TIM comparison at 120°F Ambient temps

[hxtasy]

Hey Guys, been lurking on the forum for awhile and figured i'd upload my own TIM test. Most of the test i see are at room temperature ambients. This test was done in a chamber held at 118°F / 48°C. I would also like to study the long term affects on the TIMs in higher temperature environments.

Processor: Intel E5-2440 2.4 GHz (dual)
Motherboard: Supermicro X9DBL-3
1U Server chassis
Passive cooling

Software: Proprietary software running CPU at 100% load and also stressing RAM.

TIM's:
Prolima PK-1
Shin-Etsu X23-7762
Coolermaster ice fusion

The coolermaster stuff was laying around for awhile and probably way past its shelf life.

Temps (C°) were measured from the rear processor. TIM on the front processor remained unchanged for more accurate results of the rear processor. Core temps also fluctuated +/- 2-3°C. i didnt average the fluctuations, this was just for a quick test.

the shin-etsu stuff didnt spread very nicely, is this normal or could the package i bought be out of date?

 

[petteyg359]

That picture shows effin' WAY TOO MUCH TIM... Just one of the dabs from the edge would be more than enough.

 

[jcw122]

As petteyg suggested, did you make sure you followed the manufacturers application method? A lot of TIMs have specific methods of application, and it does look like you overapplied, which can lead to poor temps.

 

[RnRollie]

must ... not ... comment ... on ... the ... CM ... package ... resemblance ... of ... holistic ... wart ... removal ... cream ... .


djee, what were they thinking at CM? Provide a lifetime supply of paste in one go?

 

[Tech Tweaker]

Way too much paste!

You need far less than that to be "covered," as it were.

 

[NiHaoMike]

120F ambient is very unreasonable for a desktop PC. It is realistic for a "naturally" cooled datacenter or a car PC, or maybe even a home server in a garage or attic.

BTW, a friend of mine told me a story when she had many PCs (including some Prescotts) Folding in her dorm room. One night, the A/C failed but the PCs kept on running. She woke up in the middle of the night very sweaty, at which point she took a cold shower and then left the shower running (with the door between the dorm and bathroom open) in order to provide emergency cooling. To her surprise, all of the PCs were still running. She didn't know the actual temperature but it was probably in the high 90s.

 

[usp8riot]

The test wasn't in optimal conditions but thanks for doing the work and the time you spent. It's a good first post.

 

[hxtasy]

I have found the X shape to be the best at spreading from center out decreasing air bubbles.

http://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/

There was another test i saw with the X and four dots, i will see if i can still find it, it was very persuasive.

Keep in mind this is a production environment, it is safer for an assembler to make an x shape and use too much instead of too little. the rest is just going to squeeze out, depending on the viscosity of the paste and the clamping pressure of the heat sink and processor. Squeeze out with this method was enough that it didn't go over the IHS.

I will do another test with the dot method or spreading it if anyone is interested.

Also these servers are going to be used in defense applications with very high ambient conditions. I thought it would interesting to post something other than another 25°C paste test.

 

[hxtasy]

As petteyg suggested, did you make sure you followed the manufacturers application method? A lot of TIMs have specific methods of application, and it does look like you overapplied, which can lead to poor temps.

this is simply not true.

 

[Conumdrum]

The amount of paste you applied is silly excessive and not good for temp reviews on an enthusiast site where proper amounts are applied. You can say this id what the manufacturer uses, but how are you to know? I know Manu uses lots due the the fast pace of building PC's, but they don't use a mustard bottle to spray it.

Thanks for the input, but it really doesn't say much overall.

The PK has a fully different app. How did you do that app? Make a mess or apply as the manu recommends?

Science, don't see it here. Methodology, documentation meaning pics etc.

 

[jcw122]

I have found the X shape to be the best at spreading from center out decreasing air bubbles.

http://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/

There was another test i saw with the X and four dots, i will see if i can still find it, it was very persuasive.

Keep in mind this is a production environment, it is safer for an assembler to make an x shape and use too much instead of too little. the rest is just going to squeeze out, depending on the viscosity of the paste and the clamping pressure of the heat sink and processor. Squeeze out with this method was enough that it didn't go over the IHS.

I will do another test with the dot method or spreading it if anyone is interested.

Also these servers are going to be used in defense applications with very high ambient conditions. I thought it would interesting to post something other than another 25°C paste test.

