Increasing pressure on waterblock to lower temps

[murray13]

Actually #Rotor there is as much change in 'Contact Thermal Resistance' (as per your graph) from 25N to 100N as there is from 25N to about 9N and the same change again from 9N to 3N of mounting force.

As to what this says? There is a point of diminishing return. But saying outright that more pressure does not give better results is not completely true. There is a point at which it would be very difficult to measure any difference while increasing pressure to the limits of what we can apply to the core.

This disscussion is all based on the properties of the TIM. Grease type thermal interface materials are very poor conductors of heat. So the thinner you can get the TIM the better it will perform.

Some of the newer phase change TIM's out there (like the one I mentioned in a diff post from Thermagon, T-lma) have very different properties. Their mounting requirements are very different as the material itself is different.

[bigben2k]

Originally posted by #Rotor
I have some old EDO ram here, if you need some ya know... to get them memory banks up to scratch again.......[jk]

3.3 or 5v?

[Al Kaseltzer]

Originally posted by #Rotor
This perceived gain is as a result of the diminished length of the thermal compound barrier( or layer), between the two surfaces.This length-reduction as result of added pressure, is however finite, once the two surfaces are touching (remember, they are not going to deform each other) the compound will not be able to get squeezed any more, right. That is where the optimal seat is achieved, more downforce from this point on, will not help, as there is nothing more to give way.

a graph.... I hope it makes sence.

(graph removed, see previous page in thread)

note that anything from about 25N and more, gains are virtually flat-lining.

You seem to be ignoring the other line in the graph - In dry conditions (no thermal interface material) the thermal resistance is still reduced with pressure. The surfaces are obviously touching, so SOMETHING is changing with pressure.

The fact is that there is no such thing as a smooth surface. As you measure on a smaller and smaller scale, no matter how much you've lapped and smoothed, there will be imperfections. These peaks and valleys on both the mating surfaces will be deformed by the contact pressure, squeezing the air (or TIM) out and increasing the contact area.

There are a couple different limits to this decrease in thermal resistance - the most likely we will see is the difference will quickly become smaller than any of our abilities to measure it. Also, with high enough pressure all voids will be removed down to a molecular scale, there will still be a thermal resistance because of the two different materials. If it was 2 pieces of copper instead, this type of pressure would be enough to cold weld them together. However, before the pressure gets that high, it's likely that one of the materials will fail as it goes beyond it's maximum compressive strength.

[#Rotor]

aaaa indeed. and once we are in that scope of where we can not perceive the difference for a given action. I like to think of it as not making any difference at all. Sure theoretically there is still gains, but the more likely result would be catastrophe. The KISS philosophy, very handy when brain matter is to be wrapped around these higher things.

reason I say that the surfaces are not deformed by one another, comes from the fact that by the time enough pressure is applied, to deform the copper significantly enough to produce a measurable difference, your CPU would already be in a better place.

as for ignoring the dry joint scenario, hhmmm yip, Guilty as charged. But for a very good reason....

[edit]

murray13

Yes you are correct....
BUT look at how much more pressure one needs to add, to get from 25N to 100N in order to get allmost the same deviation, acquired from the previous 16N added...... and somewhere in that 75N there is a "SNAP_CRACKLE_POP" that's watching you.
Also note my devious way of using the word " virtually"... as in "virtually flat-lining."

[/edit]

[Cathar]

Finally got back to this thread, but yeah, Al Kaseltzer and murrary13 are raising the same points I would have raised.

Surface area is also important. The pressure applied really needs to be defined in terms of force per area, rather than simply force.

Increased pressure will result in better thermal transfer performance, both with and without pastes (but maybe not for all pastes). The point of diminishing returns seems to occur at around 100PSI, but gains are still seen beyond this point.

However, and completely tangential to this pressure vs thermal interface thing, is the mount pressure vs tubing torque.

Most water-cooling setups have the tubing simply hanging off the CPU block in a completely unsupported fashion. When the case is sitting still this applies a fixed rotational torque that tries to lever the block away from the CPU leading to uneven clamping pressures against the CPU die.

Now 8kgs of pressure against the CPU, or really about 2kgs of pressure for a single spring located at around 3cm distance from the CPU will be greatly affected (relatively speaking) by 0.2kg of hosing torque tilting with 10-15cm of leverage. ie. the hosing torque being applied equates to around half of the spring torque being used to hold the block in place at any one corner/spring.

So this leads to a secondary reason for why increased mounting pressure is beneficial. By raising the pressure applied to each spring to around 10kgs each the relative effect of hosing torque acting against any one spring is diminished, typically leading to a more even pressure being applied to the CPU.

