Petar writes a 7 page primer on building a watercooling system from scratch.–Petar Lazarevski.
Have you ever tried to hold the tip of your finger pressed to the core surface of the last generation CPUs, such as the AMD Athlon and Duron or Intel Celeron while switching on the computer? Those who did (I’m one and not too proud of it) know that no one could withstand the heat for more than a few seconds.
The CPU heats up so quickly that if there is no heatsink attached to it, the damage could occur quickly – even after a few seconds – especially in case of AMD products.
Although the core voltage is as low as 1.5 V, the processor generates as much heat as a medium sized light bulb. That heat has to be moved away in a most efficient way if you want your computer to work properly. High working temperatures are one of the greatest problems for the people who wish to push their machines to the highest limit – those people are often called overclockers.
If you are interested in overclocking (and I suppose that you are since you have chosen this site to visit) there is a good chance that you are going to find something of interest in the text that follows.
The typical CPU cooler consists of a metal heatsink with a fan on top of it. The efficiency of air-cooling depends on the heatsink material, mass, surface and contact with the processor core, as well as the surrounding air temperature. Let’s see how these five elements affect the cooling process:
The heat could be transferred by means of conduction, convection and radiation. In a case of PC cooling, conduction and convection are much more important than radiation and we’ll focus on them.
Conduction is a way heat is transferred through solid bodies as well as gases and fluids which are not in motion. Convection is a transfer through gases and fluids in motion. Having this in mind, one could easily conclude that in the case of air-cooling, there is a thermal conduction between the CPU core and heatsink and through the heatsink to the surface, and thermal convection between the heatsink surface and the surrounding air.
First point of interest in the heat transfer is the CPU-heatsink contact surface.
There are three factors that influence the conduction there: CPU core material, heatsink material and evenness of both surfaces. Since we cannot do anything about the material the processor core cover is made of (and very little or nothing about the evenness of it’s surface unless you are ready to do some CPU lapping), we’ll focus on the heatsink material and evenness of its surface.
The ability of the material to transfer heat is shown in the coefficient of thermal conductivity [W/mK]; the higher the better. Silver is a winner with 417 W/mK and copper is very good second with 395 W/mK. The material that is commonly used for heatsinks, aluminum, is much worse with 229 W/mK. However, it has lower density and is easier to machine, which, considering lower cost, makes it an overall winner for cheap coolers.
Picture 1 represents a drawing of highly magnified, uneven surfaces of a CPU and heatsink in contact. The gaps between the two (painted white) are filled with air. The thermal conductivity of air is only 0.6 W/mK, almost 400 times less than aluminum (!) and it is obvious that the heat transfer from CPU to heatsink is greatly reduced due to presence of that air.
There are two solutions to this problem and the combination of both gives the best results:
- Lapping the heatsink surface, and
- Filling the gaps with thermal paste.
A fine-grain silver powder would be perfect for this, but it has to be mixed with a paste.
Now let’s see what happens with the heat after it has passed from the CPU core to the heatsink:
It is well known that the heat always “travels” from warmer to cooler areas of material. Since the surface of the sink is cooler than the core, the heat spreads towards the surface, which is in contact with surrounding air. However, it is not that simple – heatsink material and mass are of great importance. We have to know something about a second important property of the material – thermal capacity.
In brief, it is the ability of a material to accumulate heat, shown in numbers by a heat quantity, that it can absorb per unit of mass if it’s temperature has risen by 1°K (or 1°C). The material with best thermal capacity is magnesium with 1.017 kJ/kgK, followed by aluminum with 0.896 kJ/kgK. Copper is much worse with 0.383 kJ/kgK.
However, since it is obvious that the total capacity differs with mass (higher mass – higher total capacity), we have to take into account that copper has over three times higher density (mass per volume) than aluminum.
The conclusion is simple: If we compare two heatsinks identical in shape and dimensions made of copper and aluminum, the one made of copper is going to perform better since it will have both total thermal conductivity and total capacity higher than the other one.
Why is the thermal capacity so important? In a few words, if the heatsink base and fins are too thin and light, they will not be able to absorb enough heat without a substantial rise in temperature, making the CPU run too hot. It is very likely that, in case of heatsink comparisons, the heaviest one with the largest total surface is going to be an overall winner if they are all made of the same material.
The last stage in this brief analysis is “cooling of the cooler” – heat transfer from the heatsink to the surrounding air. There are three factors that influence this transfer (convection):
- Temperature difference between the heatsink and surrounding air;
- Heatsink surface in contact with air;
- Coefficient of Convection (heat transfer from solids to surrounding gases or fluids)
The presence of fan is almost unavoidable since the warm air is not able to move away from the hot heatsink surface by itself fast enough.
