X-bit labs

Water-Cooling Systems Roundup: 16 Systems Reviewed! (page 4)

Category: Coolers

by Kirill "ALT-F13" Balalin

[ 03/30/2006 | 08:48 AM ]

Testing Methodology

The ambient temperature remained constant at +23°C during our tests. We used the same thermal compound for the CPU water-blocks of all the water-cooling systems included in this review.

The heater itself is a copper plate with a 35x35mm square in its top part the test sample is taking heat off from. The surface of the square is finely polished on a machine and finished. The heater is installed on a 10mm getinaks plate which has holes for all the main fastening standards (Sockets 939, 775, 478, 462, etc). The tested sample is secured with screws and special brackets.

The actual source of heat is a pair of 2SC 5200 transistors which are soldered up to the bottom side of the plate. The transistors are placed as close as possible to each other. The thermal field they create is shaped like two overlapping circles, the hottest point being in the center of the overlap. It is in this center that the temperature sensor is installed. Two transistors ensure very high combined power dissipation, yet we may be sure about reliability and stability of the heater’s parameters.

Thermal Ssensor Location and Heater Operation

Between the 2SC 5200 transistors and on the same line with their dies, a hole is milled right through the transistors’ cases. It goes into the copper sole by 4.5 millimeters and it is there that we insert a thermal sensor with thermal paste. After we make sure the sensor works fine, we fix it and fill the hole with thermal glue.

This design of the heater only approximates the heat dissipation of a real central processor, so it is incorrect to extend the temperature and power showings we get on our testbed to a specific CPU. The thermal sensor is not placed right in the heater’s core, unlike the sensor in a real CPU, so the temperature is going to be somewhat lower. This fact should be kept in mind when you are analyzing the performance measurement results. It also should not be wondered that some of the tested water-cooling kits managed to dissipate 300-400 watts of thermal power on our testbed. This is not an error or mistype. These are real numbers we got on our testbed since the generated heat energy is not concentrated in a certain point, but is dissipated from a larger, 35x35mm area.

The heater itself provides all the basic characteristics to carry out a correct comparative testing of different cooling systems. After this test and review the heater will be further improved to get closer to real CPUs in its parameters.

And here’s a description of the test procedure: we apply some thermal paste and secure the water-block on the heater. Then we make a trial measurement by choosing a thermal power of 100 watts and monitoring the core temperature. When the temperature stabilizes, we write down the value, shut down the testbed, reinstall the tested sample and make another trial measurement. If the stabilized temperatures are the same in both cases, the water-block is ready for tests proper. If the temperatures differ by more than 1°C, we make more reinstallations and trial tests until we get the real temperature at 100W thermal power. It usually takes no more than 3 reinstallations for the temperatures to coincide.

And then we perform the actual test. The measurement instruments tell us exactly how much energy is being dissipated as heat: the multiplication of current by voltage gives us the amount of power. The voltage and current are being constantly controlled and the instruments are connected in such a way as to ensure the highest accuracy of measurements in the circuit. There are no detachable connections; the power and measurement cables are soldered right to the instruments’ circuits. The power cables all have a triple operating current reserve.

The temperature-measuring tool was assembled in our labs. It consists of a data decryption unit (the data are received encoded from a Dallas Semiconductor thermometer) and units that process and output these data in a readable form on the display. The device also emits a command to reset the emergency protection relay installed in the power circuit of the high-voltage section. The protection threshold is set up from the keyboard and is written into the tool’s memory. We set it at 85°C for our tests: contemporary microprocessors (other than military or special-purpose ones) cannot be stable under so high a temperature.

However perfect a testbed may be, the results will be different on it than with actual processors. That’s why we took one of the latest and most expensive (at $65) air coolers, Zalman CNPS 9500LED, and used its results as a reference point. The cooler is equipped with an 80mm fan and performs well enough, producing a moderate noise at the maximum fan speed.

The performance of each water-cooling kit was compared against the performance of this cooler. The cooler’s performance graph is added into the performance graph of each tested kit.