cpu computer water-cooling watercooling water cooling cool Always try to improve society rather than just take from it. Until then, lawyer stuff. Those using this information do so entirely at their own risk. Errors and omissions excepted. Contents subject to change without notice. All material herein is subject to copyright, patent and other intellectual property rights. All rights reserved. Copyright (C) J.Partridge. 2004.

Please note: In these days of litigation taking the place of natural justice, it is necessary that those wishing to read the monograph must understand that they must not, nor will ever hold the author responsible for any damage or injury. This is because the author does not want a parasitic lawyer coming on heavy because some damn fool thinks he or she is some brilliant engineer, or stubbed their toe and wants a fortune because they happen to be stupid.
I don't mind putting my knowledge on the web, after all, we should help make a better world, but as lawyers get richer, society gets poorer. So Please use your vote to keep lawyers out of politics. Have a nicer day :)

As just one of tens of thousands of science graduates and engineers being wasted in Britain. I write this website to help others and as a C.V. - hoping it may lead to a job. Gizzajob.

Always remember that real engineering is doing for a fiver, what it takes a big corporation to do for a thousand quid.

I first got into cooling computers when the large custom chip in my BBC 'B', used to overheat and cause the computer to crash. I soon started developing coolers and discovered thermal paste 25 years ago, along with alloy radial cooling fins and small fan systems. The biggest thing I learnt is just how easy it all is. Modifying computers is still fun.

Beyond the hype:

The other hype is that the more you spend, the better the system, which is total and utter rubbish. The best systems I know cost pennies.
A friend who builds high specification computers recently put a top commercial water-cooling system in his best computer and it did very little to improve the running temperatures. The water-cooling was in my opinion, failing in the simplest design needs and mostly a waste of money. I would happily replace most commercial systems components.

I have friends who still refuse to accept the underlying physics, but rather prefer to accept the hype and high price tags. This is not for them, but for those who prefer to try building something better, rather then to open their wallets.
I truly dispair of much of the fancy crap they often buy for massive costs and yet have no useful advantage. The poor level of schooling today has led most people into knowing very little and all too often taking hype and bullshit as the truth. The ability to think for oneself is lost in modern Britain. For most people, only the big shops and internet catalogues have the modern form of truth: the more you spend, the happier you will be.
If you want to buy a fancy, bullshit computer, then stop reading now, grab your credit card and go surfing.

If you have an open mind, then you may wish to read on.

This web page is large.
It starts with a lot of description of how present systems can be improved.
If you want to build your own, then the later sections will show how many different ways can be achieved for no money and with far greater style then the present commercial junk.
Building your own pumps, radiators and cooler headers is also mentioned later. including a few radical direct water cooling designs which are cheap to make, but hard to design correctly.

About watercooling.
Water-cooling is simply a means to move heat from one place to another. Luckily, water is a superb transport medium, ideal for regulating beautiful planets. Water it can absorb a great deal of heat energy without raising its own temperature by any vast amount. This can then be transported to a device which can remove the heat energy out of the system.

Water-cooling is in my opinion, open to far too many myths. Admittedly water-cooling of home computers is definitely in its infancy, as is much of the computing world, so many myths have naturally built up.

One of the biggest myths is the amount of coolant flow which is often greatly overestimated.
I started by calculating just how much water is needed to cool a typical 70 watt processor remaining around 45 degrees. This assessment of the basic arithmetic led to many insights and design possibilities.

Please note that there are thermodynamic booby traps in calculating such systems. The first is that the ambient air temperature in the room is the limiting factor and any system temperature must be above the room temperature. If you live in a centrally heated, double glazed house, then expect the overall temperature of the system to be higher than most others, as you cannot remove so much heat in a warm environment. If you live in a draughty cabin in the Arctic, then an identical system will run a lot cooler, as the heat difference between the room air across the radiator and the warm coolant will be able to transfer far more heat.
Just speeding up the radiator fan will not make things cooler, nor create a great difference in temperatures once the system has settled down.
The laws of thermodynamics state that you just cannot cool down below the temperature of the room. To learn more, check out the concept of entropy. Ask your science teacher, as it's basic science.

I originally thought the many experts out there knew their stuff when they said a pump needs hundreds of litres per minute, but gradually doubt crept in. As described later, I built a small pump for nothing, which pumped 1 litre per minute at 12 volts and half a litre at 5 volts. Surely this pitiful little pump was of no use in a computer.

As part of a four year engineering apprenticeship, I happen to have studied thermodynamics, so decided to do a little basic maths. In the second part of this monograph, I've kept the maths simple, so all can follow.
It turns out that the fancy pumps are not needed. Indeed, even my little test pump could be run at low speeds. If the experts got the coolant flow wrong, I wondered what else did they may have got wrong. The answer was of course is that they got it all wrong.

Looking closer, it turns out that the experts don't know their subject very well and prepared to sell selections of poorly matched, or inadequate components which will indeed water-cool a computer, but not very well. They are like most British experts, good at taking money, but not good at physics. They make a lot of money because no one seems to bother to understand what they buy. Even the hobbyist computer magazines with articles on water-cooling make some massive bloopers.
With poor education, we will continue to buy crap. British education flounders due to poor maths and science teachers. Half of British maths teachers are not trained as such. Sorry kids, the likes of me with B.Ed and B.Sc, are out there applying for jobs to make learning far more interesting, but being white, male and over forty, we are considered fit only for shelf stacking. Don't blame the kids, nor the thousands of science teachers looking for work.

Trained as an engineer, fitting superheated steam systems, a draughtsman and HNC refrigeration, steam generation and central heating, a science degree and such like, blah, blah blah, and even a couple of the totally crap GNVQ's for my sins, I've noticed many computers with water-cooling and all seem to have less than ideal set-ups. The many problems seem to occur with poor plumbing and its consequences, some no better than air cooling and a few that are actually worse. I could easily see a few simple ways to improve such systems. It really is not that difficult with a little background knowledge. (As I was highly commended for my curriculum and lesson plans during my B.Ed studies, I've written lesson plans for 14 yr olds as a set of six technology lessons on heat, with three different water cooled computer systems which can be made for pennies, to allow the pupils to make working measurements and learn about heat, basic thermodynamics and a modicum of engineering. The lessons plans are available by email.)

To prove the point that water-cooling can be done much better and a blatant rip-off, much of the following is done for no cost other than a little time and effort. Be it science or anything else, I don't accept the excuse that money is a barrier to making the best computer, nor a barrier to making the best pupils. (Making projects for pennies is part of a many Brit teachers jobs, as there is no working money in our education system, so the better teachers scavenge and teaches the pupils to scavenge if they want to learn more, than just passing exams.)

Having recently studied a GPU heatsink from a top computer cooling firm, well, not so much study, as to laugh at it, I was appalled by the poor design and atrocious build quality. The connecting pipes cause so much restriction to the flow because the cross sectional area was less than half of the pipe bore. The metal body was poorly designed and made. I just would not want this level of build in any of my systems for fear of it failing. Simply not acceptable.
If you are going to spend three figures on a CPU and another three figures on a graphics card, then at least these expensive components deserve something genuinely decent.

This webpage does NOT include is how to spend money.
What is does include is how to use your brain and hands.

As you read through, try to understand the underlying thinking behind water-cooling, so you can do it properly, or at least a lot better than done by many. My last friend who bought a water-cooling kit, soon removed it because it was not very good, yet it was from one of the better manufacturers.
Don't get stung like him, start to learn what is happening, and how to do the job sensibly and effectively. This and many other ideas are available as complete teaching projects with lesson plans and homework etc, for secondary schools, as part of my teaching studies, so if one day, you ever get me as a teacher, expect technology and science to be a lot more fun.

Let's start with a few simple considerations.
Decide if you need water-cooling; perhaps a hot running processor, or a small case, or overclocking. If not, then simply make your life easier and save the hassle and your money. Air cooling is not perfect for getting temperatures very low, but for most users and most computers, air-cooling is perfectly adequate.
A small case should not run hot, even with air-cooling, as you should be able to control the airflow easily. I build very compact cases and they still run no hotter than full towers. So check the internal aerodynamics of a small case before resorting to water-cooling.
From assholes driving 4x4's on the school run, to full tower computers, bigger is not always better, merely an excuse for incompetence or not understanding the point adequately.
Try some cardboard corn flakes packet and adhesive tape to improve the airflow first, it's a lot cheaper and far more reliable than adding more, expensive fans.

As to silicon, always remember that if you want to run faster, get a faster processor and graphics card, it really is that simple. As Harley owners say, 'there ain't no substitute for more cubes'.
If you can't afford more cubes, or you have all the cubes that are available, then it's time to tweak what you can afford.

Water-cooling is now considered a serious option as overclocking utilities are often supplied with the better graphics cards and motherboards.
Please remember that much overclocking is NOT due to a hotter chip, but because the silicon manufacturing process is giving high outputs of almost perfect silicon, and so some chips have to be down graded as slower, just to fill the market niches. - Sometimes the GPU is high spec, but rated much lower and will overclock without hassle, especially if it's an end of the line processor, where the fabrication is almost perfect.
Other systems are crippled with poor memory so that overclocking the GPU is a waste of time.
Likewise, a fast CPU with moderate graphics does not make a good games machine. By choosing the right GPU or CPU and memory, you can have a perfectly good, much faster item for standard prices and water-cooling is not always needed.

Always know the bottom line and starting points before venturing forward.

Assuming you have a real need for water-cooling, then here are some incredibly simple facts that some builders never seem to understand.

Water is a superb transport medium, but merely carries the heat from one place to another.
The amount of heat water can carry is phenomenal and far better than any normally occurring fluid on this planet. If using water temperature control with oceans is good enough to keep this gorgeous planet under control, then it's good enough for a computer. If there is a God, then he/she/it should be phenomenally proud of the power of water.
Water-cooling is superb and apart from refrigeration systems.

The amount of heat transferred depends upon the amount of water in contact and the difference in temperature. The greater the heat difference between hotspot and coolant, the greater the amount of heat that can be transferred per unit area in contact with the water passing over it. For a set difference in temperature, then the greater the coolant flow, the greater the amount of heat that can be transferred in a system. The cooler the coolant, the cooler the CPU.

But the little area between coolant and CPU or GPU is where the game lies. The rest of the system is to get this small hot spot as cool as possible.

You do not need to run a processor at sub zero as it is designed to run happily at up to 70 degrees, so this is all you need to play with. Overclocking yet running at 40 degrees is a happy piece of silicon. Running at a few degrees above room temperature is an almost perfect system.
Sub zero is theoretically better, but for playing games then this is overkill, as the chip will still generate heat, so push it by all means, but don't worship it. Sub zero is for posing in the pub with your friends, - if you have any friends after building a refrigerated computer. Therefore you must aim to present the hot CPU with the coolest water to be able to transport away the maximum amount of heat and do so sensibly.

Removing heat from the coolant is usually done by a radiator by transfer into an airflow.
Always remember a fundamental part of this game: The radiator cannot reduce the coolant temperature down to the ambient temperature. (of the room).
The better the radiator, the better the system. Poor radiator, poor system.
A poor heat removal system will always make the rest of the system equally poor. Unless all parts do a good job, then the whole system will be poor. - Your coolant chain is only as good as its weakest link. The radiator is the weakest link, - if it cannot remove the heat, then all the other parts are worthless. The vast range of radiators and how to make and modify them is discussed later.

