I am all done, finally. For those of you that just want to see random results without knowing what I did, skip to the bottom.
To create the heater, I soldered 7 x 1.8 Ohm resistors together in series. I then wrapped the resistors with copper braid and soldered the braid to a small steel plate. The steel plate was purposely unpolished in order to see the performance of the TIM's. A power supply of 19v was attached to the resistors providing an output of 28.6W. This may seem small to some, but the results can be scaled up easily, and this was the maximum the resistors could handle. Using a multimetter, the exact values for voltage and resistence gave an output of 29.8W.
The heatsink is an old OCZ model that has three heat pipes passing through a copper block at the cpu end. The heat pipes then pass through aluminum fins at the fan end.
I actually didn't have a motherboard that could mount my heatsink so I used a piece of wood, I hallowed out a cavity in the wood for the resistors to fit into and then mounted the steel plate on top of the wood. I then made anchors for the heatsink to attach on either side of the steel plate. The contact area between the steel plate and the heatsink is 3.5cm X 3cm.
On the steel plate off to one corner was securely attached a thermocouple which ran to a multimeter to read its output. I drilled a small hole in the copper block of the heatsink and ran another thermocouple inside. This thermocouple was connected to a dedicated display module. I used a small amount of Ceramique on the thermocouples to ensure good contact. Each thermocouple was calibrated such the results at any given temperature differed by less than 0.3F.
- Air - I ran the system completely dry as a control.
- Dow Thermal Fluid - Dow Corning make their own special TIM. It is just silicone oil and zinc, but it is very viscous (90,000cs).
- Pure Silicone Oil - Also from Dow Corning but has a viscosity of 10,000cs. DC is very generous with their free samples.
- Silicone and Diamond Slurry - This is the DC silicone oil with 100,000 mesh diamonds add with a ratio of 1:1 oil to diamind by weight.
- Dow Fluid w/ Diamonds - This is the DC thermal fluid with 100,000 mesh diamonds add with a ratio of 3:1 fluid to diamond by weight.
- Inventgeek.com Remake - Lastly, I attempted to follow the procedure of the inventgeek article. I used 100,000 mesh diamonds and for the silicone grease, I used one with PTFE particles already in it: click. The diamond to everything else ratio was 5:2 by weight.
To make the DIY TIM's, I placed a small glass container onto a scale, measured out the amount of diamond and grease, and then mixed it by hand using a toothpick. There was no danger of inhalation as mentioned in the inventgeek article. The particles were sufficiently dense and clumped together that they didn't go anywhere. I then finished the mixing using a dremel and a special attachment that made it like a batter mixer.
For all of the TIM's that I made for myself (except for the inventgeek batch), I placed them inside of my vacuum chamber to remove air bubbles. (warning, I'm going to get nerdy here) I wanted to know what pressure I need to pump my chamber down to in order for a majority of the bubbles to be removed. In order to do that you need to combine the Young-Laplace equation with the Ideal Gas Law. You end up with a third order polynomial. I solved the equation with matlab for different bubble radii. As you can see below, even with absolute zero pressure, some bubbles will only double in size. Meaning that in a very viscous fluid, they will not float out that much faster.
With the vacuum chamber I created, I can get down to 26 to 26.5 in Hg vacuum. This means that for my altitude most of the bubbles greater than 10 um will join and float out. When vacuuming the TIM's I left them in the chamber for an hour to get the majority of the bubbles out. (It is really cool when you pump down the chamber and your TIM looks like it's boiling. I actually did end up boiling water using the chamber to verify that the pressure readings were correct.)
For the actual testing, I first tried out all of the different fluids on the test rig to both see how each of them would spread and to thoroughly get the surface contaminated by all of them. I wanted to make sure that the order in which I did the experiments did not effect the results. The procedure for a standard test was to clean the steel plate and the heatsink with a dry rag. Then I would clean both surfaces using rubbing alcohol and a lint free cloth. I would place a dime sized amount of the TIM in the center of the steel plate, and then press down on it with the heat sink until it could be attached to the wood. Once the heatsink was attached, the fan was turned on and allowed to run for 30 mins to cool down the entire system. The test rig was positioned in the center of an air conditioned room. The ambient air temperature was taken to be the average of the two thermal couples after the 30 min cool down. This was consistently ~71 F and was used as the starting temperature in the data. The heater was then turned on and a stopwatch was used to track the time. Measurements were taken at 1,2,3,4,5,10,15,20,25,30,and 60 mins. Each TIM was tested three times and the average of all three runs was used. The order of the testing was random with each of the three runs for a particular TIM spread out over the course of all testing.