We have finally come to the breaking point.
Let’s imagine for a moment that we could make two experiments: one is cooling the heatsink with air and the other is cooling the same heatsink with water. The air and water temperatures are identical. The question is is “How much more heat could water remove compared to air?”
The answer is simple:
About one hundred times more!
This is because the coefficient of heat transfer (convection) from metal to water is about 100 times higher than for air. Of course, this is just theoretically, since we must have a heat source that is able to produce 100 times more heat and the CPU is not.
Since the CPU produces nearly constant amounts of heat under maximum stress, we will not be able to remove 100 times more heat with water, but we’ll get something else: the CPU temperature is going to drop. Isn’t that what we are really looking for?
There is one more property that makes water superior:
One liter (quart) of water is able to absorb the same amount of heat as 4000 (yes, four thousand) liters of air with the same increase in temperature!
So, how do we make the water circulate around the CPU heatsink? It’s not that simple, so we have to make a special heat exchanger usually called a waterblock. The water pump and the water cooler are necessary parts, too.
There is a way to cool the CPU without the heat exchanger (waterblock), with the water circulating directly over the CPU core, but it is harder to achieve since there are problems with sealing. We will comment on this later in this article.
This diagram shows a simplified drawing of a watercooling system. This kind of setup is in my computer for more than a year now and I can say that it works great. It consists of the reservoir R, submersible water pump P, water cooler C with fan F, two waterblocks (1 and 2) and hoses (blue lines for the cold and red lines for the warm water).
The pump forces the warm water from the reservoir to the cooler. The airflow cools down the water, which goes into waterblocks for CPU and GPU cooling. After that, warm water flows back to the reservoir. Please note that the temperature difference between the “warm” and “cold” water is not so great; I have never measured it but I am almost sure that it is under 10°C.
The system could be even simpler in case one wants to cool the CPU only – there would be a need for one waterblock then, without the necessity for a “T” or, even better, “Y” joint water flow separation unit.
Another simplification is leaving out a water reservoir. This would mean that one has to use a non-submersible pump and only a small amount of water in a system, which is then close to the radiator’s capacity. A large diameter fill in/expansion tube would be necessary in this case.
However, I think that this is not the best solution since I consider a large capacity of water an advantage. A few liters of water have larger thermal capacity than, let’s say, a half liter. This means that a large amount of water is capable of absorbing larger amounts of heat for the same increase in temperature or, even more likely, the same amount of heat but with smaller temperature gain, which could be important.
Of course, the system could always be more complicated. One of the upgrades that I intend to make is adding a third waterblock for chipset cooling. I have noticed that the temperature of the northbidge on my Abit KT7A motherboard rises up to 50° C under load and I consider this high – definitely high enough for watercooling to be implemented.
The “heart” of the watercooling system and, as far as I see it, the most important part of it, is the waterblock. Basically, it is a heat exchanger, a unit where the heat is transferred from the CPU to the water and taken away. We could call it the “cooler” but it would not be an absolutely correct term. In fact, there is no “cold” and “warm” in physics (thermodynamics), but only more or less “warm”.
I’ll try to put this in plain words:
The waterblock (or the heatsink in case of plain cooling system) does not cool the CPU but just takes away the heat it produces, thus lowering the working temperature.
There are a few ways of making a waterblock; the cheapest one is using a small plain heatsink contacting the CPU and enabling the water to flow around its fins. Putting the heatsink in a plastic box with two nipples screwed in for hose mounting could do this. Of course, everything has to be watertight and this could be achieved with epoxy.
This kind of unit is simply drawn and shown above. The heatsink is painted black and the plastic box blue. The hose nipples painted red are shown on the top, while better positioning would be sideways. The advantage of this kind of waterblock is low price. It is also relatively simple to make at home without special and expensive tools and machines.
However, there is one big disadvantage: Low Thermal Capacity.
If the water pump breaks down, the water in the plastic block is going to heat up very fast, leading to CPU overheating and possible damage. If the temperature rises high enough, there is also a possibility of water leaks due to softening of the plastic parts. On the other hand, the people who use this kind of waterblock report good performance, very close to those made of copper. If you wish to use two in-line or parallel mounted pumps for safety reasons, this could be a good safety measure.
The second and, in my opinion, the best way of making a waterblock is by machining a solid piece of metal.
There are a few metals to chose from: silver, aluminum, copper and brass. The most important properties of the metals that we should take into consideration are, just like in case of heatsink material selection, thermal conductivity and thermal capacity. Of course, the ease of machining is something to think about, especially if you are not too handy with machines and tools.
