Testbed

We use a specially designed testbed called Judge MARK 300 to explore various cooling systems. Using a field transistor in a metal casing as a heating element, we can accurately control the load on the cooling system. The testbed can send from 25 to 300 watts of power to the heating element. We read the temperature from a thermal sensor installed in the center of the element, and so we get data which are close to those of a real processor with the same heat dissipation. Our results are more accurate than those you would get with a real processor and mainboard as there’s no dependence on the CPU load (not all programs can load the CPU to the full, while we can select any wattage we need), and there’s no inaccuracy associated with the hardware monitoring systems of mainboards which are not very precise. You can easily see that by comparing mainboards from different manufacturers under the same conditions – the temperatures and voltages are going to be different. We are safeguarded from such errors and get data which are compatible among different cooling systems and also reflect the reality.

Test Conditions and Methods

The ambient temperature of 20°C is maintained artificially in the room. There’s no additional air cooling (save for the default air cooling of the reviewed cooling system). We use KPT-9 paste as a thermal interface between the heating element and the cooler.

The cooling system is mounted on the testbed and the temperature is read under zero load. Then the testbed is turned on to output a min load (25 watts), and the system works for 10 minutes before we write down the temperature data. Then we are increasing the load to 50, 75, 100, 125, 150, 175, 200, 225 and 250 watts every 10 minutes or we stop if the heating element becomes more than 80°C hot. We take the temperature data at the end of each 10-minute interval and then increase the load further – our experiments show that 10 minutes is enough for the temperature of the heating element to stabilize after a load growth of 25 watts. After 250W we return to the minimal load (25W) and measure the temperature after 1, 5 and 10 minutes more to check out the inertia of the system. After that the testbed (not the system!) is turned off and we make the last measurement of the temperature under zero load, a minute after we have stopped the testbed.

We also perform a max load test. We find the precise value of load when the system maintains a stable temperature of 80°C. Our requirement is the lack of deviations into either side for 10 minutes.

We took two inexpensive water cooling systems for the comparison’s sake. A $130 all-in-one Aucma CoolRiver system cools the CPU, graphics card and the North Bridge, keeping the noise at minimum. The recently released Titan TWC-A04 is curious for its effective and original design.