated that more than once, and he wrote that in his testing.
http://skinneelabs.com/deep-cool-z3-z5-z9/

Back when I was testing heatkiller vs ek sup hf, and testing apogee xt vs hk, doing many mounts of each, first thing I tried was which tim application gave the best and most reproducible method, and that was a measured pea in the middle, but was partly tim variable. And always, first several mounts were not temp tested, but done until I could measure pea size of tim + consistent mounting pressure, and know the footprint covered the entire IHS. But each type of tim spreads differently. Mx2 or mx4 whether 4mm pea size or 8mm pea size, tested very little difference, and line vs spreading was small difference (short term testing), though I was more consistent with measured pea in middle.

Some thick, drier shin itsu tims, size of pea mattered, as even with fairly high and consistent measured mounting pressure, you could end up with thicker bond line with larger amount of tim, ie it doesnt squeeze out the sides as readily as mx2, mx4, pk1, AS5. Also manually spreading thick tims gave worse temps compared to pea size, thin tims very little difference over short term.

But the bigger issue is not the immediate effect of manually spreading, but that manually spreading traps tiny air bubbles which expand from heat and coalescing over time and may create voids which degrade performance over time.

Whether you conclude rice,pea,x,line or manual spread and whether amount matters, will vary some based on characteristics of tim you are using, and likely shape of IHS/block you are testing. Also will depend on whether you test after weeks when voids can form from trapped air bubbles, or whether testing an immediate effect. And some tims, ie grease based, manufacturers including their testing depts like dow corning claim do not suffer voids, even with manually spreading. So again tim specific.

Though I agree, mounting pressure is most important variable.

what you observed is actually expressed in our tests too.
in loose mount test, the thin film performs basically same as the cross method, same with normal mounts. I ti only until we go 'quite tight' do the thin film clearly outperforms. This can be due to many reasons, air bubbles might be one reason.

that being said, the waterblock itself also has a lot to do with how the TIM spreads. from our results, the Koolance blocks sees to have the most 'consistency' in mounting. this can be cause of the chip we used or the mounting springs we use, cannot be concluded, all we can tell is, between different mounts, some waterblocks will likely have more consistent results over others. For our experiment, we support koolance 370, other mounts might support other blocks.

and yes, airbubble will always be an issue.
and in fact what we observed 'but not tested', is that the action of squeezing the TIM out of the tube actually creates lots of air bubble in the TIM. (and probably the factory settings of putting the TIM in the tube too.) for the Pea/cross method, we try to eliminate this by spreading 'very slowly', for the thin film method, we try to eliminate this by spreading very thin and evenly. Regardless, our results agree, Air Bubble, is a terrible mix in any TIM.

However, our tests didn't like the pea method. can be mount specific. or for whatever reason. But logically, given the pea only on average gets 75% of the surface covered, I can see some theoretical gaps for that method. To each their own though. of course, everyone should always test for their best results, and should only take the above test as a specific example, no more.

in fact, I think the only observation from our tests, should be = TIGHT = BETTER. heh.

RGE · Jun 4, 2012

yeah mounting pressure is no doubt king.

And I agree, all methods have a few drawbacks, and if I only got 75% coverage with pea, I would not use that method either. But I always do a few "footprint" mounts that waste tim first, ie mount and wipe off, until I can reliably cover the entire ihs with pea, because, yeah if only covering 75% that will not test the best.

bing · Jun 5, 2012

Actually its quite easy to test the tim spreading process by using a small piece of thick glass and press it against the base.

Don't use thin one, cause it might crack under the pressing force, thickness about > 8mm is ok and has small area like below example, it can even sustain almost my whole body weight when pressing it on the heatsink using my palm with pressure includes at the glass corners. :shock:

So far single drop or line method in the middle is the best even for thick with low viscosity tim, and you can see it for your self how the bubbles is forming if the TIM was spreaded.

Also if you had a chance to try it, just try to remove any formed and trapped bubbles only by twisting the glass (simulating the base twisting movement during fastening) without lifting it, trust me, its impossible. :shrug:

One small detail that could make significant different, once the tim is spreaded properly without any bubbles, try separate the contact by lifting the glass just a little, you will see how the air rushed in and when you re-apply the force again, air bubbles will be formed like crazy. Thats why during installation and fastening the base to cpu, never and ever lift the base even a bit, unless you are going to clean the whole tim and restart the process all over again.

Fyi, regarding tim application using spreading method, even you can't see any trapped bubbles, say like they're so tiny for eyes to see, try heat up the glass using hair dryer to simulate IHS, you will be surprised how those micro sized bubbles will expand so large that they will become visible.

Vic Velcro · Jun 5, 2012

Thanks, bing. I'll be using the glass method from this day forward.

EarthDog · Jun 5, 2012

rge said:
yeah mounting pressure is no doubt king.

And I agree, all methods have a few drawbacks, and if I only got 75% coverage with pea, I would not use that method either. But I always do a few "footprint" mounts that waste tim first, ie mount and wipe off, until I can reliably cover the entire ihs with pea, because, yeah if only covering 75% that will not test the best.

Ive done this before and came up with a method that gets great coverage.

I do a slightly smaller than pea size in the middle, then hit each of the corners with a bit of TIM as well. That way it reduces the chances of the corners not getting TIM. Great coverage everytime. :thup:

bing · Jun 6, 2012

Vic Velcro said:
Thanks, bing. I'll be using the glass method from this day forward.

My pleasure.

Once you've played with that piece of thick glass + tim on cpu IHS few times, pretty damn sure you will be an expert on how much tim is enough, and also will learn a lot on how tim spreads at different mounting angles or different fastening sequences, or found your own trick/style how to apply & mount for free bubbles and etc.
Its just too complicated to explain, its better to experience it yourself.

PS : If you don't want to waste those precious TIM, just use toothpaste, hell, even peanut butter will do, use the smooth one not the chunky of course.

bluezero5 · Jun 7, 2012

I think it is funny how we all went our means to go as far as we did to test these things. All in the name of a better system.

Normal people won't understand, but to us, every 0.5'C matters! haha.

bing · Jun 9, 2012

bluezero5 said:
I think it is funny how we all went our means to go as far as we did to test these things. All in the name of a better system.

Normal people won't understand, but to us, every 0.5'C matters! haha.

Agree, and personally I'm more interested with the consistency & repeatability of the results cause this matter big time when it comes to compare seriously on various TIMs performances.

Personally any TIM reviewer that did > 1 mounts always have my respect and trust.

Tank Geek · Jun 9, 2012

Ok I gathered TIM is thermal paste but what does it stand for?

bluezero5 · Jun 9, 2012

Thermal Interface Material

Conumdrum · Jun 10, 2012

TIM can be Ttape (sekisui), Tpaste (MX-4), Tepoxy(AS Epoxy, used it on my SB chip long ago), or even Tpad (like on a GPU cooler for ram etc) I guess. I've used them all over the years.

TIM application test.

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Member

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Member

Member, Water Cooling Sticky Reading Enforcement O

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Member, Water Cooling Sticky Reading Enforcement O

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Senior Member

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Senior Member

Low Profile Senior

Member

Gulper Nozzle Co-Owner

Low Profile Senior

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Low Profile Senior

Winner, Rig-o'-the-Quarter, Fourth Quarter 2012

Member

Similar threads