Price could also be an important issue, making silver, the most expensive by far although the best thermal conductor, worse choice of all. Aluminum may seem the best choice, but there may be a problem with electrolytic corrosion. If we want a long time, trouble-free waterblock operation, I think that we should focus on copper as the best overall choice. Brass is an excellent material for hose nipples, but not for the waterblock due to its low thermal conductivity and capacity.
The first waterblock I made was of a cross-drilled type, with two main channels which could be called “collectors” and four channels of a smaller diameter connecting the two main ones at the angle of 90°. Two brass nipples for hose attachment were screwed in the collector channels on one side of the waterblock.
You can see the way it looks with the hoses attached. The dimensions are 18x50x60 mm. This kind of waterblock is very useful for Slot-type CPUs and some Socket types (it fits perfectly on my ABIT KT7A motherboard). However, since the hoses have to be connected sideways, there may be a mounting problem for some motherboards.
On this photo it is clearly visible that hose nipples protrude from the side of the waterblock and that the four connecting channels are plugged on one side with short Allen head through bolts (shown partially unscrewed below, Teflon tape removed for clarity:
You can see the plugs as they are in operation in this shot:
Small pieces of Teflon tape are also visible. This kind of waterblock is very easy to make. Let’s see what exactly we have to do to become an owner of a homemade cross-drilled waterblock.
First of all, one has to find a copper plate of suitable dimensions. You can choose the dimensions according to the type of CPU or MB owned. Generally, Slot type CPUs can handle larger waterblocks and you do not have to take too much care with the nipples length and placement, since the waterblock is perpendicular to the motherboard surface and there is a plenty of room for the nipples and hoses.
On the other hand, if you have a Socket type motherboard, or you want to have your chipset watercooled, there is another type of waterblock to be taken into consideration, as shown above. In my opinion, it is easier to attach it to Socket370 or SocketA CPUs. The main feature is, as obvious, the nipples being mounted on top rather than on the side of the block.
The first rule while choosing the copper plate should be: the heavier, the better – meaning more thermal capacity. Of course, care must be taken about the static load of the retension units and slot of the motherboard. Since your waterblock width and/or length could be restricted, especially in the case of Socket motherboard, by the surrounding elements, you could always choose a thick plate and contribute to the total waterblock mass that way.
However, there are cases where the mass of the waterblock does not contribute to the overall “quality” of the system. The main advantage of the large (meaning heavier) waterblock becomes obvious in cases of pump failure (or any coolant flow restriction). I have tried to shut off the pump a number of times for experimental purposes. The CPU core temperature has risen gradually and slowly, giving me more than enough time to shut down the computer or turn the pump back on. If one uses multiple-pump setup for the purpose of maximal safety, a large mass waterblock becomes obsolete.
If you plan to cool a Coppermine or Athlon/Duron Socket type CPUs, attention is needed while choosing copper plate dimensions, because the CPU core of these processors is small and thin, meaning that the core surface area is under the level of the Socket locking mechanism, its highest point.
There are two solutions: Either choose a plate that is small enough not to touch the locking mechanism (like the one above) or prepare to mill a part of the plate in a way it would not touch the Socket after the attachment (below).
All the copper plate choosing tips mentioned above are related to the waterblock for CPU cooling only. On the other hand, if you intend to cool the GPU on your video card, a high mass plate is not necessary since the dissipation of the video processor is not nearly as high as the one for the CPU.
All you have to care about are its length and width – choose them in a way so the waterblock cannot touch any of the GPU surrounding components protruding from the PCB. The thickness of 10 mm is, as far as I see it, quite sufficient.
On the opposite side of the high-mass waterblock, there is a choice of a low-mass waterblock or even an absence of a waterblock. A light waterblock heats up quickly but cools down quickly too, which could be an advantageous feature in case of a setup with two (or even more) pumps. A watercooling system without a heat exchanger is possible too, but with a few drawbacks that make this solution problematical.
First of all, there is a sealing problem.
If one wants to make water flow over the CPU core surface, he/she has to make an element that enables sufficient water inflow while securing the water from leaving the area of the CPU core. Of course, this could be done with an adequate seal (silicone O-ring or similar) placed between the CPU base and the water capsule.
However, this solution relies on a correct pressure of the elements in contact, which is critical. On the other hand, there is a watertight adhesive bonding solution, but then there is a disassembling problem that goes with it.
The second problem relates to water flow stoppage.
The cooling efficiency of this setup relies solely on a sufficient water flow. If this flow is stopped, a rapid rise in CPU core temperature is very likely, to be followed by an inevitable destruction of the processor in a matter of seconds, especially in a case of AMD products.
When the CPU core temperature rises up to around 100°C, the water at the core surface starts to boil. This makes the heat removal much worse and the destruction of the CPU is very likely to happen.
