How To : Change the heat sink TIM on an LSI SAS card

8 posts in this topic Last Reply

Recommended Posts

GUIDE : How to remove the heat sink and change the thermal interface material (TIM) on an LSI SAS card




TOOLS NEEDED : Small pliers for push pin removal, small flat screwdriver for heat sink removal, plastic or wooden scraper to remove old TIM, Isopropanol or Arcticlean for grease removal


PARTS NEEDED : Replacement TIM (any brand), full height mounting bracket if case requires it, replacement nylon push pins if broken (see below for spec)


IDLE TEMPS : Measured in the centre of the heat sink with an IR Thermometer in an open case after allowing 10 minutes for the card to warm up. Original TIM 56C. After TIM change 50C.


As you will be aware, many of the LSI SAS boards run fairly hot in use and are even rather warm when idling. Since the main source of older cards such as the SAS9201-8i I used in this guide are server pulls or New Old Stock supplies, I thought it might be possible the old TIM would have dried and might benefit from replacement. I decided to examine how the heat sink was fitted with the intention of removing it, cleaning everything and replacing the TIM. This would allow me the option of fitting a larger heat sink if required or even preparing the card for water cooling.


The first think I did was record the card in its initial state as the heat sink is not symmetrical...



The technical specification of this particular heat sink is 40mm x 35mm x 9mm which fits in a single PCIe slot width. The mounting holes are on opposite corners of a 33mm square which gives a diagonal spacing of 46.7mm.


Looking at the back of the board, you can see the backs of the two nylon push pins.



They are pressed through the board and have two backwards pointing barbs (like an arrow head) which grip and hold the pin in place. Remembering the issues that the 2009/2010 Mac Pro towers had with the nylon push pins on their Northbridge heat sinks going brittle, I took care to be gentle with the pins when removing them.


Using a small pair of flat nosed pliers I gently squeezed the barbs together until they were pressed to about the same diameter as the hole in the board. Take great care to avoid the small surface mounted components with the pliers which you will see are very close to the push pins.



A little gentle finger pressure on the pins popped them through. They will be retained by the heat sink so ignore them now until you have the heat sink off later.

Now comes the tricky step. If you look under the edge of the heat sink you will notice that while you could easily push a screwdriver between the heat sink and the circuit board, this would apply upwards pressure on the solder joints of the chip on the board. At the time I thought the old TIM would release readily but this was not the case. Fortunately we do not need to take this risk to get the heat sink off.


Look closer at the heat sink where it joins the chip itself. You will see when you look from the side that the chip package has a centre flat portion which is flush with the heat sink and a thinner lip that connects to the circuit board. It is designed as a sort of flat "hat". Taking care to avoid small surface mounted components, place the tip of a small, long screwdriver (or small metal spatula) between the heat sink and the "brim" of the chip package and gently twist to separate the two. You want to apply the tip to the lip between the heat sink and the chip, not the chip and the board. It should give gently rather than suddenly.


One gentle twist later and the heat sink is off to reveal (in my case) a large amount of yellow, waxy TIM.




It is plastic in texture, somewhat brittle and very hard. It turns out that rather than use normal oil-based thermal compound, LSI used Thermal Epoxy which dries to this hard plastic material. It is completely immune to the solvent effects of Arcticlean which will readily dissolve normal greasy TIM so I had to use a wooden satay stick which I carved to a flat tip to scrape it off.



The heat sink had excess epoxy so I confess I used a small screwdriver to scrape the worst of it off. This is where I scratched the surface a little. Even with Arcticlean and Magic Eraser melamine block I still couldn't get it clean.



I took the opportunity to remove the push pins from the heat sink while I had it loose. They are retained by the heat sink and springs so carefully push them through the holes in the heat sink and set them aside with their springs. Once you have the two pins out, very gently tease the barbs back open a little with a small flat headed screwdriver. They don't need to be opened out much... just enough to grip the hole again when re-inserted.


If you do break one, the length of the main pillar is 11mm and they are made for a 3mm diameter hole.



That 11mm length is the amount of the push pin that stays on the "outside" of the circuit board in the springs. It does not include the extra length of the barbs.


Now is the point where you clean the top of the chip and the bottom of the heat sink with Isopropanol or Arcticlean2 to remove any remaining traces of old TIM or grease from your hands. I used a small dab of CoolerMaster ICE01 TIM which is what I had around and used a spreader to put a thin, smooth layer over the whole of the flat top of the chip. Use it sparingly but any excess will just sit in the "brim" of the chip cover.


Push the two push pins and their springs back through the heat sink if you removed them and line it up on the board. Refer to a picture of its original placement if the heat sink is not symmetrical. Gently push the two pins through while holding the heat sink flat and in place. The two pins should just go back through and grip lightly. If they refuse to grip, check the barbs are flared just enough to grip the back of the board.


I took the opportunity to change the mounting bracket from the standard low profile one for server cases to a full height one for desktop cases. These are commonly sold on eBay as suitable for the IBM M1015 cards but they fit this one fine.



Once the card is reassembled, put it back in your machine and test it. All drives on all SAS ports should be recognised and the Power LED on the card should be steady. I brought my unRAID server up and ran a parity check and all was well.


