Literally – Michael Hughes
This water-cooling project was the result of two months worth of thought and another month to build. This project came about as the result of:
- The old cooling system was running an Intel 2.53 GHz P4 with the factory HSF, which worked fine most of the time. The system’s idle processor temperatures were in the upper 40’s C. However, the load temperatures for the CPU were in the 50’s and lower 60’s, and that is on the hot side for these components.
- The computer is in a small room with another two computers. In winter, we don’t have to turn the heater on the room. However, in the summer it was akin to sitting in the vent of a hot oven.
- Finally, NOISE, NOISE, NOISE! The computer had 4 active fans in it. As a result, it could be heard from other rooms. This is especially true when compared with another computer in the room, which is an OEM PC having the redeeming quality of being very quiet.
Several notes before starting:
- Initially there were no plans to write up the project, so there are no pictures of the early stages of it.
- This project was not cheap; because of the very high quality and high performance parts used, it ran around $400 – $450.
- The list of components in the computer:
- Intel P4 2.53 Ghz
- Intel 845PEBT2
- Corsair XMS PC2700 512 Mb memory
- ATI Radeon 9700PRO
- Sound Blaster Audigy MP3
- Maxtor 7200 rpm 40 Gb ATA100 hard drive
- HP CD-writer12
- Antec TruePower 380W
- Cheap $30 no-name case that I bought off Ebay
The case with a cheap no-name power supply that was DOA and was replaced with the Antec TruePower. Sorry, there are no pictures of the inside!
The original plan was to do an in-case cooling solution, but this did not solve the heat or noise problem inside the room. It was suggested that the heat be vented outside; this was surprisingly convenient because the room in which the computer is placed is situated on top of an overhang to the exterior of the house.
But, the cooling system now required an extra design piece, a sort of plate. The plate is a wooden board that would hold the components close to the bottom of the overhang. However, the plate had electronics on it, so it needed to be put somewhere so that no liquid water (rain) could get to the connections on the pump and fans.
This resulted in the decision to place the plate behind a pair of large columns supporting the overhang. Also, because of the weight on the board, it would have to be attached in a very secure manner. Luckily, floor joists run across the width of the overhang so that the board could be mounted onto two separate beams.
As you can see, all the electrics of the fan & pump will be shielded from the weather by the overhang it self and the two columns.
The plate was the most important part of the system because it is where the pump, radiator, boiler drain and thermostatic mixing valve are located.
A thermostatic mixing valve mixes hot and cold water to get a very stable and constant water temperature. Basically, inside it there are two pieces of metal that, at different temperatures change size, thus modulating how much cold and hot water gets through. These metals are then set in relation to the hot & cold inputs via a control knob. The point of the valve in this cooling system is to avoid condensation on the water blocks when the temperature outside the house goes below the dew point of the interior of the house.
The size of the plate needed to be 1′ by 2′ by ¾”. These dimensions were required to give adequate room while not having a behemoth beneath the overhang. Those sizes are about 30½ cm by 61 cm by 19 mm.
The main problem in designing the plate was how the piping would be run, as the pump needed to be inline before the radiator and also before a “T” junction which would send coolant to the boiler drain and the mixing valve. The mixing valve also needed to be on the outlet side of the radiator. Finally, both the outlet side of the mixing valve and input side of the pump needed to have a clear shot to the ceiling. The resulting generic design:
Another consideration in the design process of the plate was how air would be moved through the radiator. The width of the radiator dictated that two 120 mm fans would be used to push air though it. Also, a wooden shroud was needed on the bottom of the board because there was no easy way to mount the fans on the radiator – having a shroud is more efficient that direct mounting.
The cooling system needed a reservoir in it to have an easy way to fill it, check the level and something to act as an air trap. Some other design parameters about the reservoir is that it had to have a very good seal and it must be easily disconnected from the rest of the system. (More about this later in construction, Part III Mark I.)
This cooling system also needed to cool the three major heat sources inside a PC; the chipset, CPU and GPU. For the system to be relatively easy to setup and to be easy to disconnect, the three water blocks are connected in series as follows: An input pipe goes to one block, out of that block into an output line which hooks into the input side of the next block, etc.
This is a diagram of the final routing design inside the computer:
The reason why the chipset cooler is first is that the greater the temperature difference between the coolant and the item being cooled, the better the heat transfer. To put this into full use, the CPU should be first in the loop.
