Importance of reapplying thermal paste

[Doug Hall]

I installed my CPU fan 90 degrees off, from where it should have been. (I thought I would have enough fan cable to be able to keep it wrapped around the fan.) So, I needed to remove it and try again. I did so without too much trouble. I haven't yet booted it, but I read in the motherboard manual that it's important to clean the former thermal paste off, and reapply, before going further. Is this THAT important? Will it form an air pocket, or something like that, which is important to avoid? I don't have thermal paste on hand.

 

[survive]

Hi Doug,

A bit out-of-scope for FreeNAS but let me tell you what I think....

I'm going to assume you have a new Intel CPU that you are using since you can rotate it 90 degrees. Personally I wouldn't worry about it...the thermal paste melts\softens when you turn the system on and sort of "squshes" out to the edges as the processor heats up. The cooler clamps onto the cpu with a surprising amount of force so there shouldn't be any air bubbles to worry about. I think the warning in the manual applies more to paste that has been "cured" by the system heating it up and melting it so it can be spread out by the pressure of the cooler.

That said, the cpu will shutdown if the processor gets to hot. First thing I would do is boot the box & get into the BIOS & see what the processor temperature is. Let it sit there for a while (10 min?) and see if it rises (for reference my proc in my desktop is reporting ~100F right now). Next thing would be to grab a copy of memtest+ and let that run on the box for a couple of passes to verify your RAM is sound, doing so will heat the proc up....if the system completes a few passes of memtest then I would think you are going to be fine.

Just be sure all 4 of the cooler attachment pins are down flush against the board...the 4th pin should be a pain to get to click into place.

-Will

 

[cyberjock]

You can also check your CPU temperature(if the motherboard supports it) with the command sysctl -a |egrep -E "cpu\.[0-9]+\.temp".

If you really want to make your CPU get hot, do a zpool scrub or something. Let the scrub run for about 10 minutes and check the CPU temp. Generally, I don't worry about any CPU temp below 160F when fully loaded. And most CPUs won't turn off until around 180F, and over 200F or so is where you hit the dangerous territory.

I wouldn't be too worried about what you did. Theoretically it may matter. But in reality it really doesn't matter that much. A friend put a heatsink on a P4 CPU without any thermal paste and got great temperatures anyways, for years. Go figure!

 

[DoulosGeek]

A tube of thermal compound is a pretty inexpensive investment in protecting a system AND keeping it running at peak efficiency.

I would absolutely remove any traces of the old compound with denatured alcohol and re-apply new compound. Look up some how-to's on YouTube. There are many opinions on how to apply the compound, how much to apply, how long it takes to cure, etc.

I like the thin line method. It's easy enough to see which one works best. Apply, lock down the heatsink and then remove the heatsink and see if you achieved a good spread and good coverage on the CPU and the heatsink interface. Remove the compound and re-apply, lock down the CPU and the heatsink.

NOTE: I've never seen old thermal compound "squish" out. It's standard operating procedure that if I break the heatsink/cpu interface to remove the old and apply new. The old compound is always "cured" (mostly hardened in place), which is good for heat transfer if the interface is not disturbed...bad for re-use when the interface is disturbed.

 

[Doug Hall]

Thanks for the advice guys. I only booted it momentarily (less than two minutes) after the first time I installed the CPU cooler. I really doubt that it had much time to melt the existing compound very much. But, I will keep my eye on the CPU temperature. If it rises above 16o degrees fahrenheit, or even comes close to that on a regular basis, I'll buy some good thermal paste and reapply.

Doug

 

[jgreco]

The heat from a CPU can very rapidly soften the phase change pad that often comes preapplied to a heat sink. It can resolidify in seconds.

Fortunately we are no longer in the era of flip chip or organic PGA packages where having absolute precision was important or you'd burn the processor (I still have some nice samples of burnt processors).

Unfortunately we now have greater TDP's.

If you remove a CPU with a phase change thermal pad and you don't see the imprint of the CPU in the pad, then either the pad is too small or the thermal material is poor (most likely cause). You don't really want a fat layer of thermal material between your processor and your heatsink. In the case of some craptacular quality pads, this does result in what might be referred to as a "squish out" around the edges.

