Water cooling a PC using Household Plumbing

Note: I wrote the following in 2010. I ran two computers sharing one such system for about four years. Everything worked perfectly and it was only a lack of a suitable water cooling gpu block for a new graphics card that caused me to go in a different direction. This is a viable and proven method!

I am fed up with my noisy computers! Water cooling offers a quiet way to cool PC’s but can be incredibly expensive if you use a kit.

I’ve done a bit of plumbing in my time and I was wondering if computers could be cooled using off-the-shelf plumbing equipment instead of the expensive and sometimes poorly constructed computer-cooling gear. So I’ve used a normal household radiator hung on a wall, a central heating pump, some pipework and valves and as few PC-watercooling bits as I can, and now have a fully working, very efficient and super-quiet cooling system working for two computers.

My motives:

  • Quiet!
  • Economical to run (<40watt pump running 24/7, but nearly all the cpu/gpu and case fans not running for 2 computers. Overall I think it’s a saving)
  • Reliable. (I don’t want to be fiddling constantly)
  • Cheap to build using plumbing parts where possible.
  • Expandable to include extra computers or extra cooling. (Just add bits to the system)
  • Potential to overclock to eke out a longer lifespan before upgrading cpu/gpu.
  • Geekpoints!

Planning

Price comes in at around £110, but includes a lot of stuff I had already. The computers are both in a fitted cupboard with 5M long VGA and USB extensions for the desktop computer, so money could be saved by not buying a case.

The shopping list…

Item Details Price Pic
Pump Grundfos Super Selectric.
These mains-powered central heating pumps are rated at 40watts at the lowest speed setting, and I figure give a rough throughput of around 1500-2000 litres/hour but you can double or triple output just by flicking a switch. These are pretty much silent in operation, ceramic bearings, designed for continuous use. Even new these are around £70 and a serious pump.The first I tried was really noisy – I’d removed it from my CH for that reason some years ago. I took it apart and gave it a thorough and aggressive clean out – loads of rust and chunks of rubbish in the pump chamber. Once put back together – perfect, and silent! That was after four years operation and I’m happy enough to fit it to this system. The adapters either side of this have ballvalves so that if it fails, it can be swapped out without draining the system.
Free (I have two left over from my old heating system) They’re about £70 normally.
Radiator 500x500mm double walled wall-hung central heating radiator. Hey, it’s cost-effective and matches the decor 🙂 I was going to use a secondhand car radiator as they’re better at shifting heat but don’t think they look too nice in the house. This is the least researched aspect of the initial build – the output figures for these rads all assume a much higher water temp than I’m expecting and I honestly have no idea, pre-build, whether this will work. If not, I’ll be getting a second rad and slapping that up 🙂

The very first design had this hung outside the building, but it’s vulnerable out there and with winter starting it would be a waste of free heating. In a deluxe system there would be internal and external rads with a thermostatic switchover to balance. Freezing risk taken care of with normal car antifreeze which also serves to eliminate any algal growth.

A note about antifreeze: It’s poison and it stains. If you can avoid drinking it and getting it on your carpets, it has another significant benefit above the obvious anti-freezing quality: It’s a wetting and flow-aiding agent. Car engines with a large amount of antifreeze run significantly cooler than those with just water. I can’t see a reason why this wouldn’t also apply to this system.

£30.59 (Screwfix)
Fittings and backend hose. Plumbing fittings – The components are a mixture of JG Speedfit pushfit plumbing fittings, including reducers from the pump to 22mm, Speedfit pushfit connectors, liners and the radiator tappings. Also includes the 6-14mm jubilee clips. ~£30 (I had a fair selection to begin with – probably another £50’s worth) (Screwfix)
Backend
Hoses.
A mixture of 15 and 22mm Speedfit hose (Rated to 105’c at 3 bar – amazing!). For tidiness I would have used copper tubing- except I already had the hose leftover from my house re-plumbing some years before.

