Cooling Capacity
Whether you have an air-, water- or low-temperature cooling system, the critical factor in determining the quality of the system is its ability to rid the CPU core of excess heat as quickly as possible. In technical jargon, this is known as the thermal resistance of the cooler and is measured in degrees Kelvin per Watt (K/W). As a rule, the smaller the thermal resistance, the more quickly heat can be dissipated from the hot CPU to the cooler surroundings. The crucial advantage in determining the thermal resistance of each cooler, as opposed to simply measuring the temperature of the CPU, is that individual coolers can be directly compared to each other. The CPU temperature alone is not sufficient as an assessment criterion, because the reading is dependent on the surrounding temperature, according to the laws of physics .
To calculate thermal resistance you first need to measure the CPU's power consumption. To measure this value while the PC is running, we modified a number of motherboards, one for each platform (CPU socket).
For most platforms, we used the CPU-integrated thermal diode, which allowed us to measure the precise die temperature. An exception was Socket 478; here, we modified the CPU so that the temperature could be measured using an external temperature gauge. To this end, the CPU's heat spreader was manipulated so that a small thermometer could be inserted into it.
To make sure that the CPU was operating at full load, we used the program Prime95 , which was originally created for calculating prime numbers. Because such calculations are very CPU intensive, the program is perfect for putting heavy loads on a processor. The stress put on a processor by Prime95 produces the highest levels of processor heat possible using standard applications. (Special tools from chip manufacturers create an additional 10-20% more load, but since these programs aren't available to the general public, using them in our tests would have produced results that were not representative of real-world experience.)
We used the thermal compound or pad that came with each cooler for the measurements. If neither one was included, we used standard silicon thermal compound for the test.
On coolers where the speed of the fan could be adjusted, we did one test at the highest speed and one at the lowest setting. In the event of real, repeatable variations between these settings - for example, if the supply voltage could be changed from 5 to 7 or 12 volts - these variations were also taken into account.