High or low pressure systems.
Many commercial systems use a sealed system which has high pressure flow, more so when the enormous flow rate is advertised. Other systems run at atmospheric pressure, usually employing a vent system.

Water-cooling systems do not need high pressure to flow, but it is the amount of coolant passing which can improve a system, but that does not mean a system pumping twice as fast will remove twice the heat, - far from it. The system will settle down to its happy system, as long as there is some coolant flow, be it reasonable flow or much faster.
The heat transfer of the CPU header is by its very nature restricted to the heat flow through the metal, and as such, even flowing the contents of Niagara through it, will only allow a certain amount of heat flow to be transferred.
Any advert touting high coolant flows, is in my opinion, utter crap.
Getting the coolant matched to the heat flow is the way to make an effective and sensible system. It is for this reason that most CPU headers use copper or aluminium casings, as these have high heat transfer rates.

Cars have a pressurised system, but only to reduce steam build up from a system which must run close to boiling point. Computers should never get anywhere near such levels of heat, as the CPU should never get hotter than 75 degrees, as your silicon can fry at around 90 degrees. If you run your coolant hot, then it canot cool the CPU as there will also be a poor thermal increase in the area between the coolant and the silicone die.
The point of water cooling a computer is to keep the silicone die temperatures safe or lower; water-cooling a computer is not the removal of heat to simply prevent a car engine from melting, but cars can get away with running high pressure systems close to boiling, as the engine does not fry at 90 degrees. Indeed, a car engine uses a thermostat to keep the engine running about 60 degrees, as this is where the engine is optimised to run, with warm oil in winter and good combustion from a warm engine. Computers should run at much lower temperatures.

A coolant design with small bores will need higher pressure to get around the system and also need to pass the coolant faster than a slower, larger bore system. All other aspects being equal, large or small bore systems could both transport the same volume of water and transport the same amount of heat. A little increase in bore and fewer restrictions can make for a far more reliable system.
Lower pressure usually equates to fewer leaks and greater reliability on a less stressed pump. Water and electricity are not the happiest of bedfellows. Therefore no one wants a leak, or a cooling design which could cause damage to the computer if it should leak. Therefore a low pressure system would be preferred. All my water systems are vented.

Water is heavier than air, so any bubbles will want to flow to the highest point in the system unless blocked by an entrapment area. A well designed system will use this to advantage to get rid of any air. Air in the system means contamination and lower heat flow.

The last thing that many forget is that pipes should always be neat, but they never seem to be as good as one would like. Therefore the designer should always consider making a neat job.

Looks and style.
The water cooling systems seem to be just a black box pump, a black radiator and a couple of plastic pipes and metal lumps.
Few systems ever seem to fit well, nor do they take maximum advantage of the various spaces and airflow available in the standard desktop, mini or midi tower case.
It could always work better and look far, far nicer.

Cost.

Many people would assume water-cooling is for overclockers to take their present processors to new temperature levels, or at least to keep overclocked silicon at safe temperatures. But not for all. For me, water-cooling is just good old engineering fun, a chance to develop designs far better than the mundane rubbish available commercially.
As a biker, I see far too many 'customs' built from catalogue parts and too few genuine customs. Even the custom 'build offs' on the TV are a load of second raters, pretty and expensive, but otherwise just more of the same old stuff.
(I prefer to build hubcentre steering recumbent bikes, some are road legal and handle far better then my Ducati. As a mountaineer, I see atrocious camping equipment and poor, bedraggled, overloaded kiddies learning that hiking is all heavy boots and a heavy rucksack. It just ain't so. I walk all parts of Dartmoor, Snowdonia and the Lakes and enjoy it without heavy equipment.)
We can all do better and that includes computer builders too.

The options beyond watercooling.
For some, it would seem ideal to be able to run the processors at just above absolute zero (-273K) but this is unlikely, as strange things happen to sub atomic particles in this region and anyway, the pumps used for getting near this temperature are strange and esoteric. Likewise liquid nitrogen is also a bit of a hassle. Only astronomers need this kind of chip cooling, and then only to reduce sensor 'noise'.
A better solution would be freezing down to minus 20 degrees or lower, but this again is unlikely with present compact refrigeration systems. Keeping the processor at an ice-cold zero Centigrade (32 Fahrenheit) with a compact freezer is more likely in the real world. This is done by using solid state Pettier heat pumps or by using specialist refrigeration liquid such as the successors to 'freon' which can change easily between liquid and gas, so the thermal changes can be used to reduce temperatures well below the normal background (room) temperature. These are the less notorious CFC's the chloro-fluro-carbons. See later.

Tip: If playing with peltier, it is probably best to use the peltier to cool the water of a system, to run it close to zero using antifreeze, rather than use it directly on the processor. The feedback control of a peltier direct on the CPU would need a very fast control system, whereas a peltier working indirectly via water-cooling will give a far greater degree of flexibility and reliability to the system. More importantly, it would allow the processor to reach a stabilised temperature, then stay there with minimal changes in temperature to reduce stress on the silicon and of course, also have a far longer safety and more gradual temperature rise should the peltier or its control system fail.

For most people, aiming for room temperatures using water is easily possible.
The plot improves with AMD and ATI processors which do not get stupidly hot. For some intel processors and the latest batch of nvidia, (if rumours are to be believed), then cooling is needed just to stay in the game. Many overclockers prefer Far Cry and therefore choose to be in the ATI clan, rather than the Doom3 and therefore nvidia gang. Whichever path you prefer, water-cooling for overclocking can go a lot further with cooler running processors, even if only used to improve reliability of an overclocked device.

Whatever you use water-cooling for, it must be sensible and applicable.
Perhaps you indeed prefer hot running processors of up to 100 watts and graphics cards which need two molex connectors to generate frightening amounts of heat. In such cases, water-cooling may be simply necessary to protect your investments, rather than as an aid to pushing the limits.

Before choosing water-cooling, it's always best to check the standard computer running temperature. Pencils rool. Write the temp on the front of the casing for later reference and as a bottom line. It's all to easy to forget the original temperature when thinking of many other aspects of the design. I may find only a five degree drop in running temperature on some after market systems and if so, I would soon be modifying the water-cooling kit, and how it is used.

Considering the water-cooling kit.

There are two main types of water-cooling systems.
The most common is the small bore systems, which need higher pressures to overcome the resistance in the systems. These are more prone to leaks but are generally neater due to the small pipes.
Personally, I don't like small bore systems as their connections are often too constrictive, but they can work well if you get the system nicely balanced. Their advantage is that they can use low volume pumps used for small garden water features, or so the catalogues would have us believe. I believe that even smaller pumps are possible yet still run with reasonable pipe bores.
The other is the low pressure systems, which use larger bore pipes and as such do not suffer from resistance in the system. These are less prone to leakage as the pressures are reduced. The large bore system needs slightly larger radiator piping, but this gives the opportunity for greater cooling.
I also use large bore systems in a special, pump free design as mentioned later.

The high pressure systems can be more compact in its plumbing, but their radiators must still be able to transfer the heat, so radiator size will still remain essentially the same for both low and high pressure systems.
In reality, the use of a less constrictive system is always preferred, but that does not demand large, low pressure pumps. What it does include is the ability to use much better, larger radiators in small systems.
Getting the heat away from the radiator is the weak link, whether the coolant is high or low pressure, with high or low flow rates.
The main point of any system is two stage: To pass cool water over the hot spots and cool the unit by absorbing the heat thus raising the temperature of the coolant, then transport the heat away. Then cool the hot coolant and return it to the system.
Both high or low pressure systems can transport the same amount of heat into the coolant over the processor heat sinks, then through the radiator, with catalogues often quoting flow rates around 20 to 60 litres per minute. I consider this unnecessarily high.

There is a fundamental thermodynamic problem with water-cooling.
As alluded to above, all factors being normal with room temperature and constant volume caused by normal atmospheric pressure, then the cooling temperature across the radiator is the limiting factor. No coolant will reach the same temperature of the cooling airflow, for there will always be inefficiency. Worse still, as the coolant is brought down near to room temperature, then the difference between coolant and the air is so slight that little heat transfer is possible. Therefore getting the maximum temperature difference across the radiator by using cooler air, will need the area to be carefully checked to get the lowest possible running temperatures.
Your water-cooled CPU is not going to get cooler than room temperature. (Unless using phase change, such as water surface evaporation and other esoteric techniques.)

The next point is to make sure the CPU gets the coolest coolant, so ideally takes the coolant from a radiator at the bottom of a case of a well designed system straight to the CPU. Therefore it can be seen that the radiator is the main area of importance to get the lowest CPU temperature. The coolest air in the room is usually at floor level, as hot air rises. If looking to drop a few degrees more than anyone else, then by making the coolant those few extra degrees lower will hopefully help to get the CPU a degree or so lower. If you live in a centrally heated house, tough luck.

If, like many of us poor Brits, and making your system from bits and pieces, then always invest in a decent pump, although modifying or making your own is easy and cheap. You may suffer with a poor radiator, but you must have a totally reliable pump. (Apart from one exception without a pump, as discussed later.)

Pumps have evolved greatly in the last year or so and although the best commercially available are still the same models as used in garden ponds for the last few decades, the better models are now becoming available in computer water-cooling kits. You can always simply pop down to your local garden centre, or read on.

The latest pumps have variable speed settings. You can always modify them yourself or make your own or get new pumps for less than 6 six quid, or for free. An average pump for a computer will pump something in the region of twenty litres a minute or more, but it is how reliably and how efficiently the heat is transported into and out of the coolant medium that is important. For overclocking very hot processors and GPU's, then look for a larger flow pump and larger bore piping. If it has a higher flow rate but a small bore pipe, then you must have higher pressure and this may cause unnecessary problems.
Pumps with variable speeds and flow rates for pennies are discussed in detail later.

The most worrying parts of some systems I have seen is simple things; like not getting air out of the CPU header. Also ensuring no potential bubbles build up in the radiator and perhaps a lack of reservoir.
Very poor radiator design, positioning, use and equally bad fan or radiator airflow is common. Commercial watercooling systems often have all these problems. So what is also desperately needed, is a simple means for getting the system installed, which should be easier than that seen on some systems. Convenience and neatness is still a major problem.

Radiators.

Types of radiators.
There are cross flow, where the coolant goes from one side to the other, or from top to bottom. Also possible is the snake piped radiators, where the coolant flows through a series of S bends through the finning. Either is applicable, but for compact systems the U bend, and for larger systems, the easier one-way cross flow is often used.

Other forms of radiator are discussed later. If you think radiators are boring then you have lots to learn.

Various designs of radiators have the inlet and exit on the same side of the matrix by using a segregated header. This causes the coolant to flow in a U, from the inlet, through half the matrix pipes to the other side, then return through the other half of the matrix to the exit header. But where the hot fins of the inlet are connected to the colder exit area, then heat transfer will not be ideal. Where the coolant makes a double pass through the radiator and ensures there are fewer dead areas in the coolant flow, but also increases internal resistance to the coolant flow.
Radiators should be positioned with the exit at the top to remove bubbles, whether the matrix pipes are horizontal or vertical. Horizontally, the outlet would be at the top, with the inlet at the bottom, but is not theoretically perfect, but perfectly good enough for normal use. When the matrix pipes are vertical in such radiators, then both inlet and exit are at the top and will also work and vent perfectly well.
It is obvious that if using snake type of internal flow, then positioning the internal piping such that air can bubble its way up and out of the radiator is going to be a good idea. Positioning a radiator so that it captures many bubbles is very dangerous, yet I've seen many computers with radiators positioned to trap bubbles and reduce their efficiency.
It is not rocket science to do the job properly.