Also some 15mm copper

Free (15 and 22mm)
Interrnal Hoses ClearClear PVC Tubing Hose Tube 10mm 3/8″ -10 meter £11.95 (Ebay)
Blocks CPU Block 1 (Server, for an AMD X2 4600+) came with an old Thermaltake Aquarius II complete system.

All these blocks came off ebay.

£20 (Block 1 – server)
£5 (Block 2 – Desktop)
£10 (GPU – Desktop)
Hard drives These will be ambient cooled. Without so many fans in the case, airflow doesn’t go over them so it’s lid-off or decase for this to be good enough.

If vibration noise becomes noticable then I may suspend by elastic. I’ve used Antec’s soft-rubber mounts which although they help, do still transfer some vibration to the case. Elastic suspension is cheap and crazy enough for me to want to try… (Inspired by Mike Chin’s article) I’m not so bothered by seek noise, it’s more the low whine I want to eliminate – whether this will help on that, I’m unsure.

Plan B, if this is still too noisy, is to fit water coolers to them and bury them in foam and blankets.

Update: They’re just fine hanging by elastic. Very little noise or whine, good ambient cooling.

Free (Elastic bungee cord found in a cupboard)
Total:
£107

Preperation…

After I got the bits (a note about screwfix – genuine free next day delivery – like everyone else should be) – it became apparent Plan A was not going to work. The 10mm pushfit manifolds did not fit the 10mm pvc tubing. They’re designed for Speedfit, which is thicker walled, so when the inserts were put in there was still too much slack. I even tried bodging with a some PTFE tape, and while that worked, it was too fragile to consider a permanent solution. The manifolds also didn’t give me flexibility in controlling the flow that the service valves do.

Ok, piccy time! (Click ’em for bigger)


I originally bought two 22mm -> 4x10mm speedfit manifolds. Alas, this did not work. The pipe sizing was wrong, they leaked even after bodging with PTFE tape.

Not acceptable, clearly I need something better.

This is what I came up with from the big pile ‘o stuff left over from previous plumbing jobs. Ignore the arrows, I made up two identical branches out of copper.

The water from the pump comes in from the bottom in 15mm copper and branches out via service valves into clear PVC. I’ve got two branches on this one with and end-cap should I want to extend for 1 or 2 more later.

The service valves allow me not only to restrict the flow to each branch and balance them but also allow me to turn off a loop entirely. These ones are copper pushfit which does not play well with the PVC, so I have put copper on both sides. If they were compression I could have gone PVC straight from the service valve.

If you’re about to ask “Why didn’t you make the trunk of this in 22mm copper reducing to 15mm on each branch? Surely that would give a better spread and more equal pressures?” Then yes, you’re right, it would. However, I’ve done some rough calcs of flow and made the decision based on that and the fact I didn’t have much 22mm stuff in my shed. 🙂

In fact, in an ideal world, everything would be 22mm copper except where it couldn’t be. Even allowing for the fact that the radiator joints are 15mm it would still improve flow, lessen wear on the pump and increase the amount of water in the system thus increasing thermal mass.

A note on pipe joints.

I’m using a variety of pipes here. All sizes outside diameter.

  • 22 and 15mm JG Speedfit pipe
  • 15mm copper
  • 10mm PVC
  • 8mm silicone

The first two are easy – Speedfit pushfit fittings, copper pushfit (only because I had them) and 15mm compression. All standard plumbing gear; cheap, easily obtained, durable. The steel grab-rings in pushfit are designed to cut into the pipe. I don’t think it’s smart to use these with the PVC or silicone as they’re both softer and more likely to tear than the speedfit plastic pipe.

So for the last, it’s a case of compression fittings with internal liners. 10mm pvc accepts a normal 15mm liner once softened in hot water, and this allows it to make a snug and very good joint in a compression fitting (with olive). 8mm needs an 8mm liner and jubilee clamps, or barbs and clamps.