I don't understand why you would do anything apart from manufacturer guidelines if it's a mission critical application. Testing is fine, but why not begin the test using the method the maker intends you to use first?

Also, your link is dead.

 

[petteyg359]

Keep in mind this is a production environment, it is safer for an assembler to make an x shape and use too much instead of too little. the rest is just going to squeeze out

There are plenty of TIM pastes used that are electrically conductive, and having a bit of that get in the CPU socket is anything but "safer".

 

[MattNo5ss]

Interesting tests, and thanks for sharing. A 120 °F ambient doesn't seem too practical for the majority of consumers though.

For production environment, it seems like the small dot in the center of the chip would be best.

- Dot is easiest to apply (saves time, therefore saving money)
- Dot uses less TIM (saves money)
- Dot gives almost identical performance as the X and spread methods according to your puget link

So, there is a single negligible downside and more positives to use dot over X or spread in a production environment.

 

[hxtasy]

There are plenty of TIM pastes used that are electrically conductive, and having a bit of that get in the CPU socket is anything but "safer".

This would be an issue. although i have never seen it run past the IHS, except for my test where i purposely put an excessive amount on.

 

[hxtasy]

I don't understand why you would do anything apart from manufacturer guidelines if it's a mission critical application. Testing is fine, but why not begin the test using the method the maker intends you to use first?

Also, your link is dead.

link should work.

i didn't see any user manuals for the PK-1, does anyone have it?

 

[hxtasy]

Interesting tests, and thanks for sharing. A 120 °F ambient doesn't seem too practical for the majority of consumers though.

For production environment, it seems like the small dot in the center of the chip would be best.

- Dot is easiest to apply (saves time, therefore saving money)
- Dot uses less TIM (saves money)
- Dot gives almost identical performance as the X and spread methods according to your puget link

So, there is a single negligible downside and more positives to use dot over X or spread in a production environment.

Very good points, the thing im afraid of is an assembler who is running low on paste, if it is difficult to order more and they are on their last bottle they are going to spread it out and make it work for the X amount order that needs to ship that day. No matter how many rules or procedures you set up they will always get broken in a production environment. Especially at more high volume places when the work is getting paid minimum wage, they really dont care.

it would be easier to setup a standard X procedure, then have inspection check for too much "squeeze out"?

It's definitely something i would like to figure out, process wise.


On this particular setup im also worried about longevity, the dot method was tested in the puget link right after applying, what about after X heat cycles and Y duration.


I understand this is an enthusiast forum and your rigs are your babies, but being super meticulous isn't always a good thing. From what i'm reading the X method seems to be more popular compared to the dot on other hardware forums.

 

[hxtasy]

Too much TIM test:

Processor heat sink clearance dry: 0.051 mm (about two thousandths of an inch)

Processor heat sink clearance excessive, almost sarcastic amount of TIM: 0.051 mm


temperature results are exactly the same as my last coolermaster graph, no need to graph it.

heat sink had shoulder screws and torque does not change clearance. this cannot be said for clamp style or other style heat sinks.

 

[doyll]

Would be interesting to see what the PugetSystems test results would be if each TIM pattern was tested 3-5 times to see if results were the same each time.

Good test hxtasy
Maybe try and 'X' with plastigauge? Would prove what clearance is over entire surface.

 

[hxtasy]

Would be interesting to see what the PugetSystems test results would be if each TIM pattern was tested 3-5 times to see if results were the same each time.

Good test hxtasy
Maybe try and 'X' with plastigauge? Would prove what clearance is over entire surface.

Good point, I definitely feel like the pressure isn't uniform. I've been wanting to do measure this anyways for a custom heat sink I built and am trying to figure out the torque needed on the heat sink screws.

Of course I've also seen/used heat sinks that have a circular shape and don't cover up the whole surface area of the square, with good luck. Although I hate using them.

 

[doyll]

My understanding is the chips make the heat and size of cover is sized to match the PCB area needed for the 1000+ pins connecting CPU to motherboard. Chips are in center of CPU or in a line. I have never seen them reach the edge. The rest of cover is there as a base for cooler to set on. The PugetSystems pea test was not centered so possible the bottom edge of chips wasn't covered by TIM. The line test appears to be running the wrong direction. I think the chip line is up and down not side to side.