Although this way of CPU cooling should be more efficient than a conventional watercooling with a waterblock, the complexity of the design and mounting make it an unlikely and perhaps risky choice.
So, we have chosen the material and dimensions for our future waterblock and purchased the copper plate. Let’s see what we have to do to turn this piece of metal into the hardworking cooler.
The waterblock is a heat exchanger and to fulfill it’s main purpose, water passages have to be made inside of it. The easiest way of doing this is to drill a few channels and attach two hoses for input and output of the cooling fluid. A good, stable drilling machine and appropriate drilling bits are all we need for now.
There are special bits for copper drilling, but the plain ones are going to work well enough if you apply lubrication and/or cooling. Before you start drilling, you have to choose the number, diameter and position of channels to be made.
The cross-section drawing of a simple waterblock is shown above. Generally, it is a good idea to make two main channels (collectors) with hose nipples attached to them and a few connecting channels which allow water flow from one collector to another – the more the better. What you should do is drill two holes of a larger diameter (9 mm or so, depending on a nipple diameter you wish to attach), shown as two vertical white areas.
The copper plate has to be properly fixed before drilling in order for the channels to be even, straight and parallel to the sides. Use some oil for lubrication during the drilling process and do not apply excessive force – copper does not respond the same way as steel and the bit could easily get stuck and broken.
Never let the drill bit go all the way through the material – stop it a few millimeters before it penetrates the surface opposite to the one you started drilling in or you’ll have to put plugs in places where they could have been avoided.
The collector holes should look similar like those shown in the diagram. Please note that they are not positioned symmetrically – the right one is further away from the side than the left one. This is because the connecting channels have to be plugged with small bolts which should not protrude into the collector channel disrupting the flow of the coolant. Your copper plate should now look like this:
After the second main channel is finished, the copper plate has to be turned 90° (counter clockwise looking at the above) and secured in that position. Now you are ready for drilling the secondary (connecting) channels. The diameter and number of these holes could be calculated in the way that the sum of cross-section areas of the channels equals the cross-section area of the collector channel times two, thus ensuring the uniform flow of the coolant.
However, I think that this is really not necessary and that any water pump will have enough power to ensure sufficient flow through all the connecting channels, disregarding their number and diameter. My suggestion is: Make as many holes as possible, taking into consideration the given dimensions of the plate you have purchased.
Of course, take care not to make the holes too close together because you could end up with problems while making plug threads. A good idea is to keep the axes of the holes apart for the distance calculated by multiplying the connecting channel diameter by a factor of 1.5. Shown above are only three secondary channels for clarity, so please do not use this drawing as a standard.
The secondary channels should be drilled in a way to connect two main ones. This means that you should stop drilling the moment you feel the bit entering the second collector – the left one on the drawing.
After this operation, your future waterblock should look like this:
The diagrams below are there just to show the look of the main and connecting channels, their position inside the waterblock and the way they intersect:
As mentioned above, the larger the diameter of the collector channels, the better. Also, the waterblock should enclose as many connecting channels as possible. However, quantity is not the only thing that counts, maybe even not the most important. Let’s think this way: Copper (or some other chosen metal) is a good heat conductor, but it still has some thermal resistance.
This simply means that increases in the distance between the CPU core surface and coolant (water), i.e. increase in thickness of the copper between the surface of the waterblock and the inner surface of the connecting channels, worsens the heat removal and, therefore results in higher CPU temperatures.
If we want to “bring the flowing water closer to the CPU” we have to either drill channels of a larger diameter or drill a number of smaller diameter channels closer to the side of a waterblock that is going to be in contact with the CPU core. The idea is shown below:
I think that this drawing speaks for itself. However, if you decide to apply this solution, expect to have trouble while drilling since the bit has to make a hole in the curved inner surface of the main channel that is not perpendicular to the longitudinal axis of the bid, i.e. axes of the bid and main channel do not intersect.
The only solution of this problem that I could think of is to combine drilling and milling, but this complicates the process a lot. The diagram below gives a better view on the channel intersection and probably, clarifies the drilling problem a little.
If the copper surface that is supposed to be in contact with the CPU core is irregular (and this is nothing to be wondering about), you should try to make it as even and smooth as possible. You could start with a file and end with a fine sanding paper. The remaining irregularities in a waterblock surface after lapping are going to be filled with a thermal paste before attachment.
After you have finished drilling, you have to make threads for hose nipples and plugs. You have probably noticed that I have made M10 and M6 (standard metric) threads on my waterblock, but you should choose the type that suits you best. Of course, the channel diameters have to match the chosen threads, meaning that, for instance, you have to drill 8.3 mm and 4.9 mm holes for the M10 and M6 thread respectively.