Without a pump, hotter water in a standard radiator rises to the top of the radiator, with the coolest flowing to the bottom. This is not very important in small systems, as the coolant flow rarely makes much difference in heat between top and bottom, but it may help.
The varying needs cause a conflict of needs between efficiency, neatness and accumulation of bubbles, so in most cases, simply have the outlet at the top or a transparent connection at the top for recognising any build up of bubbles, then be able to eliminate them.

Warning: Any system with both pipes on the base of the radiator will make the whole radiator a potential bubble trap and effectively ruin its capability to transfer heat should bubbles accumulate. As radiators are not transparent, then this is a constant, serious and hidden danger.
Therefore always have at least one upper pipe connection. If you must have both pipes at the bottom of the radiator, then I recommend you solder or epoxy a small brass pipe to the top of the radiator header and fit a clear vent pipe. If the system is sealed, then the pipe can also be sealed after purging the system, but at least a long, transparent pipe with indicator float will show if bubbles are accumulating.
The upper radiator main pipe can often be the inlet, which will allow any bubbles to be seen, but only when the system is shut down and the coolant flow is stopped. When shut down, any bubbles will accumulate and be seen before the system becomes too inefficient.
The CPU header is often the highest point in sealed systems, so bubbles will primarily accumulate here if the coolant flow is poor, but hidden bubbles in the radiator are also dangerous.
Be ye warned; any system may work perfectly well for the first week, but a poorly designed system could gradually become dangerous as you get complacent.
No matter what radiator I use, and I use many, it always gets positioned in the case with great care. No air gets trapped, and the airflow is unobstructed and cool. The exit airflow must also be free too.
If the fan or pump is mounted in the radiator, then ensure the radiator is secure, but free from vibration, possibly foam mounted, or using rubber grommets on the mounting screws.
Where the fan or pump is separate and ducted, then the radiator can be fixed in the case.

In most cases, the radiator is often to the bottom front or rear of the computer case, to be able to take advantage of the coolest air flow. Never allow the radiator to use recirculated air in the case, as this is a waste of effort. A little cardboard ducting on the upstream airflow to the case entry will help reduce coolant temperatures. In some cases, the radiator airflow can be kept separate from the rest of the computer. Far too many computers have atrocious radiator placements.
It is often preferable to have both radiator pipes curving neatly away from the motherboard side of the case, to allow plenty of room for the drive cables. With both inlet and exit pipes to the rear of the case, the matrix would often be horizontal. Where there is plenty of room in side the case, such as a midi tower case, then the matrix can be vertical or horizontal, with both pipes at the top.

To repeat: No radiator should be mounted with both pipes at the bottom. If you demand that both pipes are at the base, then epoxy or solder a small vent pipe to the upper water jacket of the radiator, with a pipe leading to a header tank or vent pipe.

CPU header.

For greatest efficiency, the coolest coolant should ideally enter the CPU header centrally, so the coolant is spread with the highest difference in temperatures for maximum heat transfer. In reality, the system will settle down to a happy heat regime across the system, with little need to get the last few hundredths of a degree out of the system.
It is not just the central flow which may be important, but also the surface area in contact. To get the maximum heat transfer, the base of the CPU header should be lapped to a fine finish, then heat transfer paste used.
To get maximum coolant area for heat transfer, the coolant should be passed over an internally finned CPU header base, so maximum surface area and flow can extract the heat as fast as possible. In reality, this is not needed, as most systems settle down to a working temperature, and such refinements make minor differences to the actual running temperatures, but it all helps towards making a more refined design.

Good exit flow is important, especially to expel air bubbles, but the actual internal flow is a compromise if the header is simply a hollow casing.
As will be mentioned later, making your own header is not very difficult. In the most expensive case, having a silversmith build you a special header is possible. This also allows for a finned coolant area over the CPU, so the maximum area for heat transfer is possible, but can only remove the heat that the radiator can remove. The excellent heat transfer properties of a well designed silver interface may help protect the investment of an expensive processor and to be able to overclock to the max, but a good radiator is imperative. (I design and have such silver items made to special order.) See other options later.

As many people cannot afford silver for the CPU header, then a solid block of copper with a soldered brass water jacket can be built for a few quid and an hours work. If a transparent cover is needed, then a sheet of thick acrylic or polycarbonate can be bonded then screwed into place, or a sheet of glass can be bonded and clamped to the water jacket. I prefer to rely on the clear piping for bubble recognition unless a complex header is used. More later.

Bubbles may accumulate at the top of the CPU header or high point in the system and these must be eliminated. This picture of a commercial design is utter crap, with bubbles even in the CPU header. What a waste of money. Therefore the coolant flow must be such that the CPU header outlet pipe is at the top. Ideally this would be designed to encourage any small bubbles to leave this important part of the system. The preference for a transparent CPU header casing is not just for looks, but as a visual check that all is well. The positioning of the commercially available design opposite leaves a lot to be desired.
Making your own CPU header is described later.

Pump.

Never place a pump high in the system. There is no theoretical reason why the pump cannot be mounted high in the system, then used to continually eliminate bubbles by placing the pump in a reservoir such as a plastic food box to solve some problems but it also causes some problems. In practice, unless you have shaft leaks, then place the pump low in the system, preferably with a header tank, so that any bubbles can work their way up and out of the system and kept away from the pump.

Warning: NEVER fit a sealed coolant system with pump at the top, because the pump vanes could become filled with air bubbles and thus not be able to pump any coolant.
In a sealed system, always fit the pump at the base, so that all available coolant in a leaky system will flow to the bottom and at least get pumped around, even though much of the upper system may have air in it; what coolant is left will still get pumped around and help protect the processors.

A cheap and effective primary test:

Build or buy just the CPU header and fit in into the computer with very long inlet and exit pipes.
Take the computer to kitchen sink or the garden shed, then make up a header tank for the system such as a water tank or bucket on the top of the typical midi tower computer case, and allow it to flow water through your system. Just two pipes from the CPU header, and suck the bottom to allow the water to flow down into the sink. You will have to fit the CPU cooler fan to the motherboards to ensure the Bios reads that there is a fan present.

A simple clamp restrictor on the bottom tube will suffice to reduce the flow. Completely clamping at the bottom retains coolant in the whole system. Relax the clamp to give approximately 5 litre (one UK gallon) per minute flow.
You can use a small bowl and sucking the bottom pipe to syphon the water to start it flowing, then using a jug to top up the upper reservoir is quite adequate.

When a steady flow, fire up the computer and allow it to settle down while running your hottest game. Measure a set amount of water over one minute, and measure the temperature compared to that of the incoming water flow.
The bios screen will show the CPU temperature and allow a comparison with normal running temps.

Get a reasonable flow rate, run the computer and measure the output flow when running a hot game. Allow it to stabilise and measure the heat output of the water with a thermometer. Then compare the inlet and exit temperatures and the flow per minute. The water will be cooler than you will get from a sealed system, as the water is colder from the tap if this is used, but the heat difference is important.
During the test, you can reduce the flow rate by constricting the pipe while keeping an eye on the CPU temperature.

Now carefully stop the flow and when the restriction is such that the CPU temperature begins to climb, then release slightly to get the ideal working temperature with the minimum flow. Measure the restricted flow rate by timing into a measuring jug. You may be pleasantly surprised just how small a coolant flow is needed.
This is with cold water. If the ideal CPU running temp is 35 degrees, then add hot water to the bucket to bring the water to 30 degrees. Now run the test again and see how much coolant is needed to maintain the CPU at 35 degrees. Again, wait until the CPU reaches 35 degrees then slowly release the restrictor to get a stabilised coolant flow at a constant 30 degree CPU. This should not need much coolant flow to stablise.
You now have a very good working reference for your choice of CPU header - the flow rate and working coolant temperature. By using warm water close to the CPU temp, then this is a worst case scenario and thus the actual system may run much cooler.

This is the best way to prove that you only need a small pump flow. You may want to throttle the flow by partial clamping at the base of the pipe work using a clothes peg or other simple clamp or partial plug.
You may wish to further restrict the flow until the CPU runs at a happy temperature, then see how much flow is actually needed. I was happy with less than 1 litre per minute. If you have an intel, then you may need a higher flow rate.

After the test, you can tell from the computer running temperature the optimum and minimum flow for a decent CPU temperature.
Although the tap water is probably cooler than the room temperature, the difference between inlet and exit flow will give you a good idea of the difference in temperature across this system and at the various measured flow rates.

Running temperature:
If the inlet temperature is 30 degrees and the exit temp is 32 degrees, then you have the possibility of having a system capable of running at 2 degrees above normal room temperature.
In practice, the inlet water would run slightly hotter in a normal, closed cooling system, so the actual temperature would be more then the 2 degrees above room temperature. It may be four degrees above room temperature, due to the computer settling down with slightly warmer coolant. But the difference in temperature from this basic test will give a good idea of the working temperature to be expected from the system.
If you used a jug to refill the upper tank from the warmer bottom on, then you have essentially made a closed coolant loop and your readings will be closer that to be expected in a fully enclosed system.

When testing this way, by restricting the flow rate until the CPU temperature begins to rise, you will also have measured the minimal flow rate needed for the system and also found an optimum flow rate, where any extra coolant flow does not lower the CPU temperature.

During the test, adjust the flow rate by measuring the temperature rise of the water as a guide as to whether you need to go for a bigger coolant flow pump or not.
Never use tap pressure, but a head (pressure) equal to the height of a typical small garden pump - then the flow through the CPU header and any radiator will give the true resistance of the system.

You should now have an expected temperature rise for the optimum flow rate. From this you can decide if water cooling is worth the effort.
You can now decide if you need to return to the garden centre for a pump, or brave enough to go to the car parts department to try modifying a new, six quid screen washer pump capable of just 1 litre per minute.

Plumbing.

Remember that this game is to remove the greatest amount of heat. Although the differences in pump position may only be a degree or so between the various configurations, it is up to the builder to decide what is required: Generally, the parts are mounted in the case first, and only then are the prettiest or neatest pipe routes decided, - but just swapping a couple of pipes can make a safer system.

In practice, the plastic pipes do not bend too easily, especially when cold during assembly, so it is best to assemble parts of the system with their pipes, such as the CPU so that the builder can see where the pipes can easily fit and what sort of bends and lengths are best for the system.

I always fit the pipes to my CPU header, with the exit pipe high to allow bubbles to be removed, then see where the pipes want to lie inside the case. Then I choose the best position for the pump, without undue force on the CPU header from tight bends in the pipes. In most situations, the easiest and safest is to position the pump low in the case. The layout of the typical system is such that the CPU is usually the highest point, with the radiator and pump on the bottom of the case, for easy mounting and to keep most of the system cool and away from the electrics.

