The Test

The first test was done with a 43 watt heat load - this corresponds to the typical power usage of a 1 GHz Palomino CPU. The temperature readings given here is the temperature difference between core and ambient temperature. As usual, lower temperature readings are better.

(1) the Miprocool was tested with the temperature controlled fan at maximum speed

Since in the 43W test, the top-performing coolers are very close to each other, we also publish the results of a test with a higher heat load - here, performance differences become more obvious. The second test (with the best-performing coolers from the first test) was done with a  61W heat load. This corresponds to the typical power usage of a 1.3GHz Thunderbird CPU.

Here are the results of the 61W test; remember, lower temperatures are better:

(2) from the March 01 cooler roundup

Our "simulator" would not allow testing of the less efficient coolers in this roundup at 61W, since the foil heater used would overheat. We are already working on a new simulator device that allows cooler testing with a heat load of 100-150W, to represent the cooling requirements of future CPUs. Details about this will be published before the next roundup, in a new "testing methodology" article.

By now, the AMD Athlon MP (Palomino) with its internal diode for temperature measurements has become available. However, the temperature readings taken from that diode aren't accurate enough for a meaningful performance test. If you look at how small the performance differences between some cooler models in this test are, it will become obvious that a temperature measurement with a resolution of 1°C isn't accurate enough. However, the fact that now an AMD CPU with on-die temperature measurement is available has given us the opportunity to check whether our "simulator" represents the real-world situation accurately. The difference between the C/W values calculated from our test data, and the C/W values calculated from test data obtained with the Palomino's internal diode is around 10% (based on the Silverado test data published in c't magazine, issue 17/01).

You may be wondering why our test results are in some cases (e.g. the Silverado) so much different from the test results published on other websites. We are aware of this, but there's nothing we can do about it. We've double and triple-checked our results. The only advice we can give is: Check which testing methodology is used for that particular test, and decide for yourself which review to believe. Information about our current testing methodology can be found here.

Index Noise measurements
Comments Locked


View All Comments

Log in

Don't have an account? Sign up now