After drilling and making threads in the copper plate, one thing that you have to do is clean the waterblock so there are no small copper particles left inside before you make your waterblock watertight. For plugging the holes, you could use short Allen head through bolts wrapped with Teflon tape or sealed with some silicone glue.
You can see the type and size of the bolt I have used to plug the secondary channels on one of my waterblocks (above).
All that is left to do so the waterblock is ready for action is choosing or making hose nipples. If possible, I suggest you purchase brass nipples of a right size for your copper plate and drill the collector channels and make suitable threads for the nipples afterwards.
However, if you can’t find a store close to you that sells these things or enjoy making them yourself, you should start from hexagonal brass bar and use a lathe to machine a perfect part that suits your needs.
These are hose nipples as they look prior to assembly, while this picture
is an example of the mounted nipple sealed with Teflon tape. Please bear in mind that the inside diameter of the nipple should be as large as possible, meaning as close to the diameter of the collector channels as possible.
It is a good idea to test the waterblock for leaks before making it a part of a system. Put it in the position with nipples facing upwards, attach two hoses (which should be in a vertical position) and fill it with water almost to the point when the water spills out of the hose ends. Leave it that way for a few hours and check all the plugs and nipples for leaks. If everything is dry, you can expect an almost eternal leak-free operation of your waterblock.
All right, you have finished your waterblock and are eager to find out how it performs. The first problem that you will have to cope with is how to attach the waterblock properly to the CPU, ensuring good contact and heat transfer. I have tried two different approaches and they both worked well.
If you have a Slot type motherboard, the easiest way of waterblock attachment is the one shown here:
This is a plain drawing, intended just to show the idea. The waterblock (painted orange) is pressed against the CPU (painted blue) by means of two metal bars held together by two long bolts. If you chose this method, please take care not to over tighten the bolts and to tight them evenly, so the two bars would stay parallel ensuring the best waterblock – CPU contact possible.
Maybe it is not a bad idea to put a piece of soft rubber on the side of the bars in contact with the waterblock and CPU (above) making the pressure distribution more uniform.
For Socket type motherboards (or Slot 1 – Socket 370 converters) I suggest you to use a simple mounting clip, as shown here:
This unit is made of solid brass bar drilled on both sides for bolts holding steel clips that should be hooked to the socket. For this purpose I used a straight part of the hose clamp
drilled on one end (so it could be fixed to the brass bar with a bolt) and shaped on the other in a way to attach itself easily to the socket. A simplified drawing is shown here:
On the right place (NOT in the center) of the bar, you should drill a hole and make a thread for the tightening bolt. Referring to this diagram,
notice that the tightening bolt (painted blue) is placed right above the center of the CPU core (painted black). This is very important since the good waterblock – CPU contact is of the essential importance for the heat transfer.
If the waterblock and CPU core surfaces are not absolutely parallel in contact, no thermal paste could compensate for this irregularity. I am sure that good contact, as mentioned above, is more important than the waterblock design itself.
I have owned three video adapters since I purchased my first PC and only the last one – Diamond Viper 770 (nVidia Riva TNT2 GPU) – had a built-in heatsink. It is obvious that, along with the development of modern GPUs, the need for a good cooling increases. I have noticed that, although I have attached a fan to the GPU heatsink, the backside of the video card PCB is very hot around the place where the processor is placed.
That fact has forced me to make the second waterblock – a smaller one this time for cooling the GPU. I have chosen a copper plate 50 mm wide and 10 mm thick – just about right to be attached to the GPU without interfering with the components of the video card.
It is a cross-drilled waterblock, made pretty much the same as the larger one, only with smaller nipples and three connecting channels instead of four.
This photo clearly shows the waterblock attached to the video card GPU, with three connecting channels plugs in place. Parts of the Teflon tape used for sealing are also visible.
If you decide to improve the cooling of your video card this way (which would, most likely, result in excellent overclocking abilities), be ready to face two possible problems: difficulties during detachment of the original heatsink and uneven GPU surface. I have had to cope with both of them.
It took a lot of courage to start detaching the heatsink that originally came glued on the processor of my Viper 770 video card. A wrong move could cost me around $100, something I could not afford and did my best to avoid. The only tool used for this job was a scalpel that I tried to insert between the heatsink and the GPU.
I used the minimum force possible and it looked like it was going to take for ever. After fifteen minutes or so, the heatsink came off suddenly, resting a few inches aside the video card. Since everything was followed by a loud pop, I thought that something was broken, but felt a great relief when the PC booted with no problem at all. It could be a good idea to put your video card in the freezer for a few hours. This should make the adhesive brittle and, therefore, could lead to easier heatsink detachment.