Wherever the system is at a slightly lower pressure, bubbles can creep in from low pressure zones, such as minor suction from poorly fitting connections, or a worn pump seal. The pipe seals can be cured with a smear of silicone sealant, but pump seals demand a new pump, a new seal (or two) or preferably a pump of better quality, or as a desperate measure - to submerge the pump underwater. See later.

If any pressure leak (outwards) is likely to occur, it would be in the higher pressure areas first.
If any bubbles occur (inwards) they would enter in the low pressure zones.
It is for this reason that I ensure the driveshaft seal is on the low pressure side of my designs of pump, and that there is a bubble removal design of plumbing in my systems. This 'belt and braces' approach eliminates any bubble problems that may occur and also reduces any potential for leaks in the coolant.
I far prefer bubbles to enter the system, then be purged, rather than for any coolant to leak out. Luckily, in a truly sealed system, the volume of coolant will not allow much air to enter, especially when it gets warm and expands slightly pressurising the system by flexing the warm pipes.

Making and setting up the system.
The pump can be fitted on the base of the machine, but if feeling less than confident, can also be mounted in a small plastic drip tray.
Always dry fit the CPU header to see which of the various ways around it will fit.
Always have the CPU header exit pipe in the highest position to remove any bubbles. To optimise the cooling, I always mount my motherboards upside down, BTX style, but I build my own cases from scratch, so that my systems are as good as possible.

Getting cooler air across the radiator is a primary part of the exercise. Even a few pieces of corn flakes packet and sticky tape will help towards cooler airflow. If you live in a centrally heated house with double glazing, then tough luck. The radiator is usually easily fitted near the base, but never allow dust to accumulate if in a dusty room, especially if the case is positioned on the floor.
Where the radiator is not mounted close to the case entry, then always duct the radiator to the outside, to prevent using recirculating warmer air in the case.
Ideally, the coolest airflow would be from near an outside house vent from a sheltered part of the house, but this is probably not worth the effort unless living in Finland and wanting sub-zero air across the radiator, and of course, you use antifreeze and that there is no chance from ice, snow or real lemmings from entering the ducting.
If your computer is to be a permanent fixture, then running a couple of small pipes through a wall or window frame to allow the radiator to be mounted outside in the shade and breeze may be applicable, but never induce too much restriction in the coolant flow. Make sure the warm and cool pipes are insulated from each other, even if only by a piece of cardboard.

Water boxes.
Completely submerging some types of pump in coolant is not needed unless demanding total protection or suffering from poor shaft seals. Making your own submerged pumps is easy as it allows any motor to be used and a simple shaft to a basic, submerged pump body, with the motor above the water level. See later.
If submerging or partially submerging the pump then consider using a 'Tupperware'(tm) style of food container and always fit the waterproof lid. A couple of rubber grommets from a car shop will seal the inlet and outlet pipes if they are through the lid or through the higher sides of the box.
Making an air hole, small bore vent pipe and filter will greatly reduce bacterial contamination of such 'open' systems. An air vent filter is easily made from a length of small length of pipe and inserting a piece of cotton wool. Adding some bath sponge prevents sloshing. Cutting the foam with a harder upper mounting flange of plastic glued in place, makes for a splash skirt and makes for a quieter system which mounts the motor free of vibration.

If wanting to eliminate bacterial ingress, use distilled water, or boil the coolant first for five minutes to destroy all bugs, or add one water purifying tablet to kill off any nasty bugs. Preferably soak any cotton vent in an antibacterial liquid, as used for wiping kitchen work surfaces. If carrying the machine around, then use open cell bathroom sponge as part of any reservoir to prevent sloshing, as this is so much neater than anti slosh baffles and far quieter too. If air contamination is common, then simply float a film of polypropylene on the surface of the coolant. (Polypropylene floats.)

Dummy runs.
Just like fixing hard drive cables, always make dummy runs and get the pipes perfectly aligned. Always try to eliminate any potential high points where bubbles may be trapped. Once the whole piping is sorted, then build the system dry, assembled into the computer to check the fitting and the run of the pipes. As most pipes are soft plastic, never allow them to rub against any abrasive areas, especially the sharp edges of some cheaper cases.
When cutting the plastic pipes, use a sharp knife and try not to squash the tube nor distort it, so it makes a neat fit into some of the cheaper connectors.
When satisfied, remove from the computer and fill the system. Run the system for a couple of hours away from the computer, using a 5 volt or 12 volt supply or whatever is required. If the pump is not supplied with its own transformer, then an old computer power supply will do nicely, or a 12 volt car or bike battery charger if appropriate. Details of building variable controls are mentioned later.

While running, use this time to improve the system. Move the components around to remove any potential bubbles or to get them to a high point were they can be removed by pouring in more coolant. While testing, the pump and radiator can be mounted on the table using blue tack or plasticiene, with the CPU header held for a while in its relevant position with a piece of string to check for any air entrapment problems.

Don't just test the cooling system, check the power supply if a transformer is used to power the pump motor. Many cheap power supplies fail due to overheating. I always open the casings, then carefully add air vent holes to help keep the transformers from cooking.
Remember that you are building a long and involved way to protect the CPU. A chain is only as good as it weakest link. Make all links as reliable as possible.
Bomb-proof is your lowest acceptable standard.

Removing any air bubbles is done by simply making the outlet of each component at the highest point and looking through the clear piping. Eventually all the bubbles will be in one convenient place, when the pipe carefully opened and topped up to remove all traces of air. If you are having lots of problems, then the final purging could be done under water, but this is rarely necessary.

If using a high pressure system without self sealing connections or a poor design, then sometimes it may be preferable to wire the pipes in place, although small hose clips are also possible. These clips are available in metal or plastic. If using tie-wraps, always use two on each connection, with the joins staggered, as the area directly under the join can make a poor seal.

Fully dry the outside of the system, run for another couple of hours and check for leaks. Use any case modding UV neon's to look for leaks if any coolant additive is UV sensitive.

When confident, install the system using the usual CPU heat transfer paste with any other careful considerations and alignment processes.
If a noisy pump, you may wish to use double sided, sticky foam pads, as used for model aircraft servo fixings or to mount it in foam to isolate it from the case.
In many cases, the radiator fan may be run from the motherboard. In other cases, the radiator fan may take its power from the hard drive Molex connectors, when your radiator fan sensor wire (usually yellow) should be connected to the motherboard to give a shut down warning. Technically speaking, the water pump is essentially the same as the CPU fan, but these rarely have a yellow sense wire for the motherboard CPU fan connection. So use a fan with a yellow sensor wire on the radiator to become the feedback for the FAN1 connection on the motherboard. This is needed, as most motherboards will dislike having no pulse from what it thinks is the CPU cooler.

When satisfied and all is well after two full checks, refit into the computer and run the pump and radiator fan for a while just to be sure, then switch on the computer.
Check that it all looks and sounds well, but be prepared to switch off immediately.

Run the CPU temperature programme, it may be in the BIOS or a separate programme on top of your operating system. Water-cooling systems take a long while to stabilise, so do not take the first temperature as the normal temperature. Preferably run the computer using your most intensive game for half an hour, then check the temperature. It is assumed your BIOS over-temperature warning or shutdown is active. If all is well, then adjust the radiator fan speed to a preferred level for noise vs. heat flow. Allow five minutes for the system to stabilise at the new fan setting.

Where you wish to run your fastest game and cannot check the CPU temperature at the same time, then set the maximum BIOS temperature very low to give you feedback. Alternatively, you can fit a temperature sensor to the CPU header or use an IR thermal sensor. Alternatively shut down the game or reboot into the BIOS for an instant assessment. Remember that BIOS temperature sensors are not so accurate as one would like, but perfectly acceptable for comparative tests.

Now test the temperature limits by running any overclocking settings and your most intensive games, after all this is why you are water cooling. Then reduce the fan speed slowly to get the best compromise. If the pump can be adjusted, then adjust this too. Remember that pumps are far cheaper than CPU's so don't run them at full power as they are often built down to a price. A slower pump motor is usually a more reliable pump motor. It is for this reason I occasionally use custom high-spec racing motors, then de-tune them, or use high quality motors designed for steady, long term running. You may often find that there is little difference in CPU temperature between the motor running at full power and at tick over, - so always check, then always adjust for reliability.

Keep the temperature programme running for the first few days.
Check every hour for any bubble collection denoting a slight leak on the suction side of the system and then daily then weekly. Also check for any coolant loss. Learn to listen for any audible changes in the coolant pump should it begin to fail early.

Finally decide if the whole exercise was worth the effort.

Compare the running temperatures with the one you wrote on the case in pencil before you began. If the cooling is not much different to air cooling, then consider removing or modifying the system.
You may find that your standard commercial system offers little or no advantage over a good air cooler. Do not worry, as the commercial systems can be improved. - Improved to the point where only the power plug remains !

Modding your coolant system.

Many people like to tinker with their computer systems, it's a big toy and lots of fun, and for some, a chance to pose.
Water-cooling is no different.
There are many small or large ways to add to, or improve your basic water cooling system. But improvements are only improvements if you have problems, otherwise they are called unnecessary complexities. The biggest problem or disappointment is spending lots of cash, with little improvement in the cooling. If all works well, then there may be a few minor problems, such as a slight leak, noise or ugliness.

Fitting a header tank or reservoir.
Perhaps your system has a small leak. Make sure you solve the problem and not apply a reservoir tank. Most water-cooling systems are sealed and work perfectly well when all bubbles are removed and do not appear afterwards. In a few cases, bubbles may enter the system, but this is not a problem if they are removed automatically, perhaps using a vent or header tank. Where coolant is being lost, then solve the problem.

A header tank is a passive device containing a higher level of coolant and only acts to prevent coolant loss and to give a small 'head' of slight pressure in the system. If properly designed, will also allow bubbles to move up to the header tank and out of the system.
A reservoir is usually an active part of the system and the coolant in the reservoir is often in the general flow of the coolant. A reservoir is common if using a submerged coolant pump.
In common practice, be it domestic heating or supplying water to a large city, the header tank is also a reservoir.

If building your own system or modding, then perhaps some air bubbles will naturally float to the upper part of the system. As the coolant exits from the CPU header which is often a high part of the system, then it is possible to place a T piece in here and have this going up to a small header tank. This would encourage all bubbles from this high point in the system into the even higher header tank. The higher coolant level will also give a reliable signal of any coolant loss over the years. If you never check these things, but simply make an occasional glance, then mark the level when the machine is warm, so you can check it at any time it's being used.
Note: Due to expansion and contraction of the various components, the true coolant level should always be checked when cool, before the computer is run in the morning, so the relative levels will give true indications of coolant level.
The header pipe will connect to a miniature reservoir which is open to the atmosphere top prevent pressure or vacuum building up in the system, and will thus be at the top of the system to give a 'head' of coolant. As most systems rarely leak, then any loss will be minimal. A simple length of pipe would be more than adequate for a header 'tank', but mainly to act as a high point and thus isolate any bubbles in a problematic system. I often use a small, clear plastic car fuel filter at the top to act as a minor reservoir and also keep out the dirt.

In low pressure systems, a suitable tall header pipe can be on the pressure side, but it is always best to place the header pipe on the suction side of the pump or at least in a low pressure section of the plumbing. Therefore the plumbing could be different if using either a header pipe and reservoir.

