It is far more difficult to test thermal interfaces than, for example, coolers because the difference between them is much smaller. You have to catch a difference of 1°C or less, which is a nontrivial task if you don't have a thermal chamber. For thermal interfaces to differ more notably, it is necessary to put them under such conditions that they are the weakest link in the heat transfer chain between the processor die and the ambient air. To do this, a high-wattage heater and a very efficient cooler are needed. There is no problem to find the former. As for a cooler, a liquid cooling system would be a perfect choice, but it would be much harder to deal with in terms of replacing the thermal greases. Moreover, this test wouldn’t have much practical value because liquid cooling systems are far less popular than air-based coolers.
So, carrying out such tests and ensuring repeatable results is not a trivial task.
For our test session we assembled the following test platform:
- Mainboard: Gigabyte GA-X58A-UD9 (Intel X58 Express, LGA 1366, BIOS F5i);
- Processor: Intel Core i7-980X Extreme Edition, 3.33 GHz, 1.225 V, 6 x 256 KB L2, 12 MB L3 (Gulftown, B1);
- CPU cooler: Thermalright Archon (2x140 mm Thermalright TY-140 fans at 800 RPM);
- Thermal interface: Gelid GX-Extreme;
- Memory: DDR3 3 x 2 GB OCZ Platinum Low-Voltage Triple Channel (Spec: 1600MHz / 7-7-7-24 / 1.65 V);
- Graphics card: ATI Radeon HD 6950 @ 6970 2 GB (GDDR5 256 bit, 880/5500 MHz);
- System drive: RAID 0 made of 2 Kingston V-series SNV425S2128GB SSDs (SATA-II, 300 GB, MLC, Toshiba TC58NCF618G3T);
- HDD for games and programs: Western Digital VelociRaptor WD3000HLFS (SATA-II, 300 GB storage capacity, 10,000 RPM, 16 MB cache, NCQ) inside Scythe Quiet Drive 3.5” silencer and cooler chassis;
- Backup HDD: Samsung EcoGreen F4 HD204UI (SATA-II, 2000 GB, 5400 RPM, 32 MB, NCQ);
- System case: Antec Twelve Hundred (front panel: three Noiseblocker NB-Multiframe S-Series MF12-S2 fans at 780 RPM; back panel: two Noiseblocker NB-BlackSilent PRO PL-1 fans at 780 RPM; top panel: standard 200 mm fan at 400 RPM; side panel removed);
- Control and monitoring panel: Zalman ZM-MFC2;
- Power supply: Zalman ZM1000-HP 1000 W (with a default 140 mm fan).
That’s a photo of my system during the tests (except for the Indigo Xtreme which was tested with the mainboard oriented horizontally):
The ambient temperature was monitored with an electronic thermometer and varied from 23.9 to 24.5°C during the tests. The test results were normalized to 24°C by adding or subtracting the deflection of the ambient temperature from that value to or from the result of each thermal interface. By the way, if I was standing next to the system case during the test, the air temperature would grow up by 0.2-0.3°C.
I tested the thermal greases using the GPU of an AMD Radeon HD 6950 graphics card transformed into a Radeon HD 6970 by rewriting its BIOS.
The Cayman processor is the largest of open-die GPUs and is also among the hottest. I don't think there is a graphics card better suitable for such tests. I also used one of the most efficient coolers available: an Arctic Cooling Accelero XTREME 5870.
The rotation speed of the cooler’s three fans was fixed at the maximum 1920 RPM.
The cooler’s base is ideally flat and smooth, even though it seems rather rough.
Each thermal grease was applied in a thin and uniform layer on the GPU die. Each was applied twice, with both surfaces being cleaned and degreased with alcohol between the tests. When applied, the thermal grease was left working for 1 hour to “break in” before the test proper. I ran Unigine Heaven 2.1 at 2560x1600 with 16x anisotropic filtering for that. After this warming up and when the GPU temperature had stabilized (this would take about 10 to 15 minutes), I would begin the test proper.
During the test the graphics card was running FurMark 1.8.2 launched from a renamed EXE-file at 2560x1600 with 16x anisotropic filtering enabled in the Catalyst driver. I used two monitoring tools: GPU-Z version 0.5.0 and MSI Afterburner 2.1.0 beta 5.
There were at least two such test cycles with a 20-minute pause for cooling down and stabilizing the temperature. I did this with each thermal grease for a total of over 50 test cycles. I didn’t overclock the computer when testing the thermal interfaces with the GPU.
I also tested some of them with my CPU. It would have taken extremely long to test every product on the CPU, so I took only four of them of that, including the Indigo Xtreme. The other three were the best, medium and worst performers in the GPU tests. Another reason why I didn't test all the products on the CPU is that the heat-spreader of my Intel Core i7 Extreme Edition i7-980X processor is not flat but convex.
I took a Coolink Corator DS cooler with Gapless Direct technology for testing the mentioned thermal greases with my CPU. It is a direct-touch cooler that doesn't have any gaps or aluminum inserts in its base. Theoretically, the thermal interface should have a higher effect on cooling with a direct-contact cooler than with a classic cooler because one heat transfer is eliminated (from the cooler's base to the heat pipes). To make the Corator DS even more effective, I replaced its 120mm fan with two 140mm Thermalright TY-140 fans which were working at their maximum speed of 1310 RPM.
To increase the amount of heat produced by the CPU, I overclocked it to 4.25 GHz, setting its voltage at 1.375 volts in the mainboard's BIOS.
The Turbo Boost technology was disabled whereas Hyper-Threading was turned on. Why did I use Hyper-Threading now but never do so in my CPU cooler tests? It’s because of the program for heating the CPU up. Using Linpack with the LinX frontend does not produce adequate results in tests of thermal interfaces because the Linpack load, although very high, is unstable and not very repeatable. When testing coolers, I can easily re-launch Linpack as soon as I spot it yielding gigaflop calculations greatly different from previous results (and this happens quite frequently I must tell you), but it would be too time-consuming in thermal interface tests.
So, I turned to the latest version of Prime95 x64 for a linear and stable CPU load. I launched it in the maximum-load Blend mode for 30-40 minutes during each test cycle.
The temperature of the CPU cores was monitored with the RealTemp utility. The average temperature across the six cores was then corrected according to the ambient temperature (24°C). Like on the GPU, each thermal grease was tested twice, the best of the two runs being regarded as the end result if the difference between the two runs was not higher than 1°C. If the difference was larger, I applied the thermal grease a third time and reran the warm-up and test cycle.
And the last thing: when applying each thermal grease for the first time, I threw away the first few millimeters of what I squeezed out of its tube to minimize the effect of the product's lifetime on its consistency and properties. Just another measure towards more accurate test results.