How to remove heatsink on Intel X520-DA2?

[TheBloke]

Hey all

Quick question: I'm in the process of making my server much quieter with low-RPM quiet fans. As part of this, I am refreshing the thermal paste on any heatsinks I can. All my components are used and potentially many years old, and I have expensive enthusiast thermal paste, so I figured at least it couldn't hurt.

Hit a roadblock with the Intel X520-DA2 - I'm not sure how to get the heatsink off. Before this I did a couple of LSI cards which have "arrow-head" metal pins that push through the motherboard and then expand outwards.

The Intel card looks like it might be similar, but the heads of the pins are small and plastic, and don't seem to easily compress when I try to do so with some pliers - they squish a bit, but that bulges them outwards, and doesn't seem conducive to them popping through the holes. I'm not putting too much force because I don't want to break them if I'm doing the wrong thing. I also tried turning the pins from the top but this doesn't seem to achieve anything, and doesn't look like it should.

Any advice from anyone who has done this before would be appreciated.

Photos:
Heatsink front:


Heatsink rear, showing the plastic pin heads:

 

[Terry Kennedy]

Hit a roadblock with the Intel X520-DA2 - I'm not sure how to get the heatsink off. Before this I did a couple of LSI cards which have "arrow-head" metal pins that push through the motherboard and then expand outwards.

Looks like center-pin expanding fasteners. On the top side, there should be a way to grab the center part of the pin and pull it out of the rest of the pin. Feel around with your fingernails under the head of the smaller part of the pin. Once you have that pin out, the 4 pieces that spread out on the bottom side should be able to move closer together and you can pull out the rest of the pin.

 

[TheBloke]

Looks like center-pin expanding fasteners. On the top side, there should be a way to grab the center part of the pin and pull it out of the rest of the pin. Feel around with your fingernails under the head of the smaller part of the pin. Once you have that pin out, the 4 pieces that spread out on the bottom side should be able to move closer together and you can pull out the rest of the pin.

Thanks Terry! Yeah, there was a sort of "top-hat" on the top of the pin which turned out to be a separate piece, going through to the centre part of the pin head underneath. I had to pull that hat pretty hard - harder than I was initially comfortable with - until it pulled right out, after which I was able to compress the head and push it through as you described.

Thankfully this NIC used standard thermal paste rather than the awful thick label-like slab I found on my LSI cards, which took literally tens of minutes to fully remove.

 

[i386]

How's the temperature delta? Was it worth the work?

 

[TheBloke]

How's the temperature delta? Was it worth the work?

Wish I could tell you Neither this NIC nor my LSI cards provide any temperature monitoring in their software/firmware. And I don't believe that monitoring the temperature of the heatsink itself (eg with an IR thermometer or a thermocouple) would tell me much, because arguably a higher temperature on the heatsink is better, because it means it's sucking more out of the chip. But then that's also dependent on the airflow's ability to remove a greater amount of heat. Overall I'm not sure if the numbers I could get from the heatsink itself would prove much, so I've not bothered doing that check.

I do still have one standard LSI card in the server, and two X520-DA1 NICs in my workstation on which I haven't yet refreshed the paste, so if I'm wrong and it would be a useful check, I could check heatsink temps before/after doing those.

However in my server I also have a Seagate WarpDrive which is a combination of 6 x SSDs and an LSI 2308 controller. That does have a temperature sensor in its software which I believe, or at least hope, includes the LSI controller (which has a standard LSI heatsink) as well as the SSDs: it provides a temp for each of the SSDs, then an overall temp, and I hope/assume that the latter is the LSI controller.

I've not yet changed its paste, but will be doing so shortly. Right now I am in the process of logging various system temperatures, so once I have some 'before' data I'll do that heatsink and let you know. It's an LSI disk controller not a NIC, but it could at least provide one data point showing that this can be a useful thing to do.

 

[TheBloke]

However in my server I also have a Seagate WarpDrive which is a combination of 6 x SSDs and an LSI 2308 controller. That does have a temperature sensor in its software which I believe, or at least hope, includes the LSI controller

Right now I am in the process of logging various system temperatures, so once I have some 'before' data I'll do that heatsink and let you know. It's an LSI disk controller not a NIC, but it could at least provide one data point showing that this can be a useful thing to do.