Warning: Always fit pipes to open atmosphere in the low pressure side of a plumbing system, never in a high pressure side for obvious reasons. Place your header pipe on the suction side of the water pump, so that, should any system blockage occur, the water will not overflow.
BUT: In some situations where the builder wants INSTANT warning from the coolant, it may be possible to build the system 'wrong'. If the header pipe is placed on the pressure side of the pump, then any blockage after the header pipe, such as in the radiator or CPU header, will cause the coolant to be pumped out of this higher point which is open to the atmosphere. Building this to indicate by splashing over the user, or to fill a water sensing device connected to the computers 'off' button makes for an excellent safety system. If you prefer such an open pipe and it is suitably positioned, this can act as a safety warning device. If such a blockage occurred, then having the header pipe exiting at the front of the case will give an instant alert to imminent failure. The heat reserve in the header coolant should be enough to allow the machine to be shut down within twenty seconds or so. It is always best to let the system warn you, rather than you having to keep a regular eye on it. If you fit an over flow float valve, then this can switch your machine off using a very simple second connection to the ATX power button.
The switch will need to be kept down for four seconds or more, which is not a problem on a float valve, especially if using just a brass contact and a piece of wire to make the contacts. These can also be connected to the power switch wires of an ATX motherboard and all is safe should the coolant get low or spring a leak.
The ATX power switch can be controlled by connecting to two pins to be shorted by a metal object. A closed cell foam ball covered with aluminium cooking foil inside the header pipe, with two pins pushed into the pipe above the level for an overflow and two pins pushed in below the minimal coolant level will give a simple and effective safety system for blockage and coolant loss.

Placing the header pipe just before the pump in a low point in the system, would not eliminate bubbles. Therefore the system will need to have the header pipe at a high point, perhaps just after the CPU header exit. This will ensure the header pipe is in a safe, neutral pressure zone in the system, between CPU header and radiator. Therefore this system would be pump-CPU-radiator, with the header pipe on the neutral pressure area between CPU and radiator on the highest point in the system to remove bubbles and also be on the safe side of the pump.

Any coolant reservoir can be used, but just a simple length of coolant pipe will do, as the coolant loss should be zero or thereabouts. Just because bubbles are entering the system, does not mean that coolant is leaking out. It is this reason I prefer low pressure systems for minimal coolant loss. On a properly built system, only evaporation will cause long term problems and this can be minimised to give extreme longevity.
For a filter, the open end of the pipe can be simply fitted with a cotton wool stopper to prevent evaporation, yet allow atmospheric pressure to prevent an over pressurised system. Another good choice is to use a small, transparent car fuel filter with a paper filter element. If carrying the machine around, then use open cell bathroom sponge as part of the reservoir to prevent sloshing. If air contamination is common, place a film of polypropylene on the surface of the coolant. (Polypropylene floats.)

Visual warnings.

There is nothing to stop the header pipes running outside of the case, or beside the window against a contrasting or zig zag background to allow constant coolant level inspection. This would be particularly good if you mount your ATX motherboard upside down with the CPU and thus the coolant system low in the case, to give a nice, tall header pipe.

For those who do not trust pump motors, a light could be fitted across the wires to the pump, allowing a warning light to be activated if the pump stops turning. Alternatively, a small cheap, digital tachometer could be used and mounted on the front of the case, perhaps beside the motor speed controller. See later. The fan would be monitored by the motherboard in the usual manner. Building in a pulse wire which can interface to the motherboard second fan connection will also offer an extra warning device. This could also be interfaced to a number display, or my preference, a colour bargraph display.
Even a simple cardboard rotating disc pushed onto the end of the shaft with a spiral drawn on it will give a perfectly usable visual warning device.

Neatness is difficult with plastic pipes.
If needing a sharp bend in a pipe, simply and neatly wrap a thick copper or a finer, steel wire around the outside of the pipe and bend to shape. A small welding rod with the flux removed, or a small knitting needle or wire coat hanger is ideal. As bending around the pipe is not recommended, as it can be crushed, always pre wrap the wire using a piece of doweling. Adding a flourish of a tie-wrap loop or a stand off mounting can make for a neater system.
Another way to make neat bends is to slip small plastic angle bends over the pipe to allow neater corners. Alternatively use brass pipes, but making such bends is not easy unless a trumpet or saxophone maker. In such cases, simply make the brass pipe with an open middle section and slide the pipe in place, then bend the centre, with the two sleeve ends holding the pipe. See picture. You may wish to solder on some support brackets before inserting and bending the plastic pipe, to help make a really neat case. This also makes the CPU and GPU headers far easier to position in the case.

At this point the reader should have a system working reasonably well. It may even seem perfect, even if you bought all the most expensive bits from the best catalogues and websites.
But a much nicer systems is most definitely possible.

A truly reliable and basic design.

The simple car system is where there was no water pump. Just a smooth, large bore pipe leading up to the radiator with a decent fan. Because the system was designed for hot water to rise easily and the cold water descend to fill its place, the whole system can be simple and effective, reliable and quiet. Adding a simple large bore radiator and a suitably designed CPU header with one pipe high in the header and the other low, would make a very quiet computer which would only have the one radiator fan and no inherent mechanical problems. Always use a radiator with the pipes vertical and the hot riser entering the top of the radiator.
Surprise, surprise, this will work better if the case has the motherboard upside down, BTX style, with the CPU at the bottom and the simple, single pass radiator mounted high in the case.
Because it works without a pump, you can add some flaky chrome pieces of neutrally buoyant plastic which will highlight and check the coolant flow. Try scraping the back of a toasted CD then with light oil to get the buoyancy right.
If you have a tall case, with your CPU near the bottom with one pipe low and one high in the header, AND it is a large bore system, AND the radiator is a single design where the hot coolant can flow into the top and cool by flowing down though the radiator, - then remove the pump and give this a try. Use just use one fan set low on the radiator. The radiator is placed high and vertically in the case. Insulate the hot pipe to encourage hot water to rise.
It may only take half an hour to get the pipes running smoothly vertical. If it works, and your CPU keeps suitably cool, then you have no need for a pump and have a far more reliable and quiet system. More later.

Coolant additives.

Pretty flowing water.
Many people would like to see if the coolant is actually flowing. This is difficult, as placing plastic balls or other detritus in the coolant for observation is not going to do the pump vanes any good. If an indicator is needed, then fit a length of contrasting plastic thread in the pipe. Place carefully and securely in the system, held by both ends should it break, and placed in a long, straight length of pipe with a simple and non constricting clip. Never cause unnecessary restriction to the coolant flow. A snug fitting plastic sleeve with a piece of nylon sewing thread may suffice for most people who do not trust their pumps or who just want to see the pretty water flowing. A simple loop of nylon thread inserted at the same time as a clear pipe is fitted over the pump outlet would be ideal, plus a splodge of silicone sealant. As the cord will want to stay beside the side of the pipe, always make a knot in the loop, then keep the knot positioned in the middle of the flow, so the cord trailing downstream from the knot will remain in the central flow.

If you want a prettier indicator and your pump is a basic design, such as an old car screen washer unit, (see below) then you can but some glitter glue pens, which have the every fine flakes of glitter which should not clog the pump vanes, then wash the glitter glue and dry it to remove the glue. The flakes can then be added to the coolant and if illumined properly, can give a pretty, dynamic display for checking the coolant flow. As they are heavier than water, then they will sink to the bottom and hopefully re-enter the plumbing from the radiator, but probably best forgotten. I have experimented with them, but finally ended up with simply addling a whole lot of glitter and live with variable levels of flow indication.
In a few cases, directing a safe, 5 volt pocket laser onto the coolant flow will give some interesting effects on reflective glitter when it all works as intended. Using glitter flakes from larger sources may cause radiator or pump blockage. Adding a dye can have some effect on laser light.

If an open system, with the pump submerged in a reservoir, then the return pipe can simply cascade like a small waterfall into the reservoir to indicate that coolant flow is correct and also add a background noise indicator.

If you have a clear side panel, then this could also be the reservoir, where the coolant from the radiator enters at the top and the whole side panel becomes the coolant flow indicator. Making entry and drain holes in such panels is not easy, but plenty of epoxy resin and some car windscreen washer connectors will usually suffice. If building a waterwall (c), then you can add a vane indicator near the radiator exit pipe. See later.

A temporary solution (in more ways than one) to check coolant flow is to mix oil and water together as the coolant, this will separate when not used and if different colours, then will allow the first minute or so of pumping to be seen as flow. Eventually this will be mixed by the pump and may not be discernible until the system has stopped again. This indication is only applicable for the initial few seconds. Making sure the oil and the water are not easily emulsified will depend upon the oil and the design of pump. Try it if you must, but do not expect to be too successful, as heat, oil and water can make something akin to mayonnaise. Best used in waterwalls using thermo syphon.

Pump Circuit choices.

I do not like many power connections leading to the computer, when one will do. Keep it tidy.
If you decide to use a separate power supply for the coolant pump, which is highly recommended, then you may be able to open up the computer's power supply and add a couple of mains wires soldered to the back of the mains socket connection, then feed the wires out with the other PSU cables and add a fuse in this line. This can then be used to power a very reliable, overrated, understressed power supply transformer for the water pump.
I prefer to exit these wires directly from the bottom of the PSU case, through a separate grommet and add some alloy cooking foil around these AC mains wires to reduce frequency problems from any nearby 12 and 5 volt wires.

If the pumps' power supply has the three mains socket pins sticking out the back, then these can be cut off, sealed and the wiring connected internally for safety and a new, smaller mains connector used. This wire keeps the whole system integral in the computer and automatically switches the coolant pump on with the mains to power any specific power supply transformer for the pump. I always disassemble these secondary power supply cases to add some cooling holes, as they often fail due to excessive heat in the transformer case.

If you keep the computer switched on all the time, with just the ATX soft off switch used, then the monitor and motherboard are still alive. - But you may only want the pump to work with the standard, single small computer power switch which is connected to the motherboard on ATX systems. If so, then simply connect the mains power supply for the pump via a 12 volt relay switch for a couple of quid. This can then be operated via one of the hard drive power connectors to trigger the relay, without having the transformer and pump on all the time.

For long term reliability of a pump it can be reduced in power, as running flat-out is not always needed. Unless cooling the GPU and playing Call of Duty, Far Cry or Doom all day, the pump motor could be turned down using a rheostat, such as a ceramic potentiometer which can take the power rating of a small pump motor, or a simpler device such as a switch with a resistor in the line or to switch between the 5 volt and 12 volt rails, or using the switch in any separate power supply transformer. If using a powerful motor, something more than a video motor, then use a potentiometer for a rotary control with a TIP 120 power transistor which can handle up to 60 watts to control the motor, as described later. Reducing the motor power may also reduce noise.

You may notice on some air cooled CPU coolers, that by reducing the airflow, the heat transfer remains just as good, as slower, smoother airflow across the fins works more efficiently. My CPU runs at 42 degrees with the fan at 5,000 rpm and 43 degrees with the fan at 3,000 rpm - far quieter and the fan is probably far more reliable and less likely to collect dust. Likewise with my water-cooling systems, reducing the coolant flow rate may cause a smoother flow over the CPU header, reduce turbulence, slow the speed through the radiator and may thus work just as well. Experiment in this area to assess your system, but always remember that a water cooled system will take much longer to settle down to its working temperature than for an air cooled system.

