|Noctua NH-U12P Heatpipe CPU Cooler|
|Reviews - Featured Reviews: Cooling|
|Written by Olin Coles|
|Saturday, 03 May 2008|
Page 4 of 6
Thermal Interface Material Application
Over the past several months, I have read an unreasonable number of discussion forum posts which offer inaccurate and often times incorrect information. It's not really all that surprising to read poorly conceived information on the Internet, which seems to be a anonymous means of passing off opinion for fact. As a general rule we don't let too many things go untested, and the advice of wanna-be experts is not doing the hardware enthusiast and overclocker community any good. In this article, Benchmark Reviews dispels myth and establishes fact on the topic of proper application in our Best Thermal Paste Application Methods article.
After we wrote our 33-Way Thermal Interface Material Comparison article, many enthusiasts argued that by spreading out the TIM with a latex glove (or finger cover) was not the best way to distribute the interface material. Most answers from both the professional reviewer industry as well as enthusiast community claim that you should use a single drop "about the size of a pea". Well, we tried that advice, and it turns out that maybe the community isn't as keen as they thought. The example image below is of a few frozen peas beside a small BB size drop of OCZ Freeze TIM. The image beside it is of the same cooler two hours later after we completed testing.
After discussing this topic with real industry experts who are much more informed of the process, they offered some specific advice that didn't appear to be a "one size fits all" answer:
The more we researched this subject, the more we discovered that because there are so many different cooling solutions on the market it becomes impossible to give generalized advice to specific situations. This is where our testing comes into play. For the tests in this article, the processor received a thin layer of thermal paste which was kept consistent throughout every test. Ultimately it is the contact pressure of the coolers retaining system that is created when the elements are mounted with enough force which ensure excellent thermal conductivity between metals. With this principal kept in mind, those coolers with the stronger retaining system will often times benefit from the improved thermal conductance.
Surface Finish Impact
Here's the part I've been waiting to reveal... the importance of surface finish in relation to the impact on thermal conductivity. CPU coolers primarily depend on two heat transfer methods: conduction and radiation (heat-pipes also add convection). This being the case, let's start with conduction as it related to the mating surface between a heat source and a cooler.
Because of their density, metals are the best conductors of thermal energy. As density decreases so does conduction, which relegates fluids to be naturally less conductive. So ideally the less fluid between metals, the better heat will transfer between them. Ultimately, this means that the perfectly flat and well-polished surface (Noctua NH-U12P) is going to be preferred over the rougher and less even surface which required more TIM to fill the gaps (Thermalright Ultra-120 eXtreme).
Heat radiation is different however, and requires exactly the opposite. Because gases (air) are naturally poor heat conductors, surface area is key to the performance of cooling through radiation. This type of cooling is what you commonly see automobile radiators, which utilize large arrays of metal fins to radiate heat to be drawn away by a fan. The same is true for the CPU cooler, which needs as much surface area as possible to optimize it's radiative effects. OCZ and others have recognized that the surface of a heatsink does not have to be the sum of its overall size. By adding dimples and bends, the surface area is increased without growing the overall size.
To sum it all up, science teaches us that a smooth flat mating surface is ideal for CPU coolers so that less Thermal Interface Material is used. Because these coolers are using fans to force air over the heatsinks fins, the overall surface area of those fins should be as large and uneven as possible. In the next section we'll find out just how well all of these principals worked for our collection of test products.
Testing was conducted in a loosely scientific manner. Ambient room temperatures levels were held to within one degree of fluctuation measured at static point beside the test equipment with a calibrated digital thermometer. All coolers had their original manufacturer-supplied fan removed and replaced with our specified common test fan. Each product then received the same amount of Thermal Interface Material (specified below), which amounted to roughly a BB-sized drop placed onto the center of the CPU. The CPU cooler product being tested was then laid down flat onto the CPU, and compressed to the motherboard using the supplied mechanism. If the mounting mechanism used only two point of force, they were tightened in alternation; standard clip-style mounting with four securing points were compressed using the cross-over method. Once installed, the system was tested for a baseline reading prior to testing.
At the start of each test, the ambient room temperature was measured to track any fluctuation throughout the testing period. Lavalys EVEREST Ultimate Engineer Version 4.20.1170 was then utilized to create core loads and measure each individual CPU core temperature. It's important to note that software-based temperature readings reflect the thermistor output as recorded by the BIOS. For this reason, it is critically important to use the exact same software and BIOS versions throughout the entire test cycle, or the results will be incomparable. All of the units compared in our results were tested on the same motherboard using the same BIOS and software, with only the product itself changing in each test. These readings are neither absolute nor calibrated, since every BIOS is programmed differently. Nevertheless, all results are still comparable and relative to each products in our test bed.
One unfortunate problem is that CPU's report temperatures as a whole number and not in fractions. This in turn causes the motherboard BIOS and subsequent software applications such as EVEREST to also report to the nearest whole number. To compensate for this, our tests were conducted several times after complete power down thermal cycles. Conversely, the ambient room temperature levels were all recorded and accurate to one-tenth of a degree Celsius.
All of our tests are conducted using two different product orientations: horizontal and vertical. So far as we can tell Benchmark Reviews is probably the first website to test with this method; but it's very likely that others will soon follow our lead. At the start of our test period, the test system is orientated sideways in a flat "desktop" position which places the motherboard and processor horizontally to face up towards the ceiling. Next, the computer system is powered on and EVEREST system stability tests are started with Stress CPU and Stress FPU options selected. Then for a minimum of ten minutes EVEREST loads each CPU core to 100% usage, which drives the temperature to its highest point. Finally, once temperatures have sustained a plateau, the ending ambient room temperature and CPU core levels are recorded and the first benchmark segment is complete. Lavalys EVEREST remains running at full load into the next test segment.
The second benchmark segment begins by simply turning the test system vertically upright, so that the motherboard and CPU are facing to the side. Many of the products we have tested utilize a "U" pattern in the heat-pipe rods, and the upright system orientation favors this particular product design because it removes the effect of gravity on the heat-pipes' thermal cycle. For a minimum of five additional minutes EVEREST continues to load each CPU core, and once temperatures have plateaued the ending ambient room temperature and CPU core levels are recorded. This process was identical for all cooling solutions used in our benchmark tests segments.