Nah, no help. I did change the paste on the LSI controller part of the WarpDrive, but I get no difference in the temperature stats afterwards, and more importantly I'm now pretty sure that those stats don't come from the LSI controller.

Which I suppose should have been predictable - the WarpDrive uses an LSI 2308 which is almost certainly completely identical to a standard card besides the custom firmware. So if that chip/card doesn't have a temp sensor in its normal product, it wouldn't here either.

The "board" temperature shown by the WarpDrive software could be another temp sensor elsewhere. Or it may simply be max( [6 x SSD temps] ), as I've never seen it show a higher value than the hottest individual SSD.

Anyway, as much as it pains me, I will just have to make the thermal paste change without anyway of knowing if it's making any worthwhile difference At the least I am fairly confident it won't hurt.

 

[Jack@UNIXPlus]

Hi Guys,

Here's a "how to" for removing the heatsink:

1st - grab a small flat head and place under the head. Twist the flat head or leverage it up to pop up the pin.


2nd - Once the pins are lifted just remove them.


3rd - Set aside then turn around to look and pin anchors.

4th - You'll need to pinch these two ends together in order to be able to push through the hole.


5th - Once you push it through the hole you its a matter of lifting the heatsink. Thats it!


Let me know if you have any questions.

 

[TheBloke]

How's the temperature delta? Was it worth the work?

The other day I did another heatsink, this time on a chip with a temp sensor, and the results were not good. Specifically, I refreshed the paste on the Intel 5520 I/O hub controller on my LGA1366 motherboard.

It's a motherboard chip rather than PCIe card, however the heatsink design is very similar (large flat, short heatsink with diagonally opposing push pins), and the original paste I found underneath was also the 'sticky pad' type design that I've seen on some PCIe cards.

Unfortunately I saw no drop in temperature at all, and in fact my first test it was showing 1°C higher - though that might be explained by new paste taking a little time to 'settle in'. But over time there's definitely not been any drop, which has to call into question the whole procedure.

However I have noticed an issue with these heatsinks: their push-pins exert very little downward pressure on the heatsink. Thermal conductivity can only be improved by pushing the heatsink as hard as possible onto the chip; my overclocker's CPU heatsink on my desktop CPU exerts a massive amount of downard force with huge screws and springs.

When playing about with the 5520 chip, I saw that it was possible to push the pin down far more than it normally sits. Because the pin heads are relatively shallow, the pins pull back a way before gripping onto the motherboard, such that the pins make little use of their springs.

This made me think that if I could insert something small and hard under the pin heads - a piece of plastic perhaps - I could get the pins to exert more force.

I may try this on the 5520 to see if I can make better use of my (theoretically) better thermal paste. It may be that this 'enthusiast' paste is only better in high-force installations, which is where it is expected to be used, eg for CPU and GPU. Operations on this chip require unscrewing and removing the motherboard, which is a bit of a pain, but it might be interesting to try and prove/disprove the possibility of temperature improvement.

I'm interested in lower temps in this 5520 not for its own longevity - its listed max temp is 90°C+ and mine rarely exceeds 45°C - but rather because, annoyingly, my motherboard uses its temperature as a decision point in raising the speed of all the case fans. If it exceeds 40°C it puts the fans up to 80% PWM even when everything else is perfectly cool, and above 45°C it goes to 100%. This can significantly increase the noise level.

So I'm trying to see if I can get the chip to below 40°C most of the time, as part of the process of quieting the server. I only need to drop another couple of degrees to achieve that during the majority of server operations. An alternative would be fitting a larger and better heatsink, but I've not been able to find any that I could attach using the same holes in the motherboard. Or I could simply stop using the motherboard's PWM control completely, but then I'd need to find some external solution which I can script to respond to fluctuating temperatures, and that looks like even more work.

