Dylan Hoen's Homepage
Electrical Engineer and Computer Programmer for hire in Victoria, BC, Canada
Main Page
Click Here
Things I Have Built
About Dylan

Things I Have Built
Title Page Camping Hot Tub Tall Bed With Storage Room Underneath
Portable Stareo Storage Bed / Media Couch Camperized Chevy Astro Van
LED Flashlight AUV Sonar Board BedWoofer
Computer Water-Cooling System Potato Cannon

Computer Water-Cooling System with Thermo Electric Cooler

Built: 2001/XX/XX
Added to webpage: 2010/07/19
During high school and early university, I often played first person shooter games against my friends, either at LAN parties, or over the internet. These games needed fast reaction time and accurate aiming. The faster the computer, the higher the frame rate, and the faster you could aim and shoot, the more likely you would win. We couldn't afford the fastest processors on the market, but we could buy midlevel ones and over-clock them. During a certain stable clock range, the voltage the CPU needs increases linearly with clock rate. The current increases linearly with both clock rate and voltage. When in the range where the voltage increases linearly with clock rate, the power consumed (and heat generated) by the CPU is proportional to the cube of the clock rate. Also, lowering the temperature of the CPU increases the maximum obtainable stable clock rate. Back in the year 2000, CPUs were just starting to produce large amounts of heat and just starting to need good heat sinks that could dissipate the heat. The heat sink manufacturers hadn't yet come up with the fancy heat sinks that we use today, so the heat sinks on the market were nowhere near as good as the ones for sale today. I built this water-cooling system as a better alternative to all of the air-cooling systems on the market, to get my computer to run as fast as possible, cool, fun project.

Below are images of the CPU water-cooling system that I made. On the CPU was a 1/4" copper heat spreader. A Peltier-Effect thermoelectric cooler was sandwiched between this heat spreader and a second 1/4" copper heat spreader that was part of a water block. An aquarium under-gravel filter power-head pump pumped water through this water block. The water then traveled through a car heater core (used as a radiator) and then, into a reservoir. The reservoir was there for water refilling, and as a place for air bubbles trapped in the pipes to finally escape. Two 120V fans blew air through the car heater core to cool it. I added a switch to electrically switch the 120mm fans between series and parallel. This switched the fansí RPM and noise levels from fairly loud to almost silent. The peltier thermoelectric cooler was a 13.8V, 156-watt unit. It was advertised as having a 68 degree temperature difference between the cold side and the hot side when 13.8V was applied across it and no heat was entering the cold side, and it was advertised as being capable of moving 168 watts of heat with a 0 degree temperature difference between the sides. The peltier consumed electricity and produced heat of its own, which the water-cooling system also had to dissipate. I hooked the peltier up to a switch that switches it between 12V and 5V. I ran the CPU and cooling system in a low power and silent mode most of the time, and turned everything up when I needed the processing power. At 12V, the peltier couldn't reach it's full potential, but gets fairly close. When not on the CPU, I measured a temperature of -20 degrees Celsius on the peltier, and, I think, 30 degrees Celsius in the water cooling system, for a difference of 50 degrees Celsius. I had an AMD Duron 600 MHz processer, which, by default, used probably around 1.6V. I over-clocked it to 1.017 GHz at 2.1V. This over-clock made the CPU produce almost 3 times as much heat as it originally did, and, at this heat dissipation rate, the peltier was almost neutral in effectiveness as the temperature difference between its sides was roughly equal to temperature increase of the water due to the extra heat that the peltier itself was creating. I decided that a safer over-clock was 850 MHz at 1.85V. At this power dissipation, the peltier could cool the CPU down to 5 degrees Celsius when the cooling system was on high, and, I think, less than 40 degrees Celsius when the cooling system was on low. I later learnt about Carnot efficiency, which is the highest theoretical efficiency of producing useful energy out of the heat flow between a hot object and a cold object, or, in reverse, consuming useful energy to move heat from a cold object to a hot object. I learned that large industrial refrigerator systems reach about 45% of carnot efficiency and peltiers reach about 5% of carnot efficiency. As the CPUs these days produce more and more heat, the days of using peltiers to aid in CPU cooling are over. Years later, I upgraded my computer and the CPU socket was in a different location, so rather than modify the watercooling system, I retired it and went back to a standard air cooler. I ended up using some parts from the water cooling system in an alcohol distilling system project I built a few years later. The pictures below were taken while I was disassembling the water-cooling system.

Below is an image of my old computer. The top modification is constructed of Plexiglas and contains the aquarium pump, the reservoir, the car heater core (used as a radiator), the 120V fans, and a few other modifications.

Below is a close-up of the top.

Below is a close-up of the front of the modification. Near the top are 2 red buttons, which were soldered to my sound card's volume control header. Below that is a headphone jack. Below that is the power switch, a peltier voltage switch, and the switch to switch the 120V fans between series and parallel. Below that, is an analog voltage dial to control the voltages of the 12V fans in the case, an LED readout for the 12V fan voltages, the power and hard drive access LED, a switch that switches which temperature sensor to read, and an LCD screen displaying the temperature of the sensor. There is one temperature sensor on the CPU, one in the case, and one in the water that can be connected to the LCD screen by the switch.

Below is an image of how the computer looked at night.

Below is an image of the CPU. Note that the actual chip is only about 1 square centimeter in surface area, and probably produced around 100 watts of heat when fully over-clocked. Also note the 2 holes in the motherboard on either side of the CPU socket with the white and orange wires coming out of them.

Below is an image of the water-block strapped to the CPU. Note the white and orange wires again. I used a pipe-clamp to tighten the wires to the appropriate tightness.

Below is an image of the underside of the water-block. Note the gap between the 2 heat spreaders where the peltier is sandwiched. Also note the 4 countersunk holes where nuts and bolts tightly clamp the 2 heat spreaders together. The upper part of the water-block is a cylinder of 1/8" walled copper pipe, which is soldered to the heat spreader, and a cap is soldered to the top. The water enters at a shallow angle, causing the water in the water-block whirlpool around, in theory causing turbulence to help the water remove heat from the copper.

Below is an image of the disassembling of the computer. At the back, right is the car heater core, and the black things to the left of it are the 120V fans that blow through it.

Below is a diagram of the analog fan speed control circuit and the 4 colored LEDs that denote the voltage used.

Below is an image of the parts layout on a circuit board with 1/10" grid spaced holes with solder pads. The parts layout differs from the schematic in that it was easier to connect a 300-ohm resister to +12v, rather than a 100-ohm resister to +5v.