If you decide on a separate power supply for the pump, then it may end up with one pair of insulated wires connected to the 240 volt wire from the PSU, with an internal fuse of about 1 amp rating. Another two wires leading to the HDD Molex connector with a micro relay integrated inside the case, and a third pair of wires leading to the potentiometer or Tip120 controller to adjust the motor speed. This is not exactly difficult, yet many commercial systems have just the motor connected to the HDD connector and can thus cause many problems if the motor fails or causes electrical noise in the power supply to the electronics, so always choose a safe way to power the pump. More later.

Building CPU and GPU headers.

So let's go a little further than just the CPU, as the costs of making a GPU water cooler when compared to buying the graphics card will be mere pennies.
GPU (Graphics processor unit) coolers are fraught with problems, as the memory is often superfast, overclocked and needs cooling too. Therefore it is probably not a good idea to water-cool the GPU unless considering the onboard memory too.

It's no one's fault, but for water-cooling, the graphics card is upside down on a midi tower. This is because of the superb backward compatibility back to the original PC, 1 MHz 8088 chip, when ye ancient legend hath it, cooling wast nay a problem.

All my custom built cases have standard ATX motherboards upside down to help CPU and GPU cooling, not unlike BTX, but far, far cheaper. It makes GPU cooling much safer. Build your own, it's very easy. See my companion monograph on the same website.

If wanting to cool a very expensive graphics processor and its memory, then always consider turning the motherboard upside down or desktop style, to help remove bubbles from near the chips. If lucky then you may only need to turn your ATX 'beige box' upside down into a pseudo BTX style, and still have it work perfectly well, albeit with a few minor adjustments.

Consider a second water cooling system so that if one system fails, you have not lost both your CPU and graphics card. It costs a lot to replace a CPU, but can be survived with a few tears. Having to replace both CPU and GPU may cause mental meltdown and chronic wallet failure. Using dual pumps also takes the load off a single system pump and makes life far easier for all items, while also increasing coolant flow due to far less restriction in the systems. Separate systems allow cooler coolant to both CPU and the GPU, rather than coolant first to the CPU, then warmer coolant to the GPU. It also allows a blue dye for CPU and a red dye system for GPU. After all, this may not be hidden.
If using just one system, then buy a system with a large bore and coolant flow, as any restriction in one part of the system is a restriction to the whole system.

Build your own CPU and GPU headers.
Please note: When disassembling the heat sinks from the graphics card, always make sure they can be replaced. Always be very careful with the small plastic retainers. As to serious overclocking and reviving pipelines, look to the wwweb for specific details on your GPU. Powerstrip is a good beginners overclocking tool.

WARNING: The heat sink compound may contain silver and when wiped off, this can short circuit the memory and GPU pins. So always clean off any heat sink compound INWARDS, towards the centre of the chips. I use a piece of card to protect any pins.
Although the latest memory chips are surface soldered and do not show pins, the heat compound can still work its way into this hidden area, so be careful. If in doubt, clean any remaining compound scrupulously with a safe solvent and a small, stiff, non plastic brush. Always make a note of the code numbers on the memory chips, especially the last letters, as this can tell you if they are the faster types, likewise the GPU codes.

If the GPU and memory are perfectly level in their upper faces, then simply cutting a large copper sheet to cover them will suffice. To check, put a little heat transfer paste, butter or grease on the GPU and memory chips, then place a sheet of plate glass over them. If you don't have this, then an old CD clear case cover. If the memory and GPU are all at the same level, the pattern of displaced grease will show a perfectly level pattern on all the items. It is important to put central pressure over the GPU to make sure this is the centrally level item.
If they are not all the same level, then you will not be able to build a single graphics card water-cooling jacket for GPU and memory, and separate items will have to be made.
If the differences are minor, then perhaps some aluminium or copper shims may be made to build up the height of the memory chips to the level of the GPU silicon die. Such shims are possible by measuring the difference in thickness between memory and GPU, using a set of feeler gauges, available from any car shop for a quid. Perhaps some carefully flattened and polished cola can could be used for thin shim, or look elsewhere for thicker shim. As the typical alloy beer or cola can is cold drawn to shape, it has inherent springiness, so keep it large, clean then mark with soap, then heat over a gas flame, until the soap turns brown, then plunge into cold water to make it malleable and easier to fit.
Never use more than one shim, so that the heat transfer path is as perfect as possible from the memory to the graphics cooler base. Make the shims then always double check using the grease and glass method. There may be a little room for alignment from the flexibility in the graphics printed circuit board, but only by a couple of thousandths of an inch, and only if decent clamps are used.
At a push, it may be possible to slot the copper so the sections over the memory can be slightly cranked to give perfect fitment, but this takes a long time to get perfect.
If in ANY doubt, then simply make separate coolers for the GPU and memory. Don't forget to check for memory on the reverse side of the graphics card.

A large copper sheet to cover it all will need a cardboard template, about 1/4 inch thick to absorb the heat and so it does not distort. (5mm thick.) Then off to your local stockist. If no copper, then aluminium will do quite nicely, especially if your system is being carried around. I always ask if I can rummage around the scrap bins, then buy anything likely for pennies.

I employ two ways to make GPU headers, the solid block and the built up version. Both work equally well, as they both have the same heat path. the solid block is better, as it gives a larger heat sink, should the system fail, but takes more engineering skill, and of course a big chunk of alloy. For those who only have access to a moderately thin sheets of alloy or copper, then a built up, sandwich version is also described later, but needing less engineering skill.

A large, square block of copper or aluminium will need to have the coolant channels added. This is best done by marking out on the sheet where you want your coolant to flow. Ideally, the coolant will enter centrally over the centre of the GPU, then flow outwards, evenly to the memory. Therefore marking out the positions will allow a well designed water casing to be built over the copper sheet.
Most graphics cards have a selection of holes perforating the printed circuit board. Look for those equally spaced over the memory chips and consider if this can be used to evenly clamp the coolers on the memory on both sides of the card, so look for a pair of equally opposing holes across the GPU for clamping the main cooler. Even a thin bar of piano wire will be strong enough to clamp them together using a spring and clamping sleeve on the wire to maintain clamping pressure.
Now make a full size sketch of the proposed coolers. Decide where the mountings and where any coolant and air entrapment problems may occur. First drill the main mounting holes to correspond with the graphics card mounting holes. Then lightly mark out the GPU and memory positions. Also ensure suitable clearance for the various components such as capacitors.

Where there are dead spots in the metal block, there is room for threaded holes for studs or screws into the heat sink sheet, and these will allow a sandwich of plastic to be clamped over the heat sink sheet. If you cannot tap a thread, then use countersunk screws protruding from below the heat sink sheet. These must be flush and not cause mounting problems. At least four, preferably six studs will ensure the water jacket sandwich will be secure.
To cheaply tap a thread, use a steel screw, then taper the ends and cut a hacksaw slot, lengthways along two sides, to make an improvised tap. This will suffice for soft metals such as copper or aluminium if done slowly and carefully.

Ideally, a thick slab of alloy would be used to carve out the water channels. Do not use plastic which may soften under heat, such as the lower melting point thermo plastics, if you must, then use thermo setting plastics.
As a thick sheet of plastic is not available for a water jacket, then it is easier to build up a sandwich jacket using 'plastic metal in a tube', available from most DIY and car shops. This can be built up around a cardboard former, with cling film over the copper sheet so it can be easily removed. This allows the mould to be tidied up prior to assembly. If preferring to bond to the copper or alloy, then simply bond it in place after roughing up the surface and scrupulous cleaning. During this process, the brass or plastic pipes must be inserted for best coolant flow and any smaller air vents included.
If keen on soldering, then the jacket could be made in brass or copper sheet. Use a high temperature plumbers solder if possible.
If wanting to view the coolant flow, then a flange can be moulded into the plastic metal, or soldered to the brass walls to allow a glass sheet to be siliconed in place and then clamped. A transparent cover is highly recommended, if only to check that all is well.

Here is a basic diagram of a one piece GPU cooler, using a copper sheet and plastic metal in a tube walls with glass cover. The ones I build are not quite so neat, so a pretty picture is preferred. The middle section is just some car body filler, used to easily mould the water channels. If you have one plate over the GPU and another over the memory, then just build it with an interconnecting brass tubes for convenience.
Note: I have gone one step further and now mould the water jacket directly into the CPU and memory chips, using brass interconnecting tubes. But this is committing an expensive graphics card to a terrible fate if it does not work as expected.

Ideally the inlet will be directly over the GPU, or if room is tight or using a glass cover, then from above as shown with the black lines. Then the coolant flows around to the outlet at the top, above the mid point between the memory banks. Making internal baffles or coolant routes will ensure the coolant flows past the GPU area, then across each memory bank in turn.
Always make sure the coolant flows in a single path. In this case, into the centre, directly over the GPU, then up to the right, down to the first memory bank, then exiting after the second, upper memory bank. Making two exit paths from the GPU, over two separate memory paths can allow one to flow easier than the other. It may make little difference in practice, but the chance for reduced flow in one channel will be the one that limits the overclocking potential.

Warning: Always remember that most graphics cards are positioned in mini towers, with the GPU on the LOWER side of the graphics card printed circuit board, so the bubbles will always want to be beside the hot spot in the heatsink and this is very dangerous. Never allow air bubbles to reside near the GPU or memory areas. This is the reason I now always mount my motherboards upside down on my home made custom cases, not dissimilar to BTX format and for better cooling of all components. Having the CPU low, the pump low and the radiator low, obviously makes for less piping, less head for the water to be pumped up to and a lot tidier. For more details, see my website on designing and building your own computer cases.

If the inlet pipe was to enter the GPU cooler from above, as shown with the black line, through the side wall, rather than the glass cover, then it could be positioned with its internal end lower in the casing, to give full flow though the GPU area, and also make for neater external plumbing. As the pipe would block the flow, the pipe would enter to the left in the picture. Another alternative would be to perforate, or slot the end of the pipe to more effectively direct the entering coolant over the GPU area. Just use your imagination for the many possibilities.
As plumbing is awkward, always have the pipes to and from the graphics card easy to mount and plumb in position. Also make them securely mounted to prevent unwanted distortion on the chips, form strained external piping. This will require making small brackets to restrain the pipes on the graphics card, and must always be done very carefully to prevent GPU track damage, yet maintain secure entry and exit piping. I find this the hardest part, so always make sure you can build up a nice soldered bracket for the pipes to and from the graphics card.

If you allow the coolant to flow in two directions to the outlet pipe, such as splitting the flow from the GPU to each memory bank, then this is asking for trouble, as there will always be an easier and a harder way through and one bank of memory may run warmer. If large differences in flow, then you can only overclock to the warmer running memory. You could simply make one big water area and cool all at once, but as the GPU is the hottest, then being able to put the coolest coolant straight over the 'hot spot' will help reduce the GPU temperature a little more.
You will also see in this example, I have included a small vent slot in the top of the GPU area, to allow any bubbles to vent up, out of the system. This may not be needed, as most GPU's are mounted horizontally in a mini tower case. In a desktop or micro system, it is a vitally worthwhile consideration should the graphics card be used in such positions where the unit is mounted vertically. The point to note here, is to know your system first, then eliminate or prevent problems before they arise.

