THE HEATSINK GUIDE: Common mistakes made by reviewers testing heatsinks

In the past years, websites about cooling have been growing like mushrooms. Obviously, the quality of the heatsink reviews found there varies a lot. It appears that some webmasters seem to think that all it takes for a heatsink review is a PC, a digital camera, and a heatsink.

So which review can you trust? This page is supposed to help you decide whether a cooler review is a good one, by showing the most common mistakes made by incompetent reviewers.

Mistake 1: CPU's thermal sensor is used for temperature measurements

The thermal sensors of CPUs are way too inaccurate for proper heatsink benchmarking. A good thermocouple thermometer should be used, such as, for example, the Omega HH23. The thermocouple must be properly placed; for CPUs with a heat spreader (all recent Intel CPUs, as well as Athlon 64 and similar CPUs), it is a good idea to glue it to the side of the heat spreader, with the aid of thermoconductive glue. If the CPU doesn't have a heat spreader (Socket A CPUs), it is quite hard to place the sensor properly; for testing coolers for such CPUs, it is better to use a simulator device instead of a real CPU (read below for details).

Mistake 2: Improper consideration of coolers with temperature-controlled fans

If the heatsink to be tested features a temperature-controlled fan, both its cooling performance and its noise level will obviously depend on the thermal load. A single performance/noise measurement will therefore only show the cooler's performance under one specific load condition; this is not enough for judging the cooler! For temperature-controlled coolers, coolers with heat pipes, and peltier-based coolers, multiple measurements under different thermal load conditions must be made.

Temperature-controlled fans can usually be forced into maximum fan speed by short-circuiting the thermal probe (if it is accessible). This way, at least the maximum cooling performance and the maximum noise level can be measured.

It is better to use a simulator device instead of a real CPU, since it is much easier to control its power usage - this allows heatsink benchmarking at different power loads. Especially for testing heatsinks for CPUs which do not have a heat spreader, a simulator device allows much more precise measurements.

Have a look at the simulator device I built and used for my heatsink reviews on AnandTech.com, for example. This setup was good at the time; but it is not suitable for simulating today's high-power CPUs, since a single foil heater cannot provide enough power. One solution is to use multiple foil heaters, as Joe from overclockers.com does with his simulator device. Another option is to use (cheaper) power resistors instead of foil heaters.

Simulators for testing heatsinks are also available commercially, for example from InnovaTek. Tom's Hardware Guide used an Innovatek simulator for some of their older heatsink reviews.

In any case, the shape of the simulator must be identical to the shape of the CPU it simulates, and it must be possible to mount heatsinks in the same way they would be mounted on a real CPU/motherboard combination.

Mistake 3: Lack of comparison data

There's no point in reviewing a single heatsink, if no other coolers are tested using the same test methodology for comparison.

Mistake 4: Lack of noise measurements

What's perceived as "loud" and what's "quiet" can be a rather subjective thing, opinions may vary. Therefore, reviewers should provide some hard data, in the form of noise level measurements. Ideally, such measurements should be done in an anechoic room, as it is found in some recording studios or universities. Renting such a room is very expensive, and I haven't seen any reviewers on the web who actually did this. A cheaper compromise is to build a sound-proof box, such as the one I used for AnandTech reviews. FrostyTech and Tom's Hardware Guide use a similar setup.
While this setup does the job, it is far from perfect. The dB(A) level, as it can be measured using cheap sound level meters, has a nonlinear scale. Values are hard to interpret. Sound level should rather be measured in Sone, which takes into account human noise perception: A Sone value that is twice as high will be perceived as twice as loud. c't Magazin, for example, provides Sone measurements in their cooler reviews.

Conclusion

Testing heatsinks is a non-trivial task; not everybody does it well. Read reviews critically, use common sense, and decide for yourself whom you trust!

All pages copyright © 1997-2010 Tillmann Steinbrecher
Legal information / Disclaimer / Impressum