It looked like that all that was left to do is to attach the waterblock and let the water flow. Unfortunately, when I put the waterblock on top of the GPU to check the contact of the surfaces, it came out as a big disappointment – the GPU top was so concave that I could stick a needle between the two. The situation is shown, with exaggeration, here:
No thermal paste could cure this. I had to flatten the plastic surface of the GPU, but since the large amount of material had to be removed, a flat file was the primary tool instead of sandpaper. After a few minutes of filing and blowing off the plastic residue, everything was ready for a final check and waterblock attachment.
The contact seemed perfect, so I applied a thin layer of thermal paste on the GPU core (black plastic part) and silicone glue on the green PCB part of the processor (above). A ten ponds weight was used to press the waterblock firmly against the GPU for some 24 hours.
If you decide to try this or a similar procedure, be careful with two things: the amount of material to be removed from the GPU surface should be as little as possible (I can’t tell how deep you can go until you damage the core) and the distribution of weight on the video card PCB during bonding should be as uniform as possible.
The latter could be achieved by removing the slot-mounting metal plate and putting the card on a soft surface (like towel or sponge). The position where the waterblock should be attached and a part to be removed before pressing the waterblock to the video adapter are clearly shown below:
The way the video adapter should look after mounting of the waterblock is shown here:
Please note that the nipples have to be sized in a way as not to interfere with video RAM or other elements on the PCB. If the hose inner diameter is small enough compared to the outside diameter of the nipples, you will not have to put clamps at all. Of course, this stands for the CPU waterblock, too.
There are a few more elements in a PC that need heat removal (chipset, hard disk, PSU, sound card), but without doubt the most critical are the CPU and GPU. If you decide to cool them both with water, you have to choose a way to connect them with the water pump.
In-line mounting, meaning water going from one waterblock to another, is much simpler, but, in my opinion, a solution with a major disadvantage: the water already warmed up in one waterblock goes through the other. This definitely reduces the efficiency of the second waterblock in the line, making either the CPU or the GPU run warmer than it should.
By mounting them in parallel, the water flow is separated in a “T” or, more preferably, “Y” joint and distributed to both waterblocks. It is a good idea to choose the input hose of a larger inner diameter than the two output ones because that ensures the better flow separation. However, if the pump were strong enough, this would not be too important. Above you can see the “T” water flow separator and both waterblocks. The supply hose is connected to the radiator outlet.
Let’s spend some time commenting on a water flow.
The water pressure drop due to the flow resistance is directly proportional to the fluid density, square of the flow speed and a resistance coefficient, which itself depends on a length and diameter of the tubing, as well as on a friction coefficient.
If there is nothing we can do about the resistance coefficient, since it is a fixed value for the PVC hoses we usually use in PC watercooling systems, we can surely use hoses as short as possible and with the as large a diameter as possible. This is going to reduce the flow resistance to a minimum, improving the water flow and, probably, cooling efficiency to some extent.
Another thing that we should take into consideration is the choice of a flow separator used in multiple waterblock setups. I have used a “T” separator type since it was the only one I could have at the moment, but simple analysis shows that we should use “Y” separators whenever possible. The flow resistance coefficient for the “T” unit is 1.00, while only 0.55 for the “Y” unit, meaning that the water pressure drop in a system with the “Y” unit is almost half of that in a system which incorporates the “T” water flow separator.
We have already mentioned the high thermal capacity of water as a great advantage. However, although water is really a good heat reservoir, it is getting warmer while the PC works, thus reducing the cooling efficiency of the system and making the CPU and GPU run hot. Of course, this implies a closed system with circulating water.
The first test-run of my waterblock was made with tap water coming from the bathroom and going back to the sink. This worked great, since the temperature of the tap water is rather low. However, I had always thought of the cooling system as a closed-loop one, so I had to incorporate a radiator and pump in a system.
I had trouble finding a good radiator dedicated to PC watercooling, since I could not buy one in the town I live in. I read that people used old car heaters for this purpose and found that more than acceptable. There is a large swap meet in my neighborhood, so I managed to buy two heaters at a very affordable price. The one that I use right now is large enough not to fit in the PC casing, but the other one (below)
that I purchased later is just about the right size to be mounted inside the casing.
In this photo one can (unclearly, unfortunately) see the old, larger radiator currently in use with the fan attached. I had an idea to make the air flow around the radiator fins even, so I used the bottom of a plastic food container to make a fan shroud:
I think that this helps the air to flow across the whole area of the radiator. Since this is supposed to be a temporary solution, I did not pay much attention to a way the fan shroud is mounted on the radiator, so I used some wide adhesive tape.