However, due to the requirements of the design, the entire computer must be cooled effectively. Since the chipset puts out the least amount of heat, it goes first, then the GPU. The coolant is slightly warmer from the chipset by this time, but the GPU is also much hotter, thus maintaining that all important temperature differential. The even warmer fluid is then passed onto the CPU, and then finally passed out of the system.
There are no flow direction arrows in this diagram,
so just keep in mind that cold coolant flows from the pump plate to the computer. It then exits the computer, goes to the reservoir, and finally the hot coolant returns to the pump plate to be cooled.
To stop the distilled water from freezing (it often goes below freezing outside in the winter) and to stop any biological growths (algae) inside the pipes, something must be added to the water. The best choice turned out to be a type of alcohol: Isopropyl, also known as rubbing alcohol, to act as both a biocide and an anti-freeze. An 80% water to 20% isopropyl mix was used to get the desired “anti-freeze” properties.
In addition to that, 5% percent Water Wetter was also added, resulting in a total system mixture of 75% water, 20% isopropyl, and 5% Water Wetter. One might ponder whether there would be any problems with evaporation of the isopropyl; however you have to keep in mind that this mixture has less alcohol in it than 80 proof Vodka (40% alcohol) which does not evaporate in a bar. The system is also not open to the air, because the only part that could be open (reservoir) is sealed using a cap and rubber seal.
This article will not explain how each part was arrived upon, just what they are and how they were used:
I. Water Blocks
a. CPU – Swiftech MCW5000-P: $50
b. GPU – Swiftech MCW50: $36
c. Chipset – Koolance CHC-A01 (The stock ¼” connections on the water block became a problem later): $24
a. New, Ford Econoline Van heatercore: $43
a. Taco cartridge circulator $80
This is a very strong pump designed for under floor heating systems. There could be 10 computers on the loop and it would not approach its heat limit. But the main point in using it is that it’s designed to run 24/7 for years on end with no maintenance – in other words, it is very reliable.
IV. Thermostatic Mixing Valve
a. Watts mixing valve: $80
a. Cheap 1 liter plastic container with a sealing lid from Target: $5
a. Tygon R-3603 for inside the computer (3 feet): $7
b. A large quantity of polyethylene tubing (around 20 feet): about $16
Note: Polyethylene is not vinyl. Polyethylene is a stiff walled tubing that does not rot, reacts to very few chemicals, is non-reactive to UV light and can withstand quiet a beating. For most of the system, this tubing was used because of things like wood and other pressure points that might collapse other tubing.
c. Some vinyl tubing for flexibility in short lengths: $3 – $5
a. There were so many visits to the nearest specialty hardware store (the major chain stores did not have a sufficiently wide selection of fittings) that no record of cost was kept, but because some of the fittings were expensive: The total was about $50 – $60 all told.
a. Two Mechatronics 126 CFM, 46 Db, 128 mm fans: $14
a. Extra Stuff: about $20
$427 – $440
The first order of business was to build the pump plate. First, the outline of the radiator was marked out and a hole cut so that the radiator rested on the nodes on either side of it.
Even though some people attach the fans or fan directly to their radiator, this is not the optimal arrangement. For the best performance, you need a shroud. So on the under side of the plate, a wooden shroud was constructed with a cover plate to which the fans would be attached.
Another consideration on the pump plate is that the pump had to be mounted level. However, the bottom of the pump was not level – this required a small wedge under the round end of the cartridge and a couple of steel “L” brackets attached to the cast iron housing to provide support (the pump by itself weighed in around 10 – 12 pounds or around 4 – 5 kg).
The only other part that required modification to the plate was the drain, which needed poke through the bottom of the board giving easy access to it. To do this, the piping was loosely assembled, the drain hole marked and then drilled out. The board was suspended using twelve feet of 1/8 inch thickness angle aluminum. Two feet of it was attached to each side of the board.
Four one foot pieces of the angle aluminum were then attached. In the photograph, two of the four vertical supports are shown, the one on the left using a lag bolt style screw put into the wood. The vertical bar on the right is bolted into the end of the side bar which extends beyond the board.
On top of vertical bar there is another six inch long piece of angle aluminum, set so that one of the flat sides lies even with the overhang ceiling. This flat side then had two holes drilled in it, giving a total of eight lag bolts holding the plate to the top of the overhang.
The last part was to attach the fans. This would be done using eight (four per fan) 2½” bolts. The fans already had the necessary holes, so that only eight holes had to be drilled in the cover plate. The two fans were then bolted in place.
Here it is all done!
