Heatsink Mounting

Some factors affecting performance – Bll Adams

Question: “What is the mounting procedure to ensure a consistent contact with the die?”

Having fought the TIM joint battle for over a year now, I do have some thoughts on the subject; and while the literal answer to this question would be limited to the mechanical aspects, the interrelationship between the variables is a bit more complex.

A TIM (Thermal Interface Material) joint is an element of every heatsink or waterblock installation, and is also the most variable part of any CPU cooling system. Not always appreciated is that the TIM joint variation can easily exceed the difference between different waterblocks, so its control is essential for comparative testing.

The joint consists of the CPU substrate (surface), the thermal grease or PCM (Phase Change Material) tape, and the heatsink or waterblock substrate;
and obviously the clamping mechanism by which the joint is held together.

The CPU substrate is normally used as-is, since any DIY ‘treatment’ will result in a less flat surface.

Where an IHS (Integral Heat Spreader) is used, some have ‘announced’ that theirs was concave – and that they sanded it flat – without apparently understanding that IHSs are thin and designed to deflect to a flat position under the recommended clamping force (but who ever heard of a DIYer that did not know more than Intel?).

The heatsink and waterblock substrates are a whole ‘nother kettle of fish; however, their inadequacies are well known. Lapping is indeed worthwhile, but it takes the right materials and good technique (my present preference is 15 micron diamond powder on a new Class B granite surface plate, followed by 3 micron on a different plate – but I’m quite into all this, eh).

Some general observations on waterblocks:

ONLY one waterblock manufacturer ships a FLAT waterblock – Swiftech. ALL the rest (that I have seen – many) are crappy. And while the Swiftechs are flat, I have found their finish less than optimum (but if one presumes to improve it, VERY good technique is needed – for most best left as is).

The remaining joint component is the grease or PCM (frag) tape,
of which the tape is the far better solution in terms of consistency and long term reliability. The proper application of grease is VERY technique sensitive, and is not the same for all greases.

Without getting into a grease superiority wrangle, I would observe that after very extensive comparative testing, I have concluded that Cooling Flow is superior for my testing purposes due to the reduced variability of the assembled thermal impedance of the TIM joint itself (note that my test setup enables the measurement of such, to 0.01°C).

So, finally, returning to the initial question: How to ensure consistent contact?

Different manufacturers have different solutions to this problem:
My preference for the most elegant is the German design with the springs integrated into the 3-lug retaining clip itself, or the LiquidCC ‘bullet’ spring compression limiters. As I use the 4 mobo mounting holes with springs as a ‘universal’ mounting method, two aspects must be addressed:

• The springs’ compression rate , and
• The actual compression as installed.

A Load vs. Deflection curve is made for each spring, and 4 selected that are nominally equal. I have found the Swiftech springs to be quite consistent, and a typical graph is shown:

Now it is a straightforward exercise to select the amount of compression desired to achieve a specific compressive load across the TIM joint. The springs are tightened incrementally and the final compression is set, and measured, to 0.001 inch – each spring.

N.B. the weight of filled hoses and any other externally applied loads must be identified, and quantified, to control the total clamping force.
My solution to this dilemma is to suspend everything and to add the weight of the die assembly and instrumentation cables to the load applied by the springs – takes much ‘tweaking’ to sort it out.

And that was the good news – the bad news is that some joint induced variability will still remain despite all efforts. Learn how to install a particular waterblock, 5 trials or more, then do it 10 times to understand the range and standard deviation; if OK, make the run – I run 3 times for real tests.

Note that a ‘flyer’ that is worse can be rationalized, a ‘flyer’ that is better, however, indicates that the setup has not been optimized – START ALL OVER!

Hope this provides some insights as to ways to reduce the assembly variability. It ain’t easy.

Be cool.

Bill Adams