If you intend to replace the heat sink with a larger one, have a look at Northbridge or Southbridge heat sinks that will be taller than a single PCIe slot. Alternatively you could investigate water blocks which some online PC stores may have in stock. Check the diagonal spacing of the mounting holes and the clearance next to the holes as the Front Panel LED jumper on the SAS9201-8i is quite close to one of the push pins.

Link to post
  • 4 months later...

Thanks very much @DanielCoffey for posting this great guide.  I found it via Google when planning to do the same task myself on my various LSI controllers.


It was very helpful to see in advance what I would experience, particularly as this was the first time I'd replaced a heatsink on a PCIe card, so I'd never before seen or dealt with the 'arrow-head' pins.  And it was very useful to know that the heatsink wouldn't simply come away from the chip and needed prising.   These sticky pads that LSI use are a bit of a pain!


On 03/09/2017 at 11:33 AM, DanielCoffey said:

IDLE TEMPS : Measured in the centre of the heat sink with an IR Thermometer in an open case after allowing 10 minutes for the card to warm up. Original TIM 56C. After TIM change 50C.



I also wanted to comment on this.  While I absolutely agree that it's worthwhile to refresh old paste, I'm not sure if this method of testing the before/after is accurate or useful.  And I'm worried that the figures you posted could actually mean the opposite of what you hoped.


Here's my thought process.  The temperature on the heatsink is a factor of two things:  1) the amount of heat transferred from the chip through the paste;  2) the amount of heat dissipating from the heatsink into the air.  By changing the thermal paste we affected 1, but we did not affect 2.   By improving the thermal conductivity, arguably we are more likely to see higher temperatures on the heat sink.  Better paste means more heat sucked out of the chip, and as we haven't improved the airflow/cooling, that should result in the same or even a higher temperature on the heatsink.   Which is OK, it's the chip temperature we want low.


The dissipation from the heatsink depends on the size/structure of the heatsink and the amount of airflow over it, which together give us a heat dissipation rate (degrees C per second or whatever. ) This dissipation rate is then also affected by temperature - hot objects dissipate heat faster then colder objects - but I don't think this effect would have a material difference in this case (I think it balances out because the chip itself is also cooler).)  Especially with only a few degrees of difference.


Therefore I'm pretty sure that we can consider the dissipation rate constant before and after the thermal paste change.  The only thing that changes is the greater rate of heat transfer from the chip through the paste.  Therefore if anything we might expect a rise in heatsink temperature.  The chip itself will experience a drop - our real goal - but as we can't measure that, we wouldn't see it.


So I'm worried that your measurement of a lower heatsink temp could actually mean that the new thermal paste is performing worse:  you didn't change the airflow, and therefore you didn't affect the heatsink's heat dissipation rate to any significant degree, so a lower temp on the heatsink seems to imply less heat being transferred from the chip and therefore a higher temperature in the chip.  Which could only be caused by lower thermal conductivity between chip and heatsink.


I'm quite possibly missing something here, but that's how it seemed to me.  I'm certainly not a physicist so do let me know if there's any flaws in my logic, assumption or facts!  In fact I debated whether to post this, as it was a great guide and I didn't want to rain on your parade by potentially indicating it has had the opposite effect.  But I'm still hoping I might be wrong and that someone will correct me.   


Even if I'm not and you have had a bad result, it gives the opportunity to try again;  the method certainly should work and is definitely a good idea, especially using enthusiast quality thermal paste that should outperform any standard stuff.  So it could be that something just went wrong on your attempt.  These heatsinks cam be a bugger to get back on, mine took me two attempts because the first time I couldn't get the heatsink back on properly and I was smearing paste all over the place.  So I stopped, re-cleaned, and did it cleaner the second time.


Needless to say that because of the above thoughts, I didn't bother testing the before/after temp of my heatsinks when I did my thermal paste refreshes.  In the absence of a chip temperature sensor, and with the worry that it can only show a worsening of heat transfer, I'm just going to assume it's worked OK :)


Thanks again for the great guide.

Edited by TheBloke
removed possibly incorrect part; restructured; shorter.
Link to post
  • 1 year later...
1 hour ago, truckerCLOCK said:

Great idea but I took a simplier route. I just purchased 2 mini 24 vdc fans and glued them to the heatsinks.

The problem this thread is addressing isn't keeping the heatsinks cool, that's a separate issue. The problem this thread is talking about is the lack of physical connection between the heatsink and the chip, when the compound dries out, the heatsink no longer absorbs heat from the chip, and cooling the heatsink more isn't going to help keep the chip cool.

Link to post
16 hours ago, truckerCLOCK said:

So cooling the heatsink more isn't going to help keep the chip cool.....hmmm. 

Depends on the quality of the thermal connection between the heatsink and the chip. This thread discussed how to fix a situation where the heatsink isn't well connected because the compound has dried out or become dislodged or is otherwise not performing the job of transferring heat from the chip to the heatsink.


If the heat doesn't get transferred into the heatsink in the first place, cooling the heatsink more won't do anything.

Link to post

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.