[Al Kaseltzer]

Originally posted by Cathar
The point of diminishing returns seems to occur at around 100PSI, but gains are still seen beyond this point.

Don't know why it took me this long, but Cathar putting a number on it made me realize something:

The correct mounting pressure is dependant on the Thermal Interface Material.

In other words, The AMD spec isn't just for the AMD core, but for the AMD thermal pad included with the retail Heat Sinks. Same for Intel.

For example, look at this graph:


(Image source from Googling)

Note that this is a graph of Thermal Conductance vs Pressure, the prior graph was of Thermal Resistance.

Notice the 2 best compounds at the top right of the graph. The Light blue compound is starting to level off as pressure increases, but the Dark blue compound below it is still improving rapidly with more pressure.

I would suspect that most TIMs marketed for use on CPUs would be designed to work with standard CPU clips, but for anyone trying more exotic TIMs, this might be an important point to consider.

[iggiebee]

Perhaps, also of interest for WB testing...

At the following Test and Research equipment manufacturer web site http://www.longwin.com/, and among the various CPU testing devices they produce, found a testing device named "LW-9052 Press Load Apparatus":

Edit: darn!! how can you insert and image on this thing?

Oh well
Here is a link: http://www.longwin.com/PRODUCTS/9052.HTM

Having read on this forum references as to the:

A) numerous re-mountings of the water block necessary to get a "best fit"

B) The required flatness and evenness of both surfaces (CPU core and WB)

C) The importance that pressure variations (below 100PSI aka Point of Diminishing Returns) can have

Ask myself if it such a device would greatly increase the accuracy of test results, when benchmarking water blocks?

[jaydee]

Originally posted by iggiebee


Ask myself if it such a device would greatly increase the accuracy of test results, when benchmarking water blocks?

Not sure why the image tags are not working?

Yes that is a nice machine, cost more than my truck and all my computers combined though I would bet (which is only about $5,000). I doubt it would "greatly" increase the accuracy as most any serious tester remounts 5+ times. Would cut time way down though if you only had to mount and test a block once....

[iggiebee]

Not sure why the image tags are not working?

Yes that is a nice machine, cost more than my truck and all my computers combined though I would bet (which is only about $5,000). I doubt it would "greatly" increase the accuracy as most any serious tester remounts 5+ times. Would cut time way down though if you only had to mount and test a block once....

On the other hand.. having taken a "good look" at this device, constructing a DIY ghetto versión does seems to be a good possibility.

Edit: added following..

Question: Also, although I have never been involed in testing WBs, do you get different results every time you re-mount + test a WB? or does it comes to a point where various re-mounts + tests do offer same or very similar results?

If question 1 holds true, then will benefit by using such a device, simply because it will standarize the re-mounting + test process IMHO

[jaydee]

Originally posted by iggiebee
On the other hand.. having taken a "good look" at this device, constructing a DIY ghetto versión does seems to be a good possibility.

Edit: added following..

Question: Also, although I have never been involed in testing WBs, do you get different results every time you re-mount + test a WB? or does it comes to a point where various re-mounts + tests do offer same or very similar results?

If question 1 holds true, then will benefit by using such a device, simply because it will standarize the re-mounting + test process IMHO

Yes remounting generally will get different results but usually by a C or less. I usually can't pick up the difference with my half assed equipment.

Here is a graph from Bill Adams testing. He mounts them 10 times.



http://www.overclockers.com/articles720/index02.asp

[bigben2k]

Within the max 0.231, and the min 0.225, assuming an 80 Watt source, the temps could differ by 1/2 degree.

Bill has also refined his application method: your results will vary

[jaydee]

Originally posted by bigben2k
Within the max 0.231, and the min 0.225, assuming an 80 Watt source, the temps could differ by 1/2 degree.

Bill has also refined his application method: your results will vary

He uses a 70watt source.

A little more info for the ones that are alergic to links.

The MCW5000 was then mounted under carefully controlled conditions 10 times and it’s “C/W” determined under ‘standard conditions’ (defined as 70W applied heat load, 3.8lpm coolant flow rate at 25.0°C, and 10kgf applied compressive load across the TIM joint). The following results were obtained:

“C/W” mean = 0.229
“C/W” range = 0.006
“C/W” standard deviation = 2.06E-03
The data is consistent with a normal distribution: P= 0.89 where the normal distribution has a mean = 0.2286 and standard deviation = 2.25348E-03. The "C/W" distribution is shown below. 'Standard Error' bars are indicated to enable an estimate of the measurement uncertainty.

The “C/W” was then determined with an applied heat load of 70W (about 100W per Radiate) under a range of flow rates from 1lpm to almost 13lpm.