Design of Experiment

Variables

Experimental Design

To test these factors, a standard PC benchmarking setup is required. An open-air test bench (such as the DimasTech one shown to the right) should be used to minimize case impact on experimental results. Temperature should be recorded via the CPU's sensors. Additionally, a thermometer should be placed elsewhere in the same room a sufficient distance from the test bench to read the ambient temperature.

To run the experiment a standard CPU benchmark such as Cinebench should be used. A common 4th generation CPU for benchmarking is the i7-4770k (kept at base clock). Since the benchmark stresses the CPU for a duration of time and an average across this time is taken, 2 trials should be enough in this scenario. After running the benchmark at each combination, allowing the temperature to drop completely to idle in between trials, and recording average temperatures, the data can be analyzed as below.

Hypothetical Experiment

To test the impact of these changes on the cooler, the following variables should be targeted for a 2 factorial design experiment:

Design Variables:

Fan type
Two different types of PC case fans are commonly available. Designated "airflow" and "static pressure", the two varieties vary only slightly in the shape and number of blades. I will refer to static pressure as the low condition and airflow as the high condition, following the relative number of blades on each.
Heatsink orientation
In swapping out a new fan the optimal heatsink structure may change. Since I want to aim for the same size footprint, the main factor left here is the orientation of the heatsink fins. This could either be radial as in the original design (and in my CAD changes) or parallel vertical fins. I will refer to these as the low and high conditions respectively.

Performance Variables:

Temperature under load (°C above ambient temperature)
Obviously, the most important factor to measure the performance of a cooler is its capability to disperse heat from a concentrated area.