Apparently applying thermal paste in X method gives best results?

[buddy_boyo]

EDIT: To be honest, a standard rice grain method is perfectly fine for simplicity sake.

 

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

[TheAmazingCookieFromSquare]

Then everyone shall use the x-shape method from now on!

[awesomes8wc3]

personally, i do the pea method on cpus, x method on larger chips like gpus.

dunno why, but when i first started many moons ago that's what i was told so i stuck with it.

 

inb4 paste method debates & lock

[Mew]

Couple of issues I see in this. As far as I can tell, each test was conducted once- not close to enough to see the variance between each run. 

There are also air bubbles in the X application.

This was only tested on the 3770k; different cpus with different major heat locations would affect the results of this test.

[connorpiper]

I used to use pea, but ever since my computer hit 80c because the pea method failed me, I use 4 dots in a square with a small dot in the middle. So, basically the X method.

[Samfisher]

 

 

Perhaps our most interesting result is that if you are going to use the tried-and-true rice sized dot, you might as well have a bit of fun and draw a little happy face instead. Not only does it cool just as well, but you can rest easy knowing that you have a happy, smiling CPU underneath your heatsink.

 

I'm gonna do this right now.  YOU CAN'T STOP ME MAKING MY CPU HAPPY!

[ThomasD]

Sorry, that limited data set is worthless, absent multiple samples from each group you cannot even being to determine standard deviation, margin of error, or significant difference.

 

Anyone who claims to appreciate science surely must see that the only possible interpretation is that they failed to reject the null hypothesis.  That being:  There is no appreciable difference in temperatures based upon TIM application pattern.

 

And if anyone balks at the idea of doing ten runs with each, then just pick one pattern and do ten runs with it, then see how tight the results are.  Because if you do that, and find a half-degree variation within that group then single run comparisons to other patterns are going to be very problematic.

[Mew]

Sorry, that limited data set is worthless, absent multiple samples from each group you cannot even being to determine standard deviation, margin of error, or significant difference.

 

Anyone who claims to appreciate science surely must see that the only possible interpretation is that they failed to reject the null hypothesis.  That being:  There is no appreciable difference in temperatures based upon TIM application pattern.

 

And if anyone balks at the idea of doing ten runs with each, then just pick one pattern and do ten runs with it, then see how tight the results are.  Because if you do that, and find a half-degree variation within that group then single run comparisons to other patterns are going to be very problematic.

 

It's like flipping a coin only once and determining that because it landed heads once, it'll always land heads. 

[othertomperson]

2 degrees C between best and worst. In other words, it doesn't matter a jot what shape you use as long as you use some.

 

 

Sorry, that limited data set is worthless, absent multiple samples from each group you cannot even being to determine standard deviation, margin of error, or significant difference.

 

Anyone who claims to appreciate science surely must see that the only possible interpretation is that they failed to reject the null hypothesis.  That being:  There is no appreciable difference in temperatures based upon TIM application pattern.

 

And if anyone balks at the idea of doing ten runs with each, then just pick one pattern and do ten runs with it, then see how tight the results are.  Because if you do that, and find a half-degree variation within that group then single run comparisons to other patterns are going to be very problematic.

 

I want to see graphs! With Standard Error bars, dammit!

 

I actually would like to see this done on benchmark runs, tbh. I know it will take a long time but I am very sick of seeing people (especially Jayztwocents) saying "buy this GPU, it is 2 fps faster in this game!".

[typographie]

It's best to take into account the shape of the die. Under the heat spreader, Skylake for example is a rectangle that is taller in the top/bottom directions. So applying a line from top to bottom to reflect that shape can be more effective than a dot in the center.

[Imakuni]

Sorry, that limited data set is worthless, absent multiple samples from each group you cannot even being to determine standard deviation, margin of error, or significant difference.

 

Anyone who claims to appreciate science surely must see that the only possible interpretation is that they failed to reject the null hypothesis.  That being:  There is no appreciable difference in temperatures based upon TIM application pattern.

 

And if anyone balks at the idea of doing ten runs with each, then just pick one pattern and do ten runs with it, then see how tight the results are.  Because if you do that, and find a half-degree variation within that group then single run comparisons to other patterns are going to be very problematic.

Can't we just agree that whatever the application method, CPU is gonna be chilling under 57 at load...

 

..oh. Wait. That's.... weird, to say the least (although "suspicious" would be a better word). I wonder what they consider to be "load"....

[ThomasD]

It's best to take into account the shape of the die. Under the heat spreader, Skylake for example is a rectangle that is taller in the top/bottom directions. So applying a line from top to bottom to reflect that shape can be more effective than a dot in the center

Excellent point.

[Snadzies]

I've always done a small dot in the middle with a thin X shape to the corners just to try to get more even coverage as, like in the article, I found just the dot in the middle often leaves sizable portions with out paste.