Thermal paste and CPU cooler test bench
Last year I published a thermal paste review on HWBot Forums. In total I tested 26 different thermal pastes on air and also using liquid nitrogen at different temperatures. Early 2015 I will publish a new and even more detailed test with new products and also liquid metal products. To provide accurate results I built a new test bench for thermal pastes and also to test CPU coolers in future.
Requirements and Basics
Following the test from 2013 a lot of thermal pastes are very close together taking a look at the performance. Differences of only 0.1 °C are very common and this is also the interesting and challenging point of thermal paste reviews.
A lot of websites just use CoreTemp and measure the differences in the CPU core temperature. The issue is that a temperature change of 1 °C, which is the smallest step, is already crashing the whole test. Also a varying room temperature of only 0.5 °C can easily ruin your results.
In the past I always measure the delta temperature between the CPU-Heatspreader and the base of the CPU-Cooler using K-Type thermal probes. However these temperature probes are quite inaccurate with sometimes up to 1 °C temperature differences depending on the quality of your probe and the thermometer.
Measuring the delta temperature eliminates the measuring errors of the changing room temperature. However the K-Type probes are still not accurate.
Changes and Improvements
To improve the accuracy I decided to build my own thermal paste test bench, on the basis of two copper plates, heating elements and very accurate thermometers.
The basic idea is to have one thick copper plate with 6 x 40 Watt heating elements. On top will be a surface area of about 30 x 30 mm which is similar to the latest core i7 Haswell CPUs. In addition to the heating elements there is a special PT100 measuring probe in the copper plate. The probe has an accuracy of 0.01 °C between 0 – 100 °C (using 1/10 DIN!).
Another copper plate will be placed on top with a second and similar probe. The delta temperature between these two probes will provide the performance of the thermal pastes.
To measure the temperature itself I use two Greisinger GMH 3750 High Precision thermometers. In combination with the 1/10 DIN the resolution is 0.01 °C with a maximum measuring error of 0.03 °C.
Both Greisinger GMH 3750 thermometers will be connected via USB to my laptop. Using the special Gresinger software you can read out the current temperature of both thermometers at the same time and also log them into a chart.
List of components
This is the list of almost all components and tools I used for building the thermal paste test bench and the price per piece:
- 2 x Greisinger GMH 3750 (260 €)
- 2 x Greisinger GTF 401 1/10DIN Probe (105 €)
- 1 x Greisinger GSOFT 3050 Software (75 €)
- 6 x 40 Watt heating elements (5 €)
- 2 x Copper plates with 60 x 30 x 60 mm (40 €)
- Several aluminium parts for the basis (100 €)
- Rubber stands (20 €)
- Polyamid screws (15 €)
- Several other screws (30 €)
- Phanteks PH-TC14 Cooler (75 €)
- 2 x 8-Pin-PCIe-Extender cable (5 €)
- 6 x Switch with 10 Ampere max (3 €)
- M3 thread former for copper (35 €)
- M4 thread former for copper (45 €)
- 12 mm milling cutter (80 €)
- 45 ° milling cutter (65 €)
- Several drills (50 €)
So total costs are about 1500 € + about 60 hours of work.
In the first step I drilled 3 holes with a diameter of 5,9 mm in each side for the heating elements with 6 mm diameter. To fix the heating elements I drilled another 6 holes in 90° angle.
Afterwards I used a reamer to get rid of the cuttings and to have the final diameter of 6 mm.
A first test and the heating elements fit perfectly fine.
To keep them at the same spot I wanted to use M3 setscrews. It’s quite difficult to cut proper threads into copper because of the high ductility and long cuttings. That’s why I use thread formers instead of thread cutters. The inner diameter of the tool itself is much bigger and it’s harder to break them. In addition the forming process will not produce any cuttings and a higher thread quality.
In the next step I took the copper block to the milling machine to cut out the upper contact surface.
And the finishing using a 45° mill.
Rubber stands will thermally and mechanically uncouple the copper plates from the base.
The 1/10 DIN PT100 probe has a diameter of 3 mm and will sit 10 mm underneath the surface area.
To ensure that the probe will always stay at the same position it will also be fixed by a long M3 setscrew.
Some industrial thermal paste will improve the thermal contact between the heating elements and the copper base.
A quick test and the sensor works and the temperature is very stable
To make the test bench compatible with different coolers each one will require an adapter plate. It will be fixed underneath the copper base. For the Phanteks cooler I had to have two long M3 screws with a mounting distance of 76 mm.
For the aluminium I also used thread formers instead of normal thread cutters.
The final adapter plate mounted to the copper base with a M3 screw.
The second plate will be mounted and aligned using two M3 setscrews which will be glued into the bottom plate. This way both plates will always be in the same position to have the same mounting.
Sanding the surface with 1200-grit sandpaper. No need to polish the area because the conditions are the same for all pastes anyway.
Cutting two grooves into each aluminium profile for the switches and cables.
One switch per heating element allows full control to have a variable heat load from 40 to 240 W in 40 W steps. Using 12 Volt power supply there will be about 3.3 A so I used switches rated at 10 A max.
Soldering the heating elements to the switches and a 8-Pin PCIe extender so I can just attach the test bench to a normal ATX PSU.
A quick test with a random paste shows very accurate temperatures and very stable after about 10-15 minutes.
And some impressions of the test bench so far. I will add some more parts soon to do CPU cooler testing.
The thermal paste results will follow soon
177,939 total views, 4 views today