Six holes needed to be drilled to make it all work. Two holes in the interior for the tubes, two outside in the overhang, plus another two (one inside & one outside) for the power cable. The four hose holes need to be just larger than 5/8″ to accommodate the 5/8″ OD polyethylene tube that was used. The holes for the electric cable were made the same size so that the cable could be run into a plug or power box.
After the holes were drilled, a fish tape was run through one hole, attached to the interior of a piece of tubing and carefully pulled back though. This procedure was then repeated for the other tube, and then the power cable.
The “power cable” is actually a 16-guage three prong extension cable bought at a local store. This cable is then plugged into a dedicated power strip for the PC, monitor, pump and fans. This cable runs into a dual socket electrical box. The pump and fans were then plugged into the electrical box using a cut to length power cable for the pump, and a transformer for the fans.
The inside view:
The reservoir is an incredibly important part of the system, as it acts as a fill point, air trap and level check. Due to the fact that this is on the inside over white carpet, the reservoir’s seals needed to be perfect. To get a good seal, two nylon compression-to-compression fittings were taken. A band saw was then used to cut one of the compression nuts into two pieces.
Once that was done, the screw part of the compression fitting was inserted it into a hole already drilled in the reservoir. A washer was then placed on both sides of the plastic walls. One half of the compression nut was then used to sandwich the rubber washers on both sides of the plastic walls. This procedure was then repeated for the other side of the reservoir. This resulted in an easy to disconnect fitting on both sides of the reservoir.
This picture is not the finished product, but it gives the general idea of the washer sandwich:
As said before, the reservoir needed fittings that enabled it to be totally modular. Thus instead of using barbs, nylon compression fittings were used. The nylon compression fittings used here hooked up to polyethylene piping – the advantage over barbs is that to make a connection, the pipe is inserted in the compression sleeve. To seal the connection, the compression nut is then just screwed onto the threads, locking the pipe in place. This allows the connection to be reusable with cutting the pipe or anything.
The problem with the seals on the reservoir is that they do not take well to stress and tend to leak if they are pushed one way or another. However, because of the 90 degree elbow out of the left side of the reservoir, nothing could grab the connection to steady it. Thus that connection was changed to a straight fitting which then connected to an elbow.
In addition to steadying the reservoir, the holder also needed to carry its weight. Also, the holder needed to match the look of the desk that it was going to be put underneath. This resulted in a design using cherry-veneered chipboard, with a bottom board to support the weight of the reservoir. The bottom board then had two boards with U’s cut in them attached to each side.
The reservoir needed to be the highest point within the system to trap air bubbles. The top of the computer case was measured and that set as the bottom of the reservoir. The reservoir holder was then attached to the wall with L brackets using eight screws total (four in the reservoir holder & four in the wall). Here is what it looks like with the reservoir holder finished and attached to the wall:
On the Radeon 9700PRO, the heatsink is held in place by two spring loaded clips that run though the PCB to the other side, where they expand so that the clips can not fall back out.
A standard size wire cutter was used to carefully clip away the little knobs on the back of the board. Once this was done, the heatsink come off the chip with a small amount of pulling. A thin, flexible and very sharp hobby knife was then used to slowly clean away the yellow thermal-tape that was stuck to the chip. Once this was done, the clean chip looked like this:
The next thing that was done was the attachment of the new block. First a thin layer of Artic Alumina was spread on the chip to help get a good thermal contact. The block was then attached to the graphics board using the provided screws.
The CPU block was the simplest to get off for two reasons:
- The attachment method is reusable and therefore is designed to be easily removed.
- It was assembled by hand and therefore came apart by hand. First, the pressure was released off the stock HSF and then it was pulled straight off.
The nice thing about Intel’s stock HSF is that the contact pad is a thin piece of aluminum with a very thin layer of tarnished silver. This was very easy to clean up compared to the GPU; all that was needed was a paper towel and it rubbed right off the integrated heat spreader. This gave the look of an almost new CPU:
Attaching the new the waterblock was also the easiest of the three. First, another thin layer of Arctic Alumina was spread on the chip and then the provided clips were used to hold it in place.
This was the hardest part of the electronic modifications by far. Intel has an interesting solution to stop you from modifying anything other than the HSF on the CPU, specifically the chipset heatsink. It is riveted in place – there were four small aluminum rods that held down the heatsink. These then went through the PCB and were soldered right into the motherboard.