If you manually apply paste using some of the directions provided by manufacturers, you may also see "squish out" style behaviours. I'm not a fan of the "vertical line" or other techniques that rely on the meeting of the CPU and the heatsink to cause the spread of compound. We've always had excellent luck spreading a very thin layer.

Arctic Silver has some nice discussions of the topic. This is essentially what we do here.

They also have some documents on other techniques, plus their reasoning for why those might be better for a given processor.

Pay particular attention to the discussion of the importance of cleaning.

 

[PenalunWil]

As a total noob can I ask what sort of CPU coolers you guys are using and which would be suitable for the X10SL7-F-B Mobo and Intel CPU Xeon E3-1230 v3 Quad Core Haswell Processor running 7HDD in a zraid3. I've been looking at coolers from the size of an ice hockey puck to a Mars orbiter, and I'm soooo confused :D Obviously I don't want over kill, but also I don't want to put in a stock cooler if its not going to cut the ice so to speak.

 

[Yatti420]

You resurrected an old thread.. I'm assuming that chip doesn't come with an intel cooler? If it does use it.. So purchase whatever supports that socket.. I doubt the CPU is going to get that hot unless you transcode / work it hard.. If your looking for silence you will have to pay for it.. Good luck..

 

[jgreco]

If you buy a boxed part that comes with a cooler, it will be sufficient unless you maybe wrap it in tape or find other creative ways to deny it moderate airflow.

For our Sandy Bridge ESXi nodes in 4U, which closely resemble hardware on the FreeNAS HCL, I wanted a more aggressive solution to reduce case temps and to be able to tolerate failure of the fan; I used a Xigmatek Loki 92MM with the fan mounted on the far side so that the chassis fans were blowing at the front of the cooler and the CPU fan was pulling air through. The heatsink has a TDP of 130W and the E3-1230, only 80W. You can probably see how that wouldn't be a catastrophe if the CPU fan failed.

A lot of coolers have emphasis on the "cool" (as in "gee, cool!") aspect - not the "cold" aspect. You can skip by most of those, some of them work great, some don't, all of them are too expensive. Figure out your design requirements and then pick the biggest functional cooler that would be reasonable. Lots of surface space (fins) and good airflow are what you're looking for. You want the heat to be quickly migrated away from the CPU. The Loki is a gorgeous example of simple design along those lines, but it is not going to fit in a lot of places :)

 

[PenalunWil]

My thinking is to go with the laws of nature... hot air rises so I'm going to pull cold air in at the bottom of the mATX case and suck/blow it out of the top. The CPU fan like the one jgreco has specified will be mounted and blowing in the direction of the air flow. I'm hoping to use 2 x Icy Box IB-544SSK 4 Bay units placed low down in the case and I'll pull the air into the case through these with help from there in built fans. I'm going to position an extract fan or two at the top of the case and block off all other air entry points into the case. The air flow from the CPU cooler will be in the direction of the main air flow. There should be a steady flow of air in and out of the case... cold in near the base of the case and hot out at the top. My case is an Aerocool Vs-9 Advance Windowed Midi Tower Case. I don't want a positive air pressure inside the case, or any devices that will inhibit the airflow through the case. I thought that a Hyper 212 EVO sitting on the CPU and blowing vertically in the direction of the airflow would be a good solution. Any comments would be very much appreciated... Thanks for your comments above guys.

 

[Starpulkka]

I have seen that not properly cleaned cpu heat spreader surface and not properly cleaned cpu cooler surface, cpu gets more than normally hotter on a year later, not sure what heat transfer compound have happened on a cpu heat spreader surface, but cpu might start throttling if its get too hot, so it is not good. I have invest in arcticlean 1 and arcticlean 2 thermal material remover and did get really good results. As for thermal compound install i have two methods what im using in my station. One is that pea method on small die, and a line method on a long die cpu (as some server cpus have long die). Also on a long die cpu its good to know line direction. Too much compound is bad and too little compound is also bad, also if you try spread compound on your black centurion card on a heat spreader i would be laughing loud.

Youtube

 

[cyberjock]

And you want to know what Arctic Silver's website has recommended for as long as I can remember?

The pea method(sometimes the line method depending on the processor you choose).

And people wonder why I tell people not to do stupid crap and justify it with "I know better than the f'in engineers that invented this shit!" and laugh at them.. to their face.

 

[Starpulkka]

Yes i know what method artic recommends, i used to do it too. (card for a cpu cooler's surface)

 

[cyberjock]

Here's how my friends do it...

