Originally Posted by Graystar
Well, itâ€™s an interesting report but in my opinion its got some serious problems. The first is that the amount of energy transferred increased as flow increased. This should not happen. At equilibrium the amount of energy transferred into the water should be the same as the amount used by the processor. In fact, some of the well-known testers check their equipment by ensuring that the energy transferred is the same under all flow rates.
Another problem is the fact that at 206 Joules the processor would have to be operating at 206 watts. It is already known that voltage and clock increases do not operate on power consumption in a linear fashion. The effects on power consumption are actually far less. Still, if you applied your 0.15 V and 1.27Ghz overclock increases linearly, then max power consumption would have risen to 128 watts. Dissipating 206 watts from a 128-watt processor is a cause for concern. Even pump heat can't account for *that* much additional energy!
My guess is that the problem lies with your temperature sensors. With such small changes in water temperature, you need sensors that are accurate to a few hundreds of a degree. Sensors like that are expensive, as is the equipment that reads them.
The range of your flow rates is where the greatest gains are experienced. This is because of increased turbulence in the radiator. But this effect starts to top out at about 6 LPM, depending on the radiator. Beyond that, increases in flow rate bring zero returns.
If you could accurately test a range of 1 â€“ 15 LPM, while insuring a consistent energy input level, that would be really interesting.
I am not sure how you figure the energy absorbed by the water will be constant. That means we could insert a pump that moves the water at a trickle and still have the same cooling performance? I don't think so. Increase flow will increase energy transfer, this is hypothesized by theoretical physics, and confirmed by my experiment. Look at the temperature of the CPU lower as heat transfer is increased. This confirms that more heat was removed!
I understand that there are high levels of uncertainties with the temperature sensors. I used what I had to work with. I also see the limit of the experiment from the limited flow rate range. This is all noted and explored in the paper. However, even with the lack of precision in the temperature sensors, there will still be a trend.
The published thermal output of a processor is low at best. Everyone knows that at stock settings processors easily put out more heat than published. When you pump up the voltage, increase the FSB, and torture the L2 memory, power consumption is going to be rather high.
I did not find any effect of flow rate on the radiator. The input temperature for the water stayed fairly consistent. This means the radiator was pretty much returning the water to the same temperatures at all times. That actually puts higher heat transfer at low flow rates on the radiator (which makes sense, the longer the water spends in the radiator, the more heat removed).