There are more than two months from my last coolers roundup. I know that I promised then a bigger cooler roundup. Well, I will try today to do my best and meet your requirements. I want to give a big thanks to CrackMaster who helped me and who can be considered a coauthor of this article.
I received many emails from people who wanted to share with me their experiences about coolers or to suggest me improvements to the testing method. I would like to give them a big thanks too. I changed several things to my testing methodology and I would like first to discuss these changes.
Many users asked me why I didn't use a CPU simulator with a more stable and measurable wattage load than a real CPU. Most people know that's impossible to predict the actual power used by a CPU, because the specs talk only about maximum values and not about averages. The main reason why I didn't use a CPU simulator was because I want the tests to be as close as possible to real world, and a heater can not predict a real world behavior.
There are several things hard to simulate:

• In real world a CPU dissipates power in motherboard socket using the pins
• A real CPU surface is hard to simulate by other materials or finishing
• The size of a real CPU can't be reproduced very easily. The die surface is not everything, the height is also important because every cooler applies a certain pressure on the CPU surface which is influenced by the vertical dimensions.
• The load versus time curve is impossible to be measured using a regular simulator because the CPU power dissipation is not deterministic. With a real CPU we do not know anything about the power load, but we can see how the cooling system is able to handle the increase heat dissipated by the CPU core.

Do not understand me wrong. I don't have anything against CPU simulators. These devices are much better for testing a cooling system performance than any CPU, but because we couldn't build one able to face our standards and pretensions we preferred to go for a real CPU. But we tried to improve the test as much as possible and the following decisions have been taken.

Tests were made on a motherboard having a thermistor able to touch the CPU back. We increased the thermal transfer using thermal compound. It's a common problem that inside the socket thermistors are very depended to the air flow so we tried to limit it as much as we were able to using teflon wool. We decided to use MSI K7T Turbo Limited because in our tests with this motherboard the results proved to be the most close to our preliminary tests conducted with a Taisol cooler and an on die Omega placed sensor. AMD documentation recommends the drill method for measuring die temperature, a very precise method, but unfortunately we wouldn't have been able to drill the required holes on all coolers in this roundup and to measure using the Omega probe.
So that all the measurements are made with this board and PC Alert III software from MSI under Windows 2000 SP2. The system was stressed using Passmark BurnIn Test. Different motherboards display different results because manufacturers use a mathematical method to approximate the sensor temperature. The method may not be the same, the sensor may not be of the same model and finally the calibration may differ. Certain motherboards are known to provide optimistic results while others are very pessimistic.
Tests were run three times in different days. The results are an average of the day results.
The temperature was recorded using software until it became constant for more than 20 minutes. We measured the temperature with a 100% load applied to the CPU.
Ambient temperature was measured in all cases. After several tests we discovered that the influence of the ambient temperature was almost linear to the CPU temperature, so we averaged it too.
All the tests were made in an open case environment, with no forced air flow around the heatsink. We have several Zalman coolers in this roundup which required a case, all these measurements were made in a regular case with no forced air flow around.
The noise level was measured inside a sound isolated case using a Super Mini Sound Meter 840014 at a distance of 30cm.
All coolers were installed conforming with our thermal compound installation using Artic Silver. Although several coolers were provided with other interface material we wiped it and tested them with the same thermal interface like the others.