 

[alex_stief]

Thought of using a thin copper plate with thermal paste on both sides instead of just thermal paste to close the gap? This works wonders on some older laptops with crappy cooler design.

 

[TheBloke]

Thought of using a thin copper plate with thermal paste on both sides instead of just thermal paste to close the gap? This works wonders on some older laptops with crappy cooler design.

Yeah that's a really good idea! That'd be a much cleaner way of increasing the force than trying to jam a thick piece of plastic under the pins.

And as luck would have it I already have two thicknesses of copper strips. A long, flexible roll of 0.3mm, and a stiff length of 0.7mm:



The 0.7mm would certainly go some way to closing the gap. And I could potentially do that as well as a small wedge under the pins - it'd be much easier inserting a thin piece of plastic under each pin head than it would a piece thick enough to make a big difference.

Thanks Alex, I think that's definitely worth trying.

 

[TheBloke]

Thought of using a thin copper plate with thermal paste on both sides instead of just thermal paste to close the gap? This works wonders on some older laptops with crappy cooler design.

Although it's now a bit OT for this forum, I wanted to update and say thanks again @alex_stief for this great suggestion.

I went ahead and did it on my motherboard's 5520 PCI controller chip, and it made a small but very useful difference.

I did it in two phases:

• First I inserted a single piece of 0.7mm copper between heatsink and chip, with a thin layer of paste on both sides.
  • Compared to no copper, this resulted in approximately 2°C of temp drop in all usage.
• Two or three days later I re-did it, this time with two pieces of 0.7mm copper, with three layers of paste (between heatsink & copper 1; between copper 1 & copper 2; between copper 2 and chip).
  • This didn't give any extra benefit in an idle test, but in a loaded test the max temperature was 1-2°C lower than when using a single piece of copper.

These are objectively very small differences, but given my unusual requirement of trying to keep the chip below 40°C, they proved just enough to make a useful difference. I still have to have a 60mm fan sitting right on top of the heatsink, but it's an enthusiast quiet fan and I can't really notice any extra noise from that one fan.

The resting temperature of the chip is now 37°-38°C when idle. It does go up to 41-42° during sustained periods of load (eg compiling GCC 7.2 with 24x parallel compilations, which takes about 30 minutes), but it will drop back below 40°C shortly after, and eventually drops low enough to trigger the PWM fans to drop down to a lower speed.

To try and bring things a little back on-topic: I don't know if this is worth doing on any PCIe card - it's very hard to tell without having a temperature sensor on any of them. But if one knew a given card was dropping out due to overheating, or if it was a RAID controller or other card that does have a sensor on it, which showed too-high temperatures, then this method certainly seems like it could help. The cards I've looked at all use the same method of heatsink attachment, using pins which don't apply a great deal of force, and it seems that can definitely be improved by inserting a layer of copper to increase the force.

It didn't make a huge difference for this chip, but it probably would make much more of a difference at the higher end of the heat range, eg if the chip was running at 95°C.

Finally I can report that my proposed method of increasing heatsink attachment force - sticking something under the pin heads to extend them - didn't work at all, at least on this 5520 chip. The pin heads were too small for me to be able to get anything to stay wedged underneath them without it just pinging out under the force from the springs. It might have a better chance of working on a heatsink that uses metal pins, like LSI controllers do. But it's always going to be hard because the pin heads are only slightly bigger than the hole, so there's very little excess area to hold anything down.

I suppose on a PCIe card there might be other options available, like replacing the pins with bolts and nuts, or perhaps some way to pull out the pins and clamp them in that pulled-out position. But I couldn't really try that on a motherboard chip, as there's very little room under the motherboard.

 

[alex_stief]

It's not much, but it's something...I got the idea from modifications to Thinkpad X201 laptops where such a copper plate could make the difference between thermal throttling and a more comfortable 85°C on the chip.
So the optimal solution seems to be a thicker copper plate, eliminating additional thermal resistance of one layer of thermal paste.
Could it be that the standoffs at the bottom of the heatsink prevent it from being pushed harder against the chip with your initial approach? Or do they fit into holes on the PCB in order to define its lateral position?