To fix this, the rivets on the heatsink had to be carefully drilled out. First a piece of cellophane wrap was put around the heatsink to catch the metals shavings, avoiding any possible short circuits in the motherboard. A 1/16″ cobalt drill was then used to slowly cut away at the rivets. Once the heatsink was off and with minimal cleaning, the Intel chipset looked like this:
To attach the new block, Arctic Alumina thermal adhesive (two-part epoxy) was used. Howeve, the small die size raised concerns about the small contact patch and the lateral stress the coolant tubes would put on this connection. To solve this problem, a 5 minute (two-part) epoxy was used on all four corners of the chipset to give it the extra strength needed.
First, a hole was drilled in the fan grid for the input on the back of the computer case using a hardened steel hole saw. Next, the hole was finished and smoothed using a multi-purpose file. Finally, a male threaded barb fitting screwed into the female to male connecter locked it all into place. This step was then repeated for the output connection in the case.
On the output side of the computer, a 90-degree elbow for the turn up to the reservoir was used. On the input side, a 45 degree street elbow was used to reduce the possibility of kinking the pipe coming up from the hole in the floor. After this was done, Tygon tubing was used to connect all the waterblocks inside the case.
The chipset waterblock really did cause a problem at that stage because of its 1/4″ ID connections. They in fact required two complicated converters to change the ID size from 1/4″ to 3/8″ of an inch. The input and output of the block go into a short 1/4″ vinyl section which goes into an 90 degree elbow, which in turn goes into a piece of 1/4″ tube. This in turn hooks into a 1/4″ ID barb which converts up to the 3/8″ ID tubing size used for the inside of the computer.
First, behind the computer two vinyl hose sections were added to give some extra flexibility in the placement of the case:
After the vinyl sections were in place, the hoses to and from the pump plate were attached. Finally, the reservoir was put in place and hooked up to the rest of the system. Now, with the hoses attached, the system was filled with distilled water via the reservoir. However, the pump could not be used to pull coolant around the system because of the way the air locked into it.
This meant using the old fashioned method of pulling liquid through pipes – sucking yummy tasting water! For those of you who do not know what this means, you pull coolant around the cooling system by sucking on a pipe like you would with a straw. Of course, being careful all the time not to get any of it into your mouth.
Once that was done, the final hose that I had been sucking was hooked up to the input side of the reservoir. The pump was then turned on. The entire cooling system was left running in this state for a couple of days to fully test everything.
There was 8 cups (64 oz.) or around 2 liters of liquid in the system. Since 20% alcohol was desired, this meant 12.8 oz of pure alcohol (20% of 64 oz). To get the amount of Isopropyl (70% alcohol & 30% water) needed, 12.8 oz. was divided by 70%. This gave the approximate amount of Isopropyl needed by the system – 18 oz.
The amount of amount of Water Wetter that needed to go in was 3 oz. This amount was figured by dividing 64 oz. by 5%.
Adding the two quantities of additive together gave the total amount of water that needed to be drained out of the system. Next, the system was refilled with the appropriate amounts of alcohol and Water Wetter. The pump was then turned on. Over the course of the next couple of days, the pump mixed everything evenly.
In summary, the basic ideas behind this cooling system was to get the heat produced by the computer out of the room it was in by moving all the basic active coolers that were in the PC outside, thus getting rid of the noise, and finally reducing the concern of overheating to nothing.
The main cooling action takes place outside on a wooden plate suspended from an overhang. In additionm, the plate was placed behind a pair of columns to protect it from the elements. The coolant temperature is regulated by a thermostatic mixing valve which then passes the coolant on to cool the CPU, chipset and GPU.
The now warm coolant then moves into the reservoir and then finally back to the pump plate to be cooled. The electrics of the pump & fans are plugged into the same power strip that the PC and monitor are hooked into, making it harder to do something stupid like turning the PC on without cooling it. Finally, the pump plate has the pump, radiator, drain, mixing valve and two fans on it.
This system has worked very effectively so far in keeping the PC cool, even when playing a game. The next test for the cooling system will be during the fall when the temperatures here start to drop. As far as changes to be made: The reservoir seals leak every time it is moved – it will eventually have to be taken out and resealed using silicon.
Also, quieter fans outside would be good since they are slightly annoying when they can be heard from 50 feet away. The only real complaint about the system is that it has enough extra capacity in it to heat up the air inside the case on summer days when the temperature gets into the 90’s.
The next thing is to get rid of the Intel board, buy one of Asus’s 875P motherboards and do some overclocking!
If you have any questions or comments, please e-mail me.