Testbed Configuration and Testing Methodology

We tested all thermal interfaces discussed in this article inside a system case with the side panel removed. Our testbed configuration remained unchanged for the entire test session and included the following components:

• Mainboard: DFI LANPARTY DK X48-T2RS (Intel X48), LGA 775, BIOS 08/29/2008
• Processor: Intel Core 2 Extreme QX9650 (3.0GHz, 1.25V, 2x6MB L2 cache, 4x333MHz FSB, Yorkfield, C0)
• Graphics card: HIS Radeon HD 4870 GDDR5 512MB / 256bit, @830 (300) / 3600MHz
• Graphics card cooler: Arctic Cooling Accelero Twin Turbo (~2090 RPM x2);
• Memory:
  • 2 x 1024MB DDR2 Corsair Dominator TWIN2X2048-9136C5D (1142MHz / 5-5-5-18 / 2.1V);
  • 2 x 1024MB DDR2 CSXO-XAC-1200-2GB-KIT DIABLO (1200MHz / 5-5-5-16 / 2.4V).
• Disk subsystem: Western Digital VelociRaptor (SATA-II, 300GB storage capacity, 10,000RPM, 16MB cache, NCQ)
• HDD silencer and cooler: Scythe Quiet Drive 3.5”
• Optical drive: Samsung SH-S183L SATA-II DVD RAM & DVD±R/RW & CD±RW
• System case: ASUS ASCOT 6AR2-B Black&Silver (ATX) with 120mm ~960RPM Scythe Slip Stream 120 fans for air intake and exhaust (the fans are installed on silicon spindles), the side panel removed
• Control and monitoring panel: Zalman ZM-MFC2
• Power supply: Thermaltake Toughpower 1500W W0218 (with a default 130 mm fan)
• Monitor: 24" BenQ FP241W (Wide LCD, 1920 x 1200 / 60 Hz)

All tests were performed under Windows Vista Ultimate Edition x86 SP1. SpeedFan 4.36 beta 15 was used to monitor the temperature of the CPU, reading it directly from the CPU core sensor:

The mainboard’s automatic fan speed management feature as well as CPU power-saving technologies were disabled for the time of the tests in the mainboard BIOS. The CPU thermal throttling was controlled with the RightMark CPU Clock Utility version 2.35.0:

The CPU was heated up in two modes. First we used Linpack 32-bit with very convenient LinX shell version 0.4.9 to heat it up to its maximum. We manually set the RAM capacity at 1850MB and recorded 15 runs.

Besides, since Linpack heats the CPU significantly, we also performed an additional test by running OCCT v2.0.0a CPU test for 23 minutes (with maximum priority):

The testing procedure on CPUs looked as follows. I applied the tested thermal interface, installed the cooler and started the system. I let it run for ~20 minutes for all system components to reach their nominal temperatures and stabilize. As a rule, this period of time there were some office applications running. Then I launched OCCT processor test. Once the test was completed, I let the system stabilize for another ~10 minutes and repeated OCCT once again. The next stabilization period was 20 minutes. After that I launched LinX test. When the test run was completed, a 10-minute stabilization period followed, and then I repeated LinX once again. In both cases, we recorded the results of the second test cycle, because the temperature readings in this case were often about 0.5~1°C higher than in the first test cycle.

However, this wasn’t the end of our test session on a CPU. After a complete cycle of tests with two applications we dismounted the cooler, removed the remaining thermal interface from the processor heat-spreader and the cooler base and degreased them with alcohol. Then we applied a new layer of thermal interface, installed the cooler and ran all the tests over again. This way, we tested each thermal interface during two 3-hour test cycles. The only exception was Arctic Silver 5 which was the first to be tested and which has already been in use for four days. The final result that you see on the diagrams is the average temperature of all four processor cores. We took the best result of the two test cycles for each thermal compound. To be fair I have to say that the results during first and second application of the thermal compound differed by 0.5~1°C at the most.

During our test session we paid special attention to ambient temperature, because it was one of the key factors to influence the final result. It was checked next to the system case with an electronic thermometer that allows monitoring the temperature changes over the past 6 hours. During our test session room temperatures varied between 24.5~25.0°C. So, the temperature deviations didn’t exceed 0.5°C. Taking into account the above described testing methodology I allow a measuring error of no more than 0.5°C.

In order to increase the dependence of the CPU cooling efficiency on the thermal interface used, we employed a highly efficient ZEROtherm ZEN FZ120 cooler. We replaced its default fan with two 9-blade Scythe Minebea Silent IC fans at ~1130 RPM each. They were attached to the heatsink for air intake/exhaust:

We decided on a ZEROtherm cooler, also because its base is not nickel-plated, like on some other cooling systems. Besides, its surface is well-finished, but not polished:

The cooler base is almost impeccably even. And the processor heat-spreader has long been perfected to the maximum (I evened its surface with 1000x sanding paper). In other words, both contact surfaces are made of copper and have been very well finished. The surfaces haven’t been polished. The cooler was always installed the same way and was fastened with retention screws evenly tightened in diagonal pairs.

Besides the tests on a CPU, we also checked the efficiency of our thermal interfaces on RV770 GPU of Radeon HD 4870 graphics card. This GPU die size is 256sq.mm. Although the graphics processor and the graphics card itself boast pretty high heat dissipation already, we increased the frequency to 830MHz (300MHz in 2D mode). According to the results of our latest VGA cooler tests, reference cooler of ATI Radeon HD 4870 is pretty efficient, however, only if you are eager to put up with relatively high noise. Since I wasn’t ready for sacrifices like that, I replaced the default reference cooler with Arctic Cooling Accelero Twin Turbo with two fans working at their maximum speed of ~2090 RPM:

I still used the lower heatsink plate from the Radeon HD 4870 reference cooling system, so it helped with cooling the PCB voltage regulator components and ensured higher stability during overclocking. We used the backplate from the Radeon HD 4870 reference cooler, the pressure was applied not to the GPU corners but to the entire die.

Let’s take a look at the cooler base and GPU top surface:

There is nothing I could add about the GPU (this is the only way it can be without the heat-spreader). As for the base of Accelero Twin Turbo cooler, it is even, has average finish quality. I didn’t improve it in any way.

Radeon HD 4870 was warmed up by two 12-run cycles of Firefly Forest benchmark from the synthetic 3DMark 2006 suite in 1920x1200 resolution with activated x16 anisotropic filtering but without FSAA. Moreover, to create additional GPU workload we used FurMark version 1.4.0 in stability test mode in 1024x768 resolution (window mode). We ran the tests twice for each thermal compound with 10~12 minutes stabilization period between the test cycles. We installed the cooler twice for each thermal compound, the same way we did during our CPU tests – after removing the remaining thermal interface and degreasing the surfaces with alcohol. We monitored the graphics card temperatures using RivaTuner v2.11 utility (created by Aleksey Nikolaichuk aka Unwinder). We took the average reading from four GPU diodes: "Core t", "Core t, display IO", "Core t, memory IO" and "Core t, shader core".

Besides 15 thermal interfaces discussed in our today’s roundup we also added the results of the 16^th participant: SilMORE thermal compound:

We decided to include this common thermal compound into our today’s test session because most manufacturers use SilMORE thermal interface to bundle with their cooling solutions. For example, Scythe, Thermaltake, Xigmatek, Ice Hammer, OCZ and several other less popular makers include a 1g pack of SilMORE thermal compound with their coolers. So, the results will help us determine how many extra degrees we can win by replacing the standard SilMORE thermal interface with one of the reviewed today.