Once the walls of the sandwich are made, a final layer is added and while still drying, the upper plastic or glass sheet is added with a piece of cling film to give a perfectly flat sealing joint face and also show up any thin areas.
Then the central part is bonded to the copper sheet with silicone sealant, in the same way as fishtanks, then the upper cover or clear plastic sheet added. If using a glass cover, then an alloy strap or plate is used to keep the glass in place, or a series of spring loaded edge clips, for which you will have decided the overall thickness beforehand, or simply remove the glass and file down or build up the thickness to suit. If plastic, then this can be drilled for the screws from the copper or alloy base to align with holes drilled though the upper cover. Minimal retaining forces are another advantage of a low pressure system.

When all is correct, then the cooler can be tested before actual use in the computer and pressure tested by plugging one hole and using a long length of pipe about eight feet to give an eight foot head of water to pressure test the device. Hang it out the window or hang the top of the pipe from the ceiling, then allow to stand for an hour to check for leaks.
I check my systems by submerging in a bowl of water and then blowing into it to look for air bubbles which are a more reliable check of any leakage.
If all works well, then it can be lapped prior to use and assembled onto the graphics card, using thermal transfer paste and careful tightening down. If all has been done well, there will be no distortion and the heat transfer from GPU and memory will be perfect. When replaced in the computer, it should be tested over many hours and checked for leaks, as the heat and poor mounting may cause some distortion.
Most graphics cards are mounted horizontally, with the best bits on the bottom. Viewing the coolant flow through the graphics card header is difficult, so a small plastic mirror placed at an angle underneath will help check the system is working properly and add an extra feature to the 'modded' machine. Including a white LED will also help. In a few expensive designs, you may have to replace the fan, to fool the graphics card or drivers that it is being cooled.

A single large GPU cooler makes for a neat design, but is often not possible.
The height of the various chips or a rouge capacitor is always a common problem. It is possible to buy an identical capacitor for pennies, - look for the capacity in microfarads, and the working voltage, - then solder on the new capacitor, but use longer legs to reposition it to clear the cooler, and sleeve the legs with plastic tubes from domestic or car wiring.

Therefore, three cooler units, one for the GPU and two for the memory units may be needed. This is very difficult and takes an awful amount of hard work, but is removable, should it not work as intended. This is engineering and most people will simply not want to play this game, so easier options are mentioned later. On standard cases with the graphics card in the poor position, I prefer this system, as it give the best cooling, with minimal bubble entrapment. Only on BTX format do I use other cooling systems. It is often easier to build your own BTX format case than do the following.
The trick of course, is to make a drawing of the GPU and memory layout, draw in the blocks, then decide the coolant flow to eliminate any bubbles, then ensure they all connect up easily without any excessive piping. Then decide the best placing for the pipes from the card to your radiator and pump. This is not so easy, especially if the graphics card is mounted vertically as in a micro system or desktop. In a mini tower, the graphics card is often horizontal, so the bubbles will probably work their way out. But never assume the bubbles will all leave happily. See the picture of all three items, below.

For simpler cooler units, just use a thick slab of aluminium (or if rich, copper) and drill holes for the inlet and exit pipes, which can be a tight push fit brass tube.
There are many forms of drilled metal slug. The simplest is a single hole - In one end and out the opposite end, but this does not give much time for the heat to transfer to the coolant. Doubling the distance will double the potential heat transfer. Therefore an internal U in the block may help. A more complex design is a Z layout with three lengthways holes, but this is not recommended as it resists the flow too much and is only suitable for the GPU block.
The Z design illustrated will have problems venting any bubbles unless placed to allow any trapped bubbles to flow upwards. You may notice that in the Z block I have added two very small cross holes which will help expel air even when the unit is placed in any position.
The drawing shows a transparent unit for illustration only. Do not attempt to build a heat block from clear plastic nor glass, as the heat transfer and material properties are not suitable.

The best GPU cooler would ideally have lots of internal channels, as shown in the CPU block under test in the picture opposite.
Many internal holes present a large surface area between block and coolant, to give greater heat transfer for the design. This is only possible if you are good at drilling straight holes. If your holes wander while drilling, then use a new drill bit, or sharpen your drill ends more accurately. Always ensure that the total cross section of the cross-drillings is slightly larger than the bore of the main coolant flow, so that the device does not constrict coolant flow. More on drilling holes later.

You may happen to notice my butane powered 'Prescott CPU emulator' in the picture. After many minutes, only a three degree rise was observed and was considered more than adequate for this approximate 400 watt heat source, - despite using a small waterwall radiator without active cooling and the pump running at very low speed. Yes, this 'simple' drilled block of alloy works a lot nicer than many CPU and GPU headers available commercially.

The illustration shows a design allowing all three items - GPU and two memory blocks to be positioned for best flow when mounted vertically. This can be hard to make and fit, so only attempt the memory if the main block was really easy to make. The plumbing must be neat, yet not interfere with the AGP or PCIe socket. The PC motherboard and case is designed to take very long cards, so the pipes could exit at the rear of the card for low profile desktops, but this brings the pipes close to the front of the case. In practice, the pipes often exit from the top of the graphics card, so the plumbing remains short and neat.
The headers must not be obstructed by the various capacitors and other items on the graphics card. I occasionally reposition awkward capacitors to make life easier. Ideally, both pipes would enter from the same rearward or upper side, according to your machine, but this is not always easy. See later.

The layout is typical of most graphics cards, with the memory banks to the right and also above the GPU. In this example, the coolest coolant enters the GPU first, as this is the hottest item, then exits high, to ensure no bubbles occur in this critical area. The coolant then flows first through one memory bank then the other, exiting at the top, where any potential bubbles could be seen to exit. The first memory block will cause a bubble trap and so this has a small cross drilling to allow any air to vet from the inlet side to the exit, but the small hole is not big enough to allow significant amounts of coolant to bypass the block.
You will note that I have not thought this sketch through well enough, as if I was clever, I would have extended the upper edge of the lower memory lump, and not have to have soldered up a brass pipe with an angle bend or need to use a curved plastic pipe. See small black lines. Luckily, small alloy blocks are cheap if you know where to look, as they are often scrap items from an engineering firm. My latest design uses a single L block with multiple channels. Therefore always think things through beforehand, because pencil and paper is far mightier than the hacksaw and drill and far cheaper too.
The other design point is making all the inlet and exit pipes line up, with a little gap between them. This gap is to keep plumbing neat, but still allow a small amount of flexibility in the plastic piping between each component, so that each item will be able to lie perfectly flat over their respective chips when clamped or bonded into position. If very close, then make the connections a single brass pipe, with a loose fit, and then epoxy the tube in place, with the coolers positioned over the GPU and memory for perfect alignment while the epoxy sets. If you do not like the short stub pipes, then simply mount the pipes outwards, then use short U bends of plastic piping.

With my latest designs, I reposition capacitors then simply carve just one large block of alloy for the memory on each side, then decide the best drill holes for the position of the card. After lapping the bases, use 2 part epoxy 'metal in a tube' to connect the tubes between GPU and memory. While setting, I clamp the items in position to allow for the differences in height for perfect heat transfer.

Once the design is decided, the alloy GPU block is drilled. As drilling neatly aligned holes is a pain for beginners, then start by making centre punch marks, an old nail will do. Then drill smaller holes, as these can be drilled larger later, especially if they do not wander from the intended path. Also drill the larger inlet and exit holes using a small drill, as there is a booby trap of drilling through the larger holes first, will cause the cross drillings to go off line. Therefore drill the many small cross holes first, then the main inlet and exit drill holes, all using a small drill bit. Then drill these holes to the required, larger diameter, where the smaller holes will act as a guide.
If your cross drill holes wander and cross into each other, don't worry, it just makes for a slightly more interesting cross coolant flow. Carefully clean out all holes and blow and wash clean. Dry thoroughly by warming, then plug the cross holes with alloy slugs, and / or epoxy resin or two-part, instant metal in a tube adhesive. Then clean out the tubes again, using a drill down the inlet and exit holes to clean out any excessive adhesive. Finally insert the brass inlet and exit pipes, using a snug push fit and more adhesive or instant metal. Pressure test by blowing air into the item whist underwater, as bubbles show up any small holes.
Tip: Use a little washing up soap to help 'wet' the internal surfaces when cleaning through, to help the bubbles leave the block easier. If the internals have an oily surface, then the small bubbles may remain during use in the computer.

If the GPU block cannot allow the water pipes to enter from the side, then you can place them at an angle, but this will need epoxy resin or 'plastic metal in a tube' alloy plugs to fill the holes. Always make sure the brass connecting pipes have sufficient length into the GPU alloy block to support them securely. Alternatively make tightly bent brass pipes by heating, then filling them with solder, bending over a soft wooden former, the melting out the solder. If you are making a supremely neat system, then fit the CPU and GPU header inlet and exit pipes last of all, so that any mounting of bends can be refined.

Very carefully lap the mating faces of the block on a piece of glass, using some brass polish or other abrasive cleaner. This will give a perfectly flat and clean surface for the heat paste. The pipes can now be connected to the pump and checked for good flow through the blocks. A simple syphoning test using two pipes and a jug of water will do the job, as for primary testing mentioned earlier.
Connect up all the bits and measure into a container over a minute. If it is appreciably lower than the intended flow, then perhaps you have too much drag in the system and be constricted. If you have a 1 litre container and testing over the sink then you can fill the container to check against the time taken.

Mounting the header can be a problem.
If finding a suitable spring clip is difficult, then make your own. I deconstruct the cheap little spring loaded plastic DIY clamps for their springs, (usually four for a quid), then make alloy or steel arms to use these springs. The final items are low profile and the springs are available in many sizes. If wanting to experiment first, then simply use the clips as bought, but always make sure they clamp over the centre of the GPU and memory, and if this is over glass or plastic then do not use a strong spring.
Where spring clips such as Terry Clips; flat as used for draughtsman drawing boards, or curved for holding tools.
If other mechanical restraining devices are not easy to retain the cooling blocks, then they can be bonded in place with thermal epoxy resin.
As glue is a one-way process, then it is better to use thermal paste on each item, but employ a dab of epoxy on the inner corner of each chip, which will allow a reasonable chance for removal without damage. Do not use superglue, as this sets instantly. This collection of items will need careful positioning and need an individual weight over each to ensure they are perfectly aligned to the chips while the epoxy sets hard. The faster setting five minute epoxy is less strong than true epoxy, but is plenty strong enough for holding this in place with a dab in the corners, with the thermal paste in the centre of each chip. It will also allow easier removal at a later date when selling the card at a car boot sale with the old cooling fan carefully replaced.
An option which is believed to work, is to mix thermal epoxy with equal amounts of normal thermal paste, to give a less strong bonding process.

The GPU is usually capable of being clamped in place with mechanical restraints acting through the printed circuit board. If making a mechanical clamp device and not bonding in place, then always add a central dimple, and a corresponding centre punch in the clamp, so the cooler remains accurately in position.
Personally, I prefer to glue just onto the inner corners of the memory chips, as this allows for differing expansion of the components with minimal distortion then clamp while setting. As there are often two memory chips, simply gluing just the inner corners will usually do the trick, but GPU coolers are best clamped using through bolts and a spring loaded arm. Don't forget to lap the alloy bases first and a dab of thermal grease.

This kind of stuff is not expensive, apart from the copper sheets or alloy blocks. The rest is simple hobby stuff, with a touch of engineering and common sense. Total cost should not exceed a fiver.

