Completely revamped test method
It's been a while since our last CPU cooler review, and there's a good reason for it. We've spent the last six months improving our test platform. This was a length process, where we tried several options and performed hundreds of reference measurements. In the end we found a platform that allowed us to get reproducible results under very varying circumstances, while also mimicking the circumstances in a PC case.
Our new test setup still uses official Intel CPU simulators, power resistors that have the appearance and dimensions of actual socket 1150 or socket 2011 processors, however a variable power supply allows us to generate a specific amount of heat so that we can mimic processors with various TDPs.
We did change the environment in which our test setup is used. This platform consists of an isolated box with a test compartment of 55 x 55 x 28 centimeters. There are three large, slow-turning fans that suck in air on the front of the compartment, while there is a single outtake fan on the back. We placed two heat elements in a folded air channel on the front of the case, each featuring a fan. A temperature sensor is placed in front of the entrance to the test compartment that measures the temperature of the air that is being blown into the test compartment. Using a closed-loop system with a self-learning PID controller we can keep the temperature very constant: the PID controller continuously measures the temperature of the air that is sucked in, after which the controller can generate extra heat if necessary through the heat elements.
We chose an air temperature of 35 degrees Celsius for our tests. This value was chosen for multiple reasons. First of all, the heat elements are capable of generating enough heat to reach this temperature under all circumstances, even if the ambient temperature is only 18 degrees Celsius. Furthermore, the ambient temperature in our test lab will (hopefully) never get higher than 35 degrees Celsius, which means that we'll never take in air that is warmer than 35 degrees. What's more the presence of some heat around the cooler is more realistic, since other components in a system, such as the power supply, motherboard and the graphics card will also produce heat. The base temperature should however not be too high, because we want to get a good image of what the CPU coolers are capable of. An ambient temperature of 35 degrees an excellent approximation of the temperature in an actual system, as we've seen in our separate tests for cases.
We're mostly interested in the efficiency of a cooler, when testing CPU coolers, in other words the cooling performance relative to the noise levels. A cooler that performs well while also producing an extreme amount of noise, is less attractive than a model with similar performance but much lower noise levels. In order to create an even playing field we test the noise levels of all coolers first. We measure the noise levels when the fan is running at 12 volts, in a soundproof box at a distance of 10 cm, after that we measure the voltage needed to produce 30 dB (silent) and 40 dB (audible, but not loud).
After that we test all coolers with two CPU simulators: one smaller model that simulates an LGA115x platform, and a larger one that simulates the LGA2011 processors. On the socket 115x-platform we test all coolers under a load of 65 and 95 watts, on the socket 2011 platform we tested at a higher load of 130 watts. Since the surface area between the coolers and the socket 2011 chip is significantly larger, the heat is dissipated easier, which results in significantly better cooling results for most coolers. The test results of a load of 130 watts on the socket 2011 platform generally don't differ much from the test results on 1150 under a load of 95 watts. All coolers in the comparison are compatible with our socket 115x-platform, but not all models could be mounted on the larger 2011 stand.
Some coolers lack test data. The reason for this is that they were not compatible with socket 2011, or that they did not produce enough noise to reach 40 dB and therefore they could not be set to run wit a noise level of 40 dB.