Each product type deserves its own testing methodology. CPU coolers, which we regularly put to the test, are no exception to this. Within this article, we introduce our latest test method, which allows us to even more accurately assess which models offer the best balance between cooling performance and noise. Keep reading to learn all about our new testing methodology, and which coolers you should purchase according to it.
Ever since the inception of Hardware.Info some sixteen years ago, we have performed at least one large round-up of CPU coolers every year. Naturally, we try to ensure that our tests are as scientifically sound as possible. Competing publications often times simply place a CPU cooler on a real processor, use certain software to push said processor to its limit, and then measure the resulting temperatures. This approach has a number of flaws, in our opinion. For one, when using real CPUs, you can never guarantee that they'll produce an identical amount of heat during each test. Furthermore, the temperature sensors that are incorporated in processors are far less accurate than monitoring software such as Coretemp and Realtemp would have you believe.
Our first few generations of CPU cooler test setups consisted of power resistors whose dimensions were approximately equal to those of contemporary processors. By means of a variable power supply, we were able to have our resistors draw a consistent amount of power, causing our setup to always generate an identical amount of heat. Because a power resistor, much like a processor, doesn't emit any light and doesn't have any moving parts, the law of conversation of energy teaches us that all of the absorbed power will be converted into heat.
A few years ago, we started using official CPU simulators, similar to those developed by Intel for manufacturers of CPU coolers. While these are essentially also a type of power resistor, they are incorporated into a real CPU package, and accurately reflect the thermal properties of a real processor.
Of course, there's always room for improvement. Up until now, our test procedure involved testing the CPU coolers both while the fans were running at 12V and at 7V. We would refer to these two settings as high speed and low speed in our tables. Using these two settings, we performed tests on both our Socket 1155 and our Socket 2011 CPU simulators. In addition, we measured the noise levels of both speed settings in our beloved soundproof box.
Achieving good cooling performance isn't particularly difficult if you completely disregard noise. Conversely, a combination of poor cooling performance and low noise levels isn't all that difficult to achieve either. Because of this, we always calculated so-called efficiency scores, obtained by dividing the temperature results by the noise levels. Naturally, we would also include a disclaimer on how we were well aware of the fact that dividing a unit on a linear scale with another unit on a logarithmic scale is not particularly mathematically sound, especially when taking into consideration that the temperature and the noise level units are pretty much unrelated to one another.
Nevertheless, these efficiency calculations did always result in interesting numbers, based on which we could draw conclusions on how well a cooler has been constructed. However, the fact that we performed our measurements on the two extremes of a scale (7V and 12V) was a little problematic, given that the fans of CPU coolers often times operate on voltages that are somewhere between these two values. Furthermore, you simply won't be using certain CPU coolers at their maximum voltage due to the amount of noise they produce, whereas other models have their fans running at extremely low speeds at 7V, which results in high efficiency scores despite a distinct lack of actual cooling. In short, our efficiency score had a number of downsides, and could certainly be improved upon.