The picture of the graphics card also shows the suitability for making the memory block as a large L shape of alloy for convenience, but only if all the memory chips are level with each other and you have a spring clip which will allows for different expansion rates.
As can be seen, the interconnecting pipes are short, so full length plastic pipes will be needed, and these are best retained by using steel or copper wire to make three turns then twist the ends for a secure seal. Do not attempt to reduce the gaps, as this is needed to ensure the blocks seat perfectly on their chips. A minimum of 5mm is usually needed, especially between GPU and memory blocks. If possible, simply reposition any obstructing capacitors and build a large single memory block. A large L block is possible if your drilling is accurate. If not, then drilling from both ends to meet in the middle is easier, plus blocking both ends. Never let the brass pipe inserts block off the coolant flow, but always make sure the ends are chamfered where needed.
Another aspect of this design is to decide the routes of the pipes to and from the graphics card.
The main point to note is to ALWAYS sit down and make a full size drawing of the graphics card, the position and height of the GPU, memory and any obstructions. Then work out where any plumbing can be positioned, in a manner which can actually be physically made, in a manner which will work efficiently, reliably and safely.

When your coolers are ready, then carefully lap the surfaces so they will be perfectly flat and thus transfer the maximum heat through the thermal paste. Lapping is done by gradually smoothing the mounting surface to a perfectly flat surface to mount onto the chip. There are lapping kits which cost money and consist of gradually finer sheets of abrasive paper. I simply make an initial flattening using fine wet and dry, while it is placed on a sheet of plate glass, which is perfectly flat, then I use a commonly available metal polish. It's important to keep the metal block level when lapping, otherwise it will become slightly domed, although this is not a problem as the centre of the dome will be perfectly positioned for the reasonable heat transfer, but it is better to have perfect heat transfer from across the whole chip. To maintain flatness, use plate glass and use a small figure of eight motion, using all the plate glass, with the pressure always acting in the centre of the block.

The alloy cooling block has a problem, as many people do not have access to such blocks of metal, nor are very good at neat drilling. It also needs a strong mechanical retention device to hold the alloy in position on the motherboard. Its main advantage is that it is easily removable so you can return to the original manufacturers heat sink.

Another alternative is to make a simple plate to mount on the GPU and memory, then build up walls and a lid.

An easy option is to make simple brass or copper plates to mount onto the memory chips, then solder a carefully crafted brass pipe to these, so the coolant draws away the heat.
A single, long, snaking pipe is ideal for the memory chips, flattened slightly where it passes over the memory pads, so a little more heat is transferred with plenty of solder for the heat path. These are much easier to make, but needs to be done carefully, with a neat, flat wooden jig to ensure the soldering process keeps all the plates perfectly level, followed with a little lapping for the heat sink paste.
To make a snaking brass pipe, fill the pipe with lead or solder. The brass pipe can be crimped at one end, then heated as the lead is poured into the pipe. Then the pipe can be carefully bent around a wood former, until it passes over all four memory chips. The pipe is then heated to remove the lead and is a 14th century trumpet making technique.
Then the memory heat sink pads are laid out on a nice, flat wooden block, the pipe laid over and heavily soldered.

You can go further.

Such cards may need over clocking above and beyond the 'call of duty' and a far cry' from the manufacturers standard specs and 'doom' may ensue.

Only those who want to push an old card will be open to a little further, er, development as mentioned next. It is also much easier and far cheaper to build and far more efficient if done well. But it needs total commitment on the part of the card.
If you are running your GPU close to burn out, then you may wish to 'cool direct', but only if NOT mounted in a mini tower fashion. The motherboard must be used upside down, BTX style. The following has NO room for any bubbles.
I place a piece of plasticiene over the centre of the processor and memory chips, such that a wall can be built direct on the chips. Then build them up with interconnecting tubes for direct water cooling, - WITHOUT an intermediate alloy or copper heat skink. This gives direct water-cooling.
You must ensure the water is permanently against the chips, otherwise they WILL burn out, especially the GPU, so clean the surface of the chips appropriately.

WARNING: Some silicone dies are set above the body of the chip, so if you can imagine it upside down in a mini tower case, then there is a very dangerous place around it for air to accumulate. If so, then leave well alone, or use a desktop case, or like me, simply turn your motherboard upside down, BTX style in a home made case.
WARNING: Some graphics chips have surface mount capacitors and on their upper surfaces, along with some laser cut connections. Therefore it is important to prevent these from water, so apply a little lacquer.
Place a block of plasticine onto the chips which can later be removed, then build up a wall using a non conductive paste such as hard setting resin or 'plastic metal in a tube'. At least 70 percent of the memory chip upper surface must be open to the water, but in most cases, the wall can be built up on the surrounding mounting area. A slight taper, plus grease or cling film helps remove it after the wall has set. Perhaps you may wish to make a neat little wall first, and then bond this in place.
If you want to get really sexy with this system, then look up the art of 'fluidics', which was important when I got a job designing some or other cooling stuff for nukler reactors.
I never touch any GPU silicon, but build the wall around the larger surrounding mounting plate, although I apply a smear of silver thermal grease to protect the silicon from long term corrosion and the coolant is always distilled water.

I prefer to slightly rough up the plastic surface of the memory to make a better adhesion of the plastic metal in a tube. I recommend that you study the memory mounting closely, as some are hidden surface mounted, some are still using legs. Either way, you must prevent any non insulating material does not intrude between the chip connections. If room is tight, use a small bead of silicone sealant around the lower edge first, to seal the electrical connections, then use plastic metal in a tube on the actual chip. This way, any conductive materials will not short out any memory legs or circuitry, should a leak occur and drain into the soldered memory chip connections. Some hair lacquer over nearby components may also help after a very neat and clean build.

Always decide first where the pipes are to go for the layout in your machine. I always cool the GPU first, then the memory in such a way as to ensure good flow to all. A small vent pipe is often required from the GPU to the outlet on the last memory chip in the coolant run, to help remove air bubbles. If you place the exit of the small vent pipe just inside the larger exit pipe, then you have active purging.
Cut and layout the large bore pipes to make the most of your card design. Some graphics cards have the GPU and memory at 45 degrees to enable a very clean coolant run in any position and eliminate bubbles very easily. ALWAYS make sure they there can be no air bubbles. - With alloy blocks, the chips will still transport the heat from the chip surface and you can get away with a few bubbles in the system. But in this design you do not have this advantage, so there must be coolant against the face of the chip at all times.
It is possible to lay your pipes in position first, possibly squashing a brass tube to keep it all neat, then build up around the tube. I seriously recommend you place the interconnecting tubes close to the surface of the chips, as they will usually be upside down in an ATX midi tower case, so any bubbles MUST be removed easily. Careful pipe work pays dividends. Include smaller venting pipes or channels where needed, as the card may later be used in a micro system, desktop, ATX or BTX or special.

Another main point to be considered is to ensure the pipes are able to take the physical handling force applied by the plastic tubing to the pump and radiator. Rough handling when fitting the interconnecting plastic pipes between pump and radiator can cause problems. This can impart a high force and cause the coolant channels to be cracked or pull away. Therefore always fit a mechanical restraint for the exit and inlet piping to prevent any force being applied to the channelling. Always try pulling on the pipes to check before inserting the card in position. Also check with a test run for leaks and safe coolant flow and ensure perfect bubble removal in the intended alignment.
When all the walls and tubes are built, you can then use more paste on the walls, then use cling film and a sheet of glass for an inspection covering. Then silicone sealant and preferably a mechanical retaining bracket.

Although this seems a permanent solution, it can be carefully removed if it does not work out too well. Carefully applying a sanding disc by hand, then rubbing some fine sand paper over the memory chips will allow the original heat sinks to be replaced. The GPU surrounds will also need some neatness, but the GPU silicon itself should never have been touched.
It is important that the positioning of the tubes is very easy to get almost perfect, so the plumbing can be built to have pipes positioned for both mini tower and desktop layouts.
WARNING: Remember that most GPU's in a miditower are upside down, so be wary so that any bubbles will be away from the GPU. If you over clock the GPU too far, then check for steam bubbles by which time the chips have probably died. But the chips should never be allowed to get beyond 60 degrees. If you push too far and there are no bubbles, then at least you will know it was not through lack of direct cooling.

It may seem a bit of a pain to regularly check the state of the cooling for air entrapment over the GPU. The simple answer is to temporarily stick an old web cam up close and add a light. If it does not focus too well, then use a watchmakers eyepiece and a splodge of blue tacky putty as used in offices.
More simply, adding an LED in the wall of the GPU and a small plastic mirror at 45 degrees will allow the ultimate in GPU 'modding', where pretty lights are for proper and sincere reasons for a 'bona fide' modded computer.
Inside the GPU hollow, I would include two white LED's connected to the 5 volt rail to allow direct visual checking, plus a couple of thermo sensors should things get hectic. Cheap LCD digital temperature sensors for general purposes use can be run from independent 1.5 volt systems for direct measurement. Buy a handful when they are cheap.

You may want to put some heat sink clips on the larger capacitors too. They are available from electrical suppliers or make your own larger finned items using scissors to make a fined alloy device using scissors and a cola can, held in place with O rings or a nylon tie wrap.

This is not real modding, as it does not include pretty lights or fans - and it is very naughty, because you may also invalidate the manufacturers warranty :)
Please do not follow the above few paragraphs, even if you have a spare graphics card that you wish to torture slowly and constantly as if in Guatanamo Gulag. You know it's not right.

Have you been reading carefully like good little boys and girls ? Let's find out:

I'm unemployed and Anglo-Saxon, (gizzajob). For these sins, I was unfortunately trained as a teacher of technology and design, (B.Ed) and happen to have a science degree, (B.Sc) and HNC in refrigeration, a registered engineer blah, blah etc, and been modding computers since BBC'B' days, but like so many today, no job in this once great country, but now a shit hole called Britain.
So let's see if you have sense of humour too !
As a final test piece, here is a truly scrumptious X800 pro in a micro case.
Please note that the graphics card has its cooler removed, but only to expose the central GPU and the four memory chips on each side for clarity. (4 memory chips on the reverse side too.) The GPU is believed to use all its memory pipelines to XT spec by joining the laser cut in the top right hand side of the GPU, and has a bios flash to XT. 'Atitool' and 'Powerstrip' on the HDD shows it may be in fear of being overclocked.

For five points, draw a possible route for the water-cooling through the graphics card. Assume the plumbing exits at the rear of the case.
For five points, consider where you would put the pump.
For ten points, choose the THREE best positions for the radiator and the THREE types of fans which could be used.
Being such a small case, decide if you would you need any other considerations.

Give yourself extra points if you do NOT use any expensive materials or processes.

You will also get points if you can over clock the machine to the fullest extent of its potential for under TEN quid. This must include a radiator and pump.

If you do really well, I'll promise not to recommend you for any daft GNVQ.

If you decide not to water-cool, - and that's valid too - then separate GPU and memory finned heat sinks could be used. The GPU would have its own fan, but if ducted well, then the memory fins on the GPU side can be cooled via the GPU fan, simply by adding cooling fins to the memory chips and cutting slots in the cooler to ensure they are in the draught. I would probably not risk shimming on this chip. A small fan such as an old 486 fan can be used for the memory chips on the other side, using some cardboard ducting from cool external air which ca