I expect to put my new radiator inside the casing. The cooling efficiency is probably going do decrease a little, partially because the new radiator is smaller and partially because the air in the casing is a little warmer than the outside air. However, the advantages (compact system, quieter operation) are big enough for this solution to be chosen. Of course, the water flow rate is going to increase a little due to much shorter hoses, so the result could be surprisingly good.
Water pump selection is an important task, but not as nearly as waterblock making. Basically, you can chose between two types (submersible and in-line) and, maybe, between power supply voltages. I did not have much choice, so I have purchased a submersible pump for aquariums that works on 220 VAC.
The main advantage of this solution over the in-line type is quieter operation. However, if something goes wrong with the power cord (which should be absolutely water-resistant), lots of damage could occur, maybe even total destruction of the PC.
I have tried to measure the electrical resistance between the waterblocks and PC components. The lowest value I have seen was 150W, which means that there is a possibility for the damage to happen if the wires in the pump coil or power cord get in contact with water. Of course, usage of distilled water reduces the risk, but if you really want to play safe, use the in-line pump.
The pump that works in my system for over a year now is shown here:
On this drawing, you can see the way three main parts of the pump look like when it is disassembled. It is a good idea to have a hose with the cross-section area matching the one of the water pump outlet.
You can connect the pump to the computer’s PSU so it would start the moment you turn your PC on, or to a separate plug with a switch. The second option (the one that I use, mainly for testing purposes) is not recommended, in case you do not want to risk your system running without water flow if you forget to turn the pump on.
This shows the way a complete system looks after assembly. The only part we have not discussed in detail yet is a reservoir. As you can see, I have used an old plastic food container. It is large enough for the pump to fit in and has a cap robust enough to withstand drilling and mounting of four nipples. Of course, everything has to be watertight, so I have used a sealant for the cap and nipples.
It is not clearly visible that there are four nipples attached to the reservoir cap: one outlet, two inlets and one (much larger) for the fill-in/expansion tube. It is strongly advisable to keep the reservoir completely filled with water, since the submersible pump should be (for cooling reasons) covered over the top with water.
If there is some water in the expansion tube, you can be sure that the reservoir is completely full. I use the expansion tube also as an outlet for the pump power cord.
You have probably noticed that there are two hoses going out of the casing. Since the radiator (in this stage of system development), due to its large size is outside, there are two nipples mounted on the casing for attachment of the hoses that are provided for the water flow to and from the radiator. Here
one can see the look of the threaded nipples (nuts and washers not shown) that could easily be mounted on a casing wall. Since there is a lot of room here, I put hose clamps for security reasons. The same type of nipple is mounted on the reservoir cap as a water outlet from the pump.
Since the threaded part of it is submerged, I suggest using high quality, stainless nuts and washers. Brass, the material used for the nipples, does not corrode.
The first upgrade for this kind of watercooling system that I could think of is adding Peltiers. There are basically two points for Peltier application: direct CPU cooling (meaning Peltier mounted between the waterblock and the CPU, with a provision for a cold plate on top of the CPU) and cooling of the water that goes into the waterblocks.
I like the second idea better, so I am already thinking about purchasing a powerful (150W) Peltier.
The second upgrade that I have already done is making a system (almost) fail-safe, i.e. an add-on that provides for secondary cooling to prevent CPU damage in case of water flow interruption. If you eliminate the possibility of leaks (only major potential problem besides the mentioned one) by using high quality materials and first-class workmanship, the only thing that you have to worry about is water pump breakdown or clogging of the water lines.
Application of the right materials (copper instead of aluminum) can minimize the possibility of corrosion, which is the main clogging reason. However, the water pump could stop working at any time and there is no way to predict that. Of course, it is possible to have a device capable to detect that the pump has stopped by means of water flow measurement (which, of course, detects the clogging of the lines, too).
Another solution is to use two pumps in parallel work; if you choose this one, you will surely not make a mistake. I have chosen a simpler, cheaper and probably not the best solution – a piggyback cooler,
as shown above. The idea is to mount a plain heatsink with fan on top of the waterblock with the mounting clip. Of course, thermal paste is applied between the two. This solution works surprisingly well, as the tests, which will be presented later have shown.
If you choose to attach a piggyback cooler to your waterblock, I suggest you to purchase a better one than I had (like an FOP or Alpha). Since you’ll have at least the same cooling capacity as you would have with the same cooler mounted directly on the CPU, this should work exceptionally well.
