Here’s my first water-cooling experience.
Aquarium water pump 400 liters/hour, 190 ATS (13 USD)
Plastic cap of a spray can, free
Wood 2 x 17x40x25 mm, free
Copper pipe OD 13 mm, 22 ATS (1.5 USD)
2 x 1.5 m from a 25 m garden hose, 10 ATS (0.7 USD)
25 l PE canister, free
The plastic cap is glued to the Celeron CPU with an inflow copper pipe
in the middle of the top pointing at the CPU core. The outflow pipe
is mounted sideways. The pump is submersed in 25 liters of water in the
canister. It pumps water through a hose, the inflow pipe going directly to
the CPU core (‘contact watercooling’) and via the outflow pipe
through another hose back into the canister.
To stably mount the pipes to the plastic cap, I put some wood on them
which would then stick with a large surface onto the cap. The piece of
wood on the outflow pipe had to be cut in a round shape so it would
fit on the cylindrical cap side. The cap surface was roughened where
it was glued together with the wood so it would hold better.
The cap came off the cpu twice, so I put some aluminium foil into a
frying pan, heated it up and pressed the open side of the cap onto the
aluminium foil. This melted the 1 mm wide edge and made it 3 mm wide.
The edge was flattened and roughened with a file.
Since I noticed that the CPU PCB had taken on some water (it had
turned a somewhat lighter green where the water was), I took a piece of
plastic foil from the cover of a fresh cheese package, cut out a
window for the CPU core and glued it on the CPU.
To protect the core from the epoxy glue, I took a piece of thin plastic foil and stuck it
onto the core with distilled water. The CPU is now waterproof. I also roughed the
area of the cheese foil to which the cap would be glued.
I then wondered if putting a copper plate or heat sink on the core would
increase or decrease thermal flow. Water is a rather bad heat
conductor so some copper might spread the heat to a bigger area and
allow the water to pick it up better. This would also decrease the
risk of steam bubbles e.g. if the pump fails.
The Radiator Design
I tried to make a radiator, but it leaked here and there so I gave it
up. But maybe someone who can build better than me can implement this design:
It is a 0.5 mm copper sheet glued to the left outside of my big tower
case. The edges of the sheet are bent 180 degrees and keep the sheet
some millimeters away from the case. There are also some zig-zag bent copper
wires across it to maintain the distance from the case.
Along the top
and bottom edge of the case there are 2 rows of holes, 6 mm wide in
the middle and getting gradually up to 11 mm at the sides. At the
inside of the case there are 2 aluminium U-profiles glued along the 2
rows of holes. In the middle of each U-profile there is a hole with a
copper pipe glued on with wood for stability. The ends of the
U-profiles are closed with acrylic windows cut from a CD cover.
The idea was to attach hoses to the copper pipes and let the water
flow through the U-profile, through the holes in the big tower case metal
sheet, between the copper and the case sheet, and then through the other line
of holes into the second U-profile. The water would dissipate heat
through the copper sheet. This would mean a radiator requiring almost no space.
The U-profiles were initially glued on with silicone but that didn’t
hold when the case sheet was bent. The U-profiles remained rigid.
So I glued blocks of wood with epoxy at the ends of the U-profile in
the corner of the U-profile and the case sheet. A
block came off the case several times. Once, when I had not glued it on well, water
leaked between the wood block and the U-profile so I had to tear it
apart and redo it.
The copper sheet was glued to the case sheet with silicone, too. At
first it appeared fine but then I tested it in the bathtub with the
pump. I tested the water pressure in the outflow hose by holding my
finger lightly on it. That caused too high pressure and I got a
leak between the copper and the case sheet.
I gave up at this point. I thought that thing was too unreliable and I didn’t want to
put my CPU on risk. However, if the design gives you ideas, if you try them something out
that works, please tell me (and us).
Nor was bending the copper sheet very easy, I ended up having to use a bench vice to do it.
A Conventional Radiator?
I started to make a conventional radiator with 3 copper pipes and some
40 lamellas. I cut the copper sheet into a bunch of 145×39 mm
pieces. I managed to get them somewhat flat by pressing them in the
I wanted to drill 3 holes through the bunch of lamellas and then
solder them onto the pipes, but I thought that too troublesome so I decided to
finding a ready radiator from a scrap dealer instead.
The scrap dealer was closed that day, though. I found some big plastic
canisters at a gas station with car wash, so I got the idea to try something else.
I now have a very simple, almost soundless, cheap watercooling system that serves
I calculated that even with no heat dissipation, the water in this big plastic canister
would only go up one degree Celsius an hour. That’s ok, since it will dissipate heat. I guess it will
level at about 5-10 celsius above ambient and take a looong time to
My Celeron 733 has some 20 W IIRC, the waterpump adds another 6
W so to be on the safe side I take a total of 30 W. 1 g water takes 4
J to rise 1 C.
I have found that in practice, the 25 liters cooling water goes up
about 3 C during a typical 12 h working day (seti@home) with my
Celeron 733@733 (ca. 20 W) (can’t OC with my motherboard :().
It then cools back to ambient while I sleep. It’s very simple and I can absolutely
recommend trying this before messing around with a radiator to anyone
who doesn’t have to get rid of, hmm, >50 W and doesn’t run 24h. A
metal canister or big can would dissipate heat better.
Let me know what you think of this.