View: http://www.youtube.com/watch?v=TnyoJtv9Cx0

 

[Starpulkka]

Here's how my friends do it...

View: http://www.youtube.com/watch?v=TnyoJtv9Cx0

Hmm.. strange way but i believe you. ^_^

 

[DJ9]

Heh. Here: View: https://www.youtube.com/watch?v=-hNgFNH7zhQ

 

[jgreco]

OMFG! Awesome (Kentucky Way).

Okay at one point I had a customer who was interested in a hardware design done here in shop, but the CEO "knew a company" (drinking buddies?) and had them do the hardware instead. This was 2006-2007 era IIRC, and it was Opteron stuff. These things started dropping like flies, and the guy tasked with operating them pulled one apart to show me one day. The heat sink grease had been just totally glopped on, had smooshed out from between the CPU and heatsink, and formed a little halo around the CPU socket on the serverboard. It was still real thick in between the CPU and heatsink, which of course is a problem for heat transfer.

I wish I had had a camera at the time, but ... secure facility and all that.

Now I know there's a lot of debate about the best way to apply heat sink grease. Some observations:

1) Applying it very wrong is bad.

2) If you're overclocking, you want the best possible application. But no one here is doing that.

3) The general consensus is that the pea method is mostly OK except when you're supposed to use the vertical line, which is often these days.

4) But the process that works for every CPU is actually what Arctic Silver calls the surface spread. This is hard to do right until you've done it a bunch of times. It has the added benefit of additional surface contact between the CPU and heatsink, but also has the potential for air bubbles especially if you are sloppy. That's part of the reason AS doesn't suggest it for most cases ... but in a shop environment it is preferable because there's less opportunity for someone to use the "wrong method."

5) As long as you've made a good attempt to do a good job, don't have a heatsink that's too small, and aren't doing dumb things, the specific application technique used probably isn't going to impact you. Pea, vertical line, surface spread... follow the directions for ANY of them and you're probably going to be just fine. This really only comes into play for the people doing overclocking or other silly things.

 

[PenalunWil]

Thanks guys for the entertainment and advice. My missus was wondering how the hell a FreeNAS forum could make me laugh... she didn't get it. Think I may be turning a bit geeky. :D

 

[jgreco]

Glad we entertained you.

 

[D4nthr4x]

Having spent a large amount of time on another forum testing thermal paste application, cleaning, and surface prep; I have a fairly large amount of experience with this. First the best thing to do is to use a coffee filter to clean off the old paste, for some reason we were seeing consistent 1-2C drops on temps doing this I think it cleans the surface better than other clothes and doesn't leave any fibers behind (or at least that is my theory), also for application it really depends on what thermal paste you buy. If it is very viscous it is recommended to heat up the tube by putting it in a cup of hot water to make it easier to apply/spread, and then generally you just want to put some in the middle spreading it out across the chip was VERY inconsistent among the people testing and if you are using newer after-market coolers buy one that lets you tighten it to a backplate and you will have no problem with spread. However, if you have some of the direct contact heatpipe coolers you need to put some extra thermal paste on and we generally would recommend putting a small line near each joint between the pipes, now on the new direct contact coolers this isn't as big of a deal but on the first/second gen ones they had large gaps there. But if you want the best temps on Haswell/Ivy CPUs then you have to de-lid the CPUs and apply a liquid metal thermal paste instead of the stock paste Intel uses since they switched away from soldered on heat spreaders. But there is always the risk of killing your CPU and your warranty is also void.

As for the best paste to use IMO it's Arctic Cooling MX-4 I get really consistent results with it, and it isn't too think. Most of the diamond thermal pastes are terrible due to their viscosity but you get a nice polish on your CPU that rubs the writing off and makes it shiny (voiding the warranty since you can no longer identify it), and things like Tuniq TX-3/4 is also fairly thick but it works ok if you heat it up first. Also I think there is either a bad or fake batch of Noctua pk-3 being sold right now because the thermal paste bundled with their heatsinks was A LOT better than the stuff that I bought separate.

Also if you go liquid cooling make sure you mount your radiators with the tubing coming out of the bottom on the pre-filled units. We were seeing 6-10C drops by flipping it around because there was a good bit of air at the top of the rad causing a problem.