When I think about a year’s experience with the watercooling system, I can say that I have noticed only two minor disadvantages. The first one, as mentioned previously, is the radiator placed outside of the casing. Since this is going to be corrected soon with the smaller radiator, there is only one thing left that I have noticed – dust accumulation on the radiator fins. This could maybe be avoided by placing an air filter in front of the fan.
Finally, if you build a system like this for yourself, you will be interested in its performance. CPU and GPU temperature affects the overclockability of your PC as well as reliability, and good watercooling improves them both. That’s why I have made two sets of tests: one with the CPU working at the default setting (700 MHz) and the other with overclocking (up to 1003 MHz). Since I do not have the option for GPU temperature measurement, I have only tested the maximum overclockability of the video adapter.
The CPU tests were made with the cooling system in full operation and with a water pump breakdown simulation (obtained by turning the pump off). The minimum and maximum CPU temperatures were recorded. Please take these results as approximate, since they were acquired only by temperature readings from the ABIT KT7A’s in-socket thermistor.
I have tried to make the reading more realistic by bending the sensor upwards a little (so it would have better contact with the CPU) and putting a piece of rubber sponge under it (which should isolate it thermally from the surrounding air). I have noticed that the temperature readings were a few degrees higher after this small modification.
The programs used to put the CPU to the maximum stress were CPU Stability Test 6.0, CPU Stability Test 1.3 and Prime95. For GPU testing, I used 3DMark2000, MS Flight Simulator 2000 and Quake3. For monitoring the CPU status, I used WCPUID 3.0 Beta 1.
The first test was made with the CPU set to its default speed of 700 MHz and core voltage 0.75V less than default (although the VIA Hardware Monitor V2.04 shows only 0.4V difference). I have chosen this voltage because the CPU works perfectly stable with it from 700 MHz to 820 MHz, so there is no need to go higher.
Turning off the pump simulated the water pump breakdown. The system worked perfectly stable in both cases, although the temperature was considerably higher when the pump was off. To make sure that the readings are as accurate as possible, I ran the tests for an hour after the temperature stabilized at its highest level. It is obvious that the piggyback heatsink is working well enough to ensure the normal CPU operation in case of water pump breakdown.
The test results are shown in the following table:
|Idle – Pump off|
|Max – Pump off|
The next test was made with the same voltage settings but the system clock was raised from 100 MHz to 117 MHz, making the CPU run at 820 MHz. The system was absolutely stable.
|Idle – Pump off|
|Max – Pump off|
It is obvious that the temperatures are very close to those from the previous 700 MHz test. If we take into account that the room temperature was 1°C higher, all the results, except for the one with the pump turned off during maximal stress, could be considered identical. The piggyback cooler proved to be more than useful during this test too.
The next test was made with the CPU overclocked to the limit. The maximum frequency at which my AMD Duron 700 works absolutely stable is 980 MHz (140MHz FSB). The maximum acceptable working frequency (PC still usable for all programs except extreme stability tests) is 1003 MHz (143 FSB). The next table presents the data for the CPU working at 1003 MHz @ 1.85V.
|Idle – Pump off|
|Max – Pump off|
The maximum registered temperature of 39°C was achieved with CPU Stability Test 6.0. However, all other programs did not make the CPU run higher than 37°C. I tried to max out the temperature while the pump was off, but stopped at 55°C since I was afraid of the CPU damage (although the PC was still working OK). It is interesting that the same piggyback cooler worked better with my old system (ASUS P2V-B motherboard with Intel Celeron 466 CPU), but this is not too strange since the Celeron 466 CPU dissipates much less heat than a Duron.
Let’s see the performance of the AMD Duron 700 @ 1003 MHz. The well-known SiSoft Sandra 2001 test shows the following results. The Norton Utilities 2001 System Benchmark shows a score of 310:
The default frequency setting for my Diamond Viper 770 video card is 125 MHz for the core and 143 MHz for memory. The highest setting that I have achieved with (not absolute) stability is 170 MHz for the core and 190 MHz for memory. I consider this a good overclock which could not nearly been achieved with a built-in heatsink, even with the add-on fan.
Quake Timedemo1 with 1280x1024x16 resolution, geometric and texture details at high, scores 31 FPS. 3DMark2000 at 1024x768x16 scores around 3700, but can’t complete the test every time with 170/190 setting for the frequencies.
My opinion is that this is an efficient and reliable cooling system, especially considering its low cost. If you decide to make one for yourself, you are surely not going to regret it. I doubt that any conventional cooler can match the performance of this kind of system.
Everything (except two details) mentioned in this article is something people are making and writing about for a long time. I started this project after reading a lot of articles, mainly on Overclockers.com. There are only two things that I am sure a lot of people were thinking about and some of them may have built, but I have never seen them in a review or article, and those are a fan shroud and a piggyback heatsink.