Why Do We Test Each Paste In Four Scenarios?
In choosing four test platforms, I incorporate feedback from our valued readers. For instance, you wanted to see us take the cooler mounting pressure into account. We're skipping the LN2-based testing and focusing on scenarios you'll encounter in the real world. For example, we're using popular pre-assembled water coolers that should ensure heat sink temperatures below 60 °C (140 °F), premium aftermarket air coolers with back plates that should demonstrate high mounting pressure, and a run-of-the-mill budget cooler with push-pin mounting (that'll give us limited pressure). Stock coolers like that let the CPU get above 60 °C/140 °F (AMD) and 80 °C/176 °F (Intel).
Depending on viscosity and composition, not all pastes are a good fit for every application, nor are they all well-suited for novices. This warning also applies to replacing the heat sink on your GPU, and we'll discuss that separately below.
First, let's take a look at the three systems I used to test each thermal compound.
Test System One: Closed-Loop Liquid Cooling | |
---|---|
Cooler | Corsair H80i |
Fan | Original H80i fan, powered from an unregulated 7 V supply |
CPU | AMD FX-8350 |
Motherboard | Asus 990FX Sabertooth |
Test System Two: Air Cooler with Back Plate (Screwed On) | |
---|---|
Cooler | be quiet! Shadow Rock |
Fan | Original Shadow Rock fan, speed set to 70% |
CPU | Intel Core 2 Quad Q6600 (Q0 Stepping) At 2.66 GHz |
Motherboard | Gigabyte UP45-UD3LR |
Test System Three: Intel Boxed Cooler (Mounted with Push Pins) | |
---|---|
Cooler | Intel Boxed Cooler |
Fan | Original Intel Fan, Speed set to 80% |
CPU | Intel Core 2 Duo E6850 |
Mainboard | Gigabyte UP45-UD3LR |
Testing Thermal Paste on a Graphics Card
This is a somewhat sensitive topic, and for safety reasons I excluded electrically conductive or liquid metal pastes from my testing. Since GPUs don’t have a heat spreader, but allow the cooler’s sink to directly sit on the die, I didn't want anyone to risk a short circuit.
I also used an older graphics card, which was convenient to test with. Its cooler was mounted using four screws and its fan speed could be dialed in to a constant value. Furthermore, I figured that an older card would be more tolerant of the higher temperatures I'd see. After all, we didn't want a cheap paste to destroy an expensive, current-gen board. Fortunately, the GPU die size and surface temperature are still comparable to modern mid-range cards.
Test System Four: Graphics Card Test | |
---|---|
Cooler | Zalman GPU cooler |
Fan | Original Zalman fan, speed set to 80% |
CPU | AMD Radeon HD 4850 |
Test Environment | Test System 1 (see above) |
Test Cycles, Test Duration and Settings
I should also explain how I took my measurements. Because the digital temperature sensors built-in to modern CPUs only give us uncalibrated Tcore values, I used the old way of measuring die temperature with a thermal diode under the heat spreader. The processors in this story still use soldered-on spreaders, so this value should be fairly accurate. I'll report the difference between Tcase and the room temperature, since that latter figure wasn't as constant as I would have liked through all of the benchmarking.
For the graphics card, I used the temperature as the GPU reported it. That number wasn't influenced by minor changes in room temperature.
Test Environment | |
---|---|
Room Temperature | Approx. 72 °F (between 70 and 73 °F) |
Results of CPU Tests | Reported in °C as an average of temperature differences (Difference between the ambient temperature and the reading of the sensor under the heat spreader) |
Results of GPU Tests | Reported in °C according to the temperature sensor on the GPU |
Test Cycles CPU | 1 x four-hour burn-in, followed by break of at least two hours 4 x one-hour measurement, with one-hour breaks Total time at least 16 hours per thermal product and cooler |
Test Cycles GPU | 1 x four-hour burn-in, followed by break of at least two hours 2 x one-hour measurement, with 30 minute breaks Total time at least 8.5 hours per thermal product |