No announcement yet.

Home-Grown Water Cooling Setup

  • Filter
  • Time
  • Show
Clear All
new posts

  • Home-Grown Water Cooling Setup

    A friend of mine recently constructed a pretty much custom-built water cooling setup for his dual processors. With the exception of the Swiftech waterblocks, pump, and a few other things he basically constructed this out of various parts found around his house.

    Below is his description and a few of the pictures. He did a very nice job with the setup.

    A good bit of the cooling gadgetry is from Swiftech but the rest is my hackery. The main bit of hackery is the heat exchanger which is usually hiding under the work bench (behind a trash receptacle). In examining the HX, you will see that I am more concerned with function than physical beauty

    The reservoir is a chunk of 4" sewer pipe (leftover from the repair of a root intrusion into a septic inlet). The system holds about two gallons of coolant (50% Prestone anti-freeze, 50% distilled H2O). The blue anodized box has two fans that pull cool air in from the floor into the box and out the top of the box and thus through the radiator (the box was left over from an unrelated car project). The box serves as a plenum which insures that air flows through all of the radiator roughly equally.

    Note that the outlet of the pump incorporates a tee with a normally shut valve. It is not evident but just after the tee is another valve at the inlet to the radiator that is normally open. These two valves, when reversed, make draining/purging the system easy - the pump essentially drains the system through the tee. I have a container to catch the coolant. When the maintenance is done, I pour the coolant back into the system through the cap at the top of the reservoir. Because of the height of the reservoir, the system never has a problem priming. I am pleased to report that I can move the boxes quite a lot and not cause leaks. I may cart it off to the next lan party.
    All of the fittings are either Pex or Watts (from Lowes) and are designed for push fit assembly. The design is such that pressure helps hold the cables in place (like the infamous Chinese finger trap). I have a few pictures of the fittings in various stages of disassembly.

    At present the CPU0 is running at 23C and CPU1 is at 19C. When CPU 0 and 1 are equally busy they converge to the same temp and have not gotten to more than 36C. The video card is usually reports about 53C and never above over 61C. The GPU is the first element that the incoming coolant hits. I have not found over-clocking the GPU to help the frame rate. I will eventually get the new Swiftech CPU heat sinks. I am not touching the video heat sink though.

    Note that I have installed a sound blaster card in the slot adjacent to the 8800GTX. I can install a second 880GTX for SLI (thus sandwiching the SB card between two videos) though I doubt I will.

    The fans are running under the control of a left-over thermostat that switches them between 7v (12v-5v actually) and 12 volts. They only run at 12v on hot summer afternoon. It is expensive to keep the room cool via HVAC (and besides, I like to be warm) which is the motivation for water cooling. On a cold day, my computer dungeon is the nicest part of the house to be in.

    Last edited by Michael; 01-23-2007, 05:19 PM.
    Michael Larabel

  • #2
    Master of Disaster

    I'd like to thank Michael for posting these images. I would like to clarify that the fans controlled by the thermostat are those within the plenum chamber that the radiator is affixed to. Being a pack-rat, I had a thermostat left over from a home-repair project. I created a circuit that use the thermostat and a diode to switch the fan from 12v to 7v according to the ambient temperature in the area of the heat exchanger (the big gizmo with radiator, plenum and reservoir).

    The fans in the system itself are 4-wire and operated by the system. These fans are: two 38mmx120mm x 120mm x 38 and two 40mm x 40mm x 25mm. Normally these fans are very high RPM and consequently noisy but my board (Tyan s2895) has controls that manages them according to temperature. Accordingly, the fans normally run at a small fraction of their rated speeds and (apparently) thick fans move lots of air quietly even when spinning slowly. The four wires of the fans are +12v, gnd, tach_out and speed_control_in. The first three wires are fairly standard but the fourth wire, the speed control input is a bit unique.

    It is an input to the fan and supplied by the system. It is intended to be a fairly high frequency pulse width modulated square wave (not a slowly varying DC voltage level). The system modulates the duty cycle of the square wave and in so doing controls the speed of the fan. One reason that the frequency of the square wave needs to be high is so that it won't cause a hum or other audible sound in the process of controlling the fans (clever, IMO). I got these fans at Skyline Graphics of Dallas Texas (they are fan specialists and have all sorts of wild and wonderful things).

    I am too lazy to create a schematic but the idea for the thermostatic control of the plenum fans is this:

    The red wire of the fans (usually +12V) is connected to a +12V supply (cabled over from the system). The black fan wires (usually ground) are connected to a junction consisting of one terminal of the normally open thermostatic switch and the "arrow" side of a diode (get'em from Fry's for under a buck, a one amp rating should suffice). The "line" side of the diode is connected to a +5V supply (cabled over from the system). The remaining terminal of the thermostatic normally open switch is connected to ground.

    It is important that arrow on the diode point toward the +5v supply. If the diode is placed to conduct current from +5v to ground, it will be a short. If you are lucky the power supply will safely shut down. If you are not lucky, the MagicSmoke may escape the power supply (and we all know, that if the MagicSmoke escapes from a component, the component will no longer work).

    When the +12v and +5v are applied, current flows from the +12v supply through the fan then through the diode and into the +5v cable connection. This current flow is allowed because the diode "arrow" points at the +5v cable connection (i.e current flows from plus to minus and the diode is oriented to point in the direction of current flow). Now the tricky bit is that when the temperature rises enough to close the normally open thermostatic switch, the current instead flows from +12v through the fan, through the now closed switch and finally to ground (because it is the path of least resistance).

    So this way the plenum fans run at either slightly less than 7 volts (just barely on, but still flowing air) or they are running at full speed according to the setting on the thermostat. Typically I set the fans to come on at about 95DF. This explains the use of the yellow cable bundled with the coolant hoses.

    P.S. If you really want to know the gory details, this setup works because the 5v supply starts flowing current to its load before the fan starts flowing a significant amount of current. A fan being a primarily inductive load, resists changes to current flow. In particular when the fan is first powered up it resists the flow of any current at all. The current ramps up slowly which is good because this gives the 5v supply enough time to establish a current flow in its load.

    If not for this effect, the current through the fan would try to flow into the +5v power supply rail. Most power supplies will interpret this a a BadThing and shut down. But since the +5v supply is flowing current by the time the fan is up speed the current that flows out of the fan actually serves to provide a tiny fraction of that required by the load from the +5v power supply rail . The resultant effect is that +5v supply slightly backs off the current it is supplying to insure that the +5v voltage is within the +5v supply regulation limits. The +12v is doing a tiny bit more work and the +5v is doing a tiny bit less work.

    If all of this is unclear, trust me it, works with any decent power supply. Just be sure that the diode is not biased to flow current from +5v to ground (else kaboom). If you try this and something breaks, I'll try to help you out but keep in mind that it is an advanced technique that you should not try if you don't understand it