In the preliminary testing, I experimented with spreading the TIM across the entire heatsink with a razor before attaching it. That always caused an drastic decrease in performance. For this reason, that method was not used.
T1 is the steel plate temperature.
T2 is the heatsink temperature.
The margin of error is +- 0.3 F.
The first figure shows the temperature difference between the heat sink and the ambient air temperature. As you can see for all the runs that even though the different TIM's caused different delays in reaching an equilibrium temperature, they all reached the same equilibrium temperature. This is expected since the amount of the heat it is outputting (31W) is the same regardless of the TIM used. The time axis is graphed in log form in order to see all the data points. The lines in between the data points are just for continuity.
Next is a figure showing the the temperature difference between the steel plate and the ambient air temperature. The worst performing TIM is obviously air. Next is the mixture mimicking the directions from inventgeek. It was the only DIY TIM that was not put in the vacuum chamber.
Lastly, we have the temperature difference between the heatsink and the steel plate. I've excluded the air run so that we can distinguish the results better. As you can see at equilibrium, AS5, Cermaique, and the Dow Corning Thermal Fluid were all very close to one another with AS5 being the best out of the group. For pure silicone oil, you can see that it is better than the inventgeek mixture. This is most likely because of the air trapped inside of the mixture. It could have also been because the mixture was so viscous that it did not thin as much as the oil. Once we add the diamonds, we improve the performance of the silicone oil but not by that much. Perhaps if I added more diamonds, the performance would increase, but the mixture was already pretty viscous and dry. When we added diamonds to the Dow Thermal Fluid, it improved its performance. It was actually the best performing out of all of them. The catch is that it did not improve it by all that much. As you can see, there isn't much room for performance anyways. The temperature difference between the heatsink and the steel plate for the best performing TIM's was in the 5-6 F range.
The adding of diamonds did improve the performance of the of both the silicone oil and DC Thermal Fluid, but it did not improve them by that much. A few things might be effecting this. The mixtures might not be mixed well enough leaving clumps of diamonds. The problem with increased mixing is that it introduces more sub mm and um sized bubble. And since we cannot remove the smallest of these bubbles, increased mixing might lead to worse performance. Also, the amount of diamonds may have been too small. The problem with adding more is the viscosity of the slurry increases. Since slurries behave like solids and liquids, this may effect the mixtures ability to thin once the heatsink is added. Playing with juts these two variables adds two orders of magnitude to the complexity of the problem and that is why I did not explore them further. However, it is safe to conclude that you will not stumble upon the optimal mixture as inventgeek seems to suggest they did.
The best performing TIM turned out to be the DC Thermal Fluid with diamonds. It preformed slightly better than AS5. One may wonder why I didn't add diamonds to AS5. That is because I'm not made of money. It may have also decreased performance if I added too much or if the mixing added too many bubbles. But in the end, performance could not be improved by that much anyways.
If we take the temperature difference between the the plate and heatsink to be the average temperature across both interfaces, then the overall performance of the best performing TIM's are 4700-5700 Watts/F*m^2. Even if you were to drastically improve that, since the temperature of the heatsink is fixed by its own performance, you'd only be able to get 4-5 F improvement at the very extreme.
These results are for ~30 Watt output but they can easily be scaled up. For 60 watts, the temperature difference for the heatsink to ambient would roughly double, 6.5F --> 13F. The same would happen for the temperature difference between the steel plate and the heat sink 5F --> 10F. So with the same ambient temperature of 71F, doubling the heat output with take the heatsink equilibrium temperature from 83F to 94F. Reducing the area, will have a similar effect.
When you look at this as cost vs performance, the Dow Corning Thermal Fluid wins hands down. They sent me 1 kg of it for free. It works better than Ceramique and will be good enough for 99% of applications.
As you can see the diamonds did not perform any sort of miracle as reported by inventgeek, nor can it. All of the inconsistencies in their procedure and advice (don't add the diamonds to a zinc oxide grease as I did for the best results), combined with their insane results which are non reproducible makes me conclude that this was all a big joke played by them.
Lastly, here are some images of my setup and experimenting. Also, I've attached the data from all of my experiments. Thank you and good night.
I didn't have a picture of the fire extinguisher before I cut it. So I included an artists rendering of what it looked like before.
Test rig front Test rig side
is the Dow Thermal Fluid with diamonds that beat AS5. I might try this later on an actually cpu and post results but I'm done for now.Edited by freddyman - 9/5/09 at 2:40pm