Liners are essential for any plastic pipework to stop it collapsing. Even if it seems tight when you first assemble it, any heat build up (and that’s what we’re doing here after all!) will soften the plastics. The white pipe here is good for seriously hot temperatures (well above boiling) and even that gets soft in heat. The PVC and silicones aren’t heat resistant and need extra care to ensure their joints are soft-proof.

Here is the Thermaltake CPU block with some 8mm silicone to the 10mm clear PVC. Note the jubilee clip which was to be doubled-up before real testing. The copper liner in the pipe is an 8mm one fitted inside the silicone.

Here’s some PVC sitting in a cup of hot water waiting for it to soften enough to push the liner (just to the right of the cup) into the end.

With the liner in place, the PVC can’t collapse even when a compression fitting and olive is tightened around it.

A white Speedfit 15mm liner is just to the left of the mug, but the copper is better with PVC.

Olives are those brass/copper rings at the top of this picture – they sit around the pipe and are pushed into the pipe and the compression fitting when the fitting is tightened. By being slightly softer than the fitting they deform to give a perfect high-pressure seal. They are single-use only though, so be sure before you fit!

The rough draft assembled in my workshop.

The black container on the left is a central heating header tank. I had one in the garden amongst the weeds so I used it.

This is the flow on speed I (40watts) through one block on the test rig. Plenty!

(Incidentally – increasing pump speed did not increase flow exponentially. It did increase pressure before the block (ie, the main bottleneck) but throughput was already at maximum on slowest speed)

I haven’t yet been able to get a good and cheap flow indicator that’s also silent. Obviously returning the water to the header tank like this would give a very audible level indicator, but it introduces a lot of air to the water and air causes corrosion from the steel rad. Anyway, this project’s about making things quieter.

Speaking of corrosion – I’m taking a risk on my system. I will probably add a central heating inhibitor as well as the antifreeze, and no doubt purists are reading this and already screeching about different metal types (steel, stainless steel, copper, bronze and cast iron) causing electrolytic corrosion but this is not really an issue with central heating systems so I’m hoping the same applies here if I can achieve a sealed water system.

The header tank is staying in the install by the way, although only connected by it’s drain point. It’s a handy filling point, an expansion vessel and a handy air-release valve!

This is why men need sheds. My wife would not be happy if I was doing this in the front room…

After this I was fairly happy in my plan and it was time to fit it!

Installation and Testing

Ok, time for the excuses. It’s messy, I’m a messy person…

This is working, cooling the white PC (server) with a cpu temp of around 28c (37c under full load using ‘stress -c 10’)

I’ll try and explain what’s going on here with more pics. The black plastic tub at the top is the reservoir – it allows me to fill the system, it gives somewhere for the air bubbles to escape. It also allows any thermal expansion if the water gets warmer. Without this expansion the pressure in the pipework would increase when the water got hotter and put more pressure on the joints, leading to leaks. (Interesting point – if you pressurise a water system you can increase the boiling temperature well above 100’c – good for car radiators, but if it gets that high in this system, something went badly wrong…)

Dog toy far left…

Rad on left. Note clip-on thermometer on the return at the bottom. Not very accurate, alas. To get real water temps you need sensors in the water path.

The copper going back up was initially the feed to the header tank. However, some air bubbles were getting trapped in the return, so I’ve taken a swept T in 15mm plastic (just to the right of the copper) up to the header tank. This catches the air bubbles which vent in the tank. The white pipe in the middle is the return to the radiator.

The water leaves the rad at the front here and takes the white 15mm plastic pipe along the back. On the far-right you can see it go 90′ upwards to the pump.

That wall is filthy! Because of the dogs shedding mud elsewhere in the house, the computers running 24/7 would suck phenomenal amounts of dust through this area coating everything in filth. Motherboards, walls, desks.

Anyway. Central heating pump mounted vertically and pumping upwards – this helps vent air bubbles which otherwise get trapped in the pump and make it noisy. The pipes in front of it are the return from the computers. At the T, the upper vents airbubbles to the reservoir while the lower feeds the rad.

