Detailed Water Cooling How-To (and other tricks)

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

After trying various air cooling combinations and not getting the results I wanted, I started researching water cooling. I was uneasy about putting water in my valued computer, but after doing all the research I felt confident I could pull it off. I didn’t want to spend $125.00+ for a “boxed” unit as I felt I could do better with my own design. This article will be highly detailed with part numbers, prices, and pictures of the process. After a month of research, two weeks of collecting parts and designing, and 4 days of building and testing I’m happy to report success.

Everyone knows that water cools better than air, but I wanted to know just how much better. After searching the web to no avail, I decided to post my question in the Forum. Steve Bage was nice enough to answer my question. He informed me that the heat transfer coefficient of air is 0.02 W/m^2K and the heat transfer coefficient of water is 0.6 W/m^2K where W = watts, m^2 = square meters, and K = degrees Kelvin. From the math you can see that water is 30 times more efficient at transferring heat than air.

I chose a BE COOLING’s Jagged Edge Copper Water Block and BE COOLING’s 6.25″ X 5″ Copper Core Radiator. Copper provides the best heat transfer properties for the money. If you decide to mix metals, such as an aluminum water block and copper core radiator, make sure to use some type of additive in the water to keep everybody happy. If no additive is used, electrolytic corrosion can occur that will destroy parts.

I decided to go all out and lap the block. From BE COOLING it was pretty flat, but I went through 600, 800, 1000, 2000, and 3000 wet dry sand paper. It tended to get smoother when used dry vs. wet. As you can see, it now has a high mirror finish:


To keep the radiator cool I used a Sunon KD1212PMB1-6A fan. This 120mm X 38mm unit produces 108 CFM, (cubic feet per minute) at 3100 RPMs, (revolutions per minute) 6.8 Watts, and 42 dBA (decibels). To cut down on noise when full fan speed isn’t needed, I decided to wire in a rheostat, Radio Shack part # 271-265B to control the fan RPMs. Rheostats are like variable resistors in that they limit current to a DC device.

For a finished look, I used a Radio Shack knob for the rheostat, Radio Shack part # 274-403A. This rheostat is rated at 3 Watts, and the fan is rated at 6.8 Watts. I figure this won’t be a problem as the rheostat wouldn’t see the full 6.8 Watt load unless I turned it down far enough to practically stop the fan. This will never occur. Half way would be 1500 RPMs which would equal around 3 Watts and I’ll likely need more speed than that to keep it cool. I guesstimate 2500 RPMs to be the sweet spot.


To pump the water, I decided on a MAXIJET 1200 submersible pump which I picked up at a local aquarium shop. It has a 295 GPH rating which is plenty. Keep in mind when choosing a pump that pressure decreases as pumping height of the water increases. The MAXIJET came highly recommended for reliability.

It’s a magnetic drive unit, which means the only moving part is the impeller. I didn’t want to have to remember to plug in or turn on the pump each time I used the computer, so I wired it into the power supply using a Radio Shack relay part # 275-218C. Information on how to wire this and the rheostat can be found HERE.


For a reservoir, I used an underground 4″x4″x4″ waterproof electrical box. The pump fit perfectly inside this box and it is 100% watertight. I drilled two ½” holes in the top of the box for the 3/8″ brass barbs for inlet and outlet and one 5/16″ hole for the pump’s power cord. I drilled the holes just slightly smaller than the threaded end of the barb so it would cut threads into the top of the box.

I sealed around the barbs with some black RTV silicone and put a rubber grommet (Radio Shack assorted grommets part number 64-3025A size ¼”) around the power cord. The grommet was a tight fit, so I used rubbing alcohol to help the power cord slide through. Once the alcohol evaporated it created a watertight seal.

I mounted the box to the bottom of the case with four screws (#8x32x1/2″) and nuts and isolated it from the case with rubber washers (# 00 flat sink washers) to deaden any noise from pump vibration. Clear 3/8″ ID vinyl tubing and stainless steel hose clamps were used for all connections. Approximately five feet of vinyl tubing is needed and 8 hose clamps. The electrical box, brass barbs, screws/nuts, rubber washers, vinyl tubing, and hose clamps were all purchased at Home Depot.