After the pump, the water comes into the backmost of these two contraptions. The return comes out from the frontmost and heads off left to the rad.

Once this was on with the CPU block plumbed in, I filled with water, watched for leaks and didn’t see any. I ran the entire system on speed III for 24 hours to test at a higher pressure and still no leaks – time to get the server down from the loft!

The day’s running also eliminated all the occasional gurgling noises as the air gradually separated from the water. The only noise now is a low “thrum” just on the edge of hearing from the pump.

Installing the first computer

You see – putting servers in lofts is not always a great idea. This spider was happily catching flies that were crawling into the case for warmth through the missing blackplate!

This server records cctv via zoneminder, runs mythtv with two pci DVB-S tuners and a couple of extra usb DVB-T dongles. It’s also a dev platform for coding work and hosts a bunch of samba shares, including the main one for the Popcorn Hour in the lounge.

Running an AMD 4600 twin-core, 2tb of disks (for cctv and tv recording)

Ironically, this HS/Fan combination (Hiper variable 80mm fan on a stock HS) is one of the quietest ones I have, but still very loud – it has to go!

Yes, that PSU does have “May be faulty?” scribbled on it. Pulled from a flaky machine, it turned out to be a bad motherboard, so PSU re-used in this.

The old Thermaltake block resting in place. A smear of thermal compound. Unfortunately none of the clamps from the old Thermaltake system fitted the AM2 socket.

Those 10mm-8mm joints are deliberately kept outside of the case – still not entirely happy about those!

I took the crossbar clamp off the old Heatsink/Fan and used that to hold the block in place – perfect fit!

I tried to touch the corner of the Northbridge heatsink too as that’s now the hottest item in the box, but I don’t think it makes much difference.

The water’s going a bit rusty from the steel rad after a few days, not a problem but if it gets much worse I’m going to need some filtering of some kind. I’m hoping it’ll stabilise as the oxygen works its way out of the water. The colour is mostly from residue on the inside of the pipe – the water’s nowhere near that mucky, if you squeeze the pipe it disperses the residue.

Time to spin it up!

The PSU fan was now the only fan in the box, but with the desktop next door I couldn’t really tell how noiseless this is.

Server started normally, straight into BIOS to watch the thermal values. All very cool as you’d expect – the CPU’s not doing any work yet.

Can’t avoid it any more – boot into Debian.

“watch sensors” allows me to monitor temps in realtime. All seems okay so I let it do its thing for an hour and a half.

Machine was running around 1.0 load, mostly CPU (Zoneminder is heavy on cpu because of the motion detection).

Ambient temp in room is 20’c. Cpu temp is 23-28’c – better than expected! (One core is always 5c below the other, no idea why that is)

The radiator temp was only about 2’c above room temp, not enough to feel a difference.

One week on…

No changes, system very stable.

I’ve tested with the linux tool ‘stress -c 10’ which hammers the cpu with 100% load. Immediately the cpu temps rise by 10’c – and stay there. Happy with that, when it was around 50’c idle with air cooling.

If I turn off the pump, the temp slowly rises as you’d expect. I bottled out at 50’c after about 90 seconds and turned the pump back on, and the temp immediately dropped again. Theoretically, you could use convected water instead of a pump but you’d need to do some serious calcs on pipe bore and head.

So far I’m very happy. With the low temps it really does open up the possibility of overclocking but mainly I want to get the desktop plumbed up too, so scouring ebay for cheap blocks…

Oh, and I de-cased the server. Neater, less space, more surrounding air, easier to work on. Note I’ve stuck a blue HeatSink onto the existing Northbridge HS with thermal adhesive. That’s also getting warm so I know it’s working. Turns out the PCI cards didn’t need so much support, they’re pretty good like that. Oh, and the hard drives. The old case was also acting as a sounding board causing a large amount of noise on seeking. Now they’re suspended like this on corded elastic, the seek noise is only just detectable.

Update!

Now added my second PC to the other loop, CPU and GPU.

The GPU waterblock cost £10 off ebay and the CPU block was £5.

Temps on this machine in this location, uncased and not under load.

CPU: Before 39c. After, 27c.
GPU: Before 53c. After, 39c.

Proof!

Another pic showing the cpu and gpu blocks being tested for leaks before fitting them;

The Australia-shaped gpu cooler is a “modified” unit whose seals had failed and sombody has re-done with silicone. I inspected thoroughly and tested to 3 bar and they’re both good. Although half acrylic (ick), the gpu block contains a large amount of copper and the thermal mass is significant. And although intended for an Nvidia 8800 fits my PNY 9600 GT perfectly, same mounting points and covers both core and memory fine.

I do have a couple of chipset blocks too now, but waiting to see if they’re neccessary. The desktop’s northbridge runs at around 85c.

My room is again quiet. The only thing I can hear with the cupboard door closed is a very faint hum of the central heating pump. If you wanted to go further, you could place this a long way away – after all, these pumps are designed to push water around an entire house.

One month on…

After a month of both computers running 24/7, I closed them down to inspect the water, as the transparent pipes were pretty coated on the insides with rust. On draining some of the water though, it was totally clear. There was the odd bit of soft floating material (fungal or dust clumps?) but these wouldn’t affect anything so I returned the water to the system with some anti-freeze to kill any life in there. (Bad idea introducing fresh water with oxygen to a closed system – it sparks off a new bout of oxidation). Obviously the rust is coating internal components, but very slowly and not remaining suspended. The running temps are all lower than projected and I’ve overclocked the GPU by 40% and it’s still running at < 30’c in a room of ambient 20’c. Overall, I class this a total success!

This is a diagram of the final working system. (Click for bigger)

Future plans…

Case fans: I am considering de-casing one or both of these machines. They’re in a cupboard, relatively safe and if I do it’ll improve ventilation significantly. The only problem in decasing a computer that I can see is supporting the PCI cards securely enough. I might need to dig out some old cases and hack them up, or do something creative with metal and wood. Update: Both machines are decased and running free. Although not as secure as I like, the gpu even with massive cooler and two monitor leads feeding off it is still pretty solid.

PSU Fans: Can’t realistically see water cooling working here (although I know it’s been done). There are fanless PSU’s but they’re still very expensive, and I know that stopping a fan without adding cooling to a PSU will melt it pronto. I think PSU fans will have to stay, but worth paying a little extra for properly quiet PSU’s. This may be the big compromise. (Update: Possibly decasing that and increasing the heatsinks may work, but that means exposed mains electrical parts inside my cupboard and I’m not a big fan of that.
Update: I removed the server’s PSU fan as that was the loudest thing in the cupboard after the desktop was plumbed in. I replaced that with the Hiper fan from that server’s CPU, declocked to 5 volts. Absolutely silent now.

Chipset Cooling: Ambient cooling, perhaps with additional heatsinks where needed but I deliberately choose motherboards without chipset fans as they are always failing and very loud, and with watercooling the case temps are also reduced, so these should be better off just by taking the cpu heat out of the case instead of just churning it around.

Disclaimer!

I don’t recommend you do what I’ve done, not because it’s difficult – it’s well within anyone halfway handy – it’s just that if it goes wrong it you’ll have rusty water everywhere and your hardware may be damaged. But if you do decide to do it or something like it, this might help a little.

Precautions; Move as much electrical away from underneath, set safety cutouts in bios or software to throttle or shutdown your CPU if it gets too hot (if the water stops for example). Have buckets and mops handy for that first test…

Credits:

Heavily inspired by the guy in America who used his swimming pool as a way of cooling his servers. Sadly his original page appears to be gone, but here’s a taste.

Since writing this, I’ve found Vonslatt who’s done something similar and he’s even avoided a pump altogether!

Facebook Comments