Mounting the water block to the processor was the greatest challenge.

I looked at the various systems available and they were either too expensive or didn’t look sturdy enough for my needs, so I decided to make one myself. I measured the block and the holes around the processor of my Abit KT7AR.

I decided to make the hold down 3″ X 2¼ “. I liked the plexiglass or lexan look, so I made yet another trip to Home Depot and looked for ¼” thick plexiglass. They had it in 18″ X 12″ sheets, way more than I required, so I asked if they had any scrap from cutting these sheets. My lucky day – they had a scrap piece and gave it to me free of charge and let me cut it to the size I needed using their machine.

I had enough to make 6 mounts should the need arise. My design is similar to the one Danger Den offers, but I didn’t want to have to take the system apart to remove the mount, so I made a channel in mine to allow it to slip on and off around the barbs.

I drilled four 3/16″ holes in the plexiglass for mounting, and one 7/32″ hole for the ¼”x20x2″ nylon bolt that would apply pressure on the block. Use sharp bits for drilling plexiglass to avoid chipping. I then cut ¼”x20 threads in the center hole. I made two mounts, one was a template to make sure everything lined up and to ensure I got the pressure right above the processor core. The ¼x20x2″ nylon bolt can be found at Home Depot.


On the finished product, I sanded the edges of the plexiglass with 1000 grit wet dry sandpaper (used dry) to make them clean and smooth.


I used nylon thread stock from Mc Masters (#10×24) to mount the hold down to the four holes around the processor in the motherboard. I cut four 3″ lengths out of the 24″ thread stock and put #10×24 nuts on one end with a dab of glue to make sure they didn’t come loose, then turned them until 4 threads were showing. I put these through the four motherboard holes then put #10 nylon washers and #10×24 nuts on the front to hold the thread stock in place.

I then threaded four #10×24 nuts down to act as a support, making sure to measure all four to get them even. Then the plexiglass is mounted and four more #10 nylon nuts are threaded down to provide support on the top. The #10×24 nylon nuts and washers can be found at Home Depot.



J. Orrand

Now it was time to start modifying the case.

I decided to do numerous mods I have read about at the same time, since the system was already taken apart. I started by cutting a 92mm hole (3″ hole saw works fine) in the side of the case above the PCI slots and in close proximity to the CPU. This 92mm Sunon part number KD1209PTB2-6 moves 48.5 CFM @ 2600 RPM and 33dBA using 2.4 Watts. It is used to exhaust hot air out of the case. A fan guard is used to keep fingers out of harms way.


I then made my own rounded cables for all IDE and SCSI devices. This is pretty simple, but TAKE YOUR TIME. I used a Xacto knife to make a small slit between indentations in the wires, then gently pulled by hand to separate the wires. On ATA 66/100 cables I made slits about every 8th wire. On SCSI and regular IDE cables I made slits every 4th wire. I then wrapped the bundled wires in black wire wrap. I figured every bit of increased airflow would help.


Next I decided to cut the grill off of the power supply fan and replace it with some less restrictive gutter guard material (Ace Hardware). Dremel tools make this an easy task.


I then cut the hole in the front of the case frame for the radiator. I traced the radiator then used a “T Square” to ensure straight lines. Drilling 3/8″ holes at each corner will make cutting accurately easier. I again used a dremel tool with heavy duty cutting wheel for this. Have a couple of replacement wheels on hand.

One thing I kept in mind was how the reservoir would block airflow. I compensated for this by moving the fan/radiator mounting off center so maximum air flow around the reservoir would be achieved.


Next, I placed the reservoir in the case and used the motherboard as a guide to make sure nothing was blocked to decide where the reservoir would be mounted. Once satisfied with it’s location, I marked the box mounting tabs and drilled the four 3/16″ mounting holes.


Then I positioned the relay pump control relay and made sure all cables would reach. Once I was sure everything was in order, I marked the sides of the relay on the bottom of the case and drilled two 3/16″ mounting holes. This would allow the relay to be securely mounted to the bottom of the case with a zip tie.

This is a good position as it allows for easy access to the relay in the event maintenance is required. At this point I also drilled a 1/2″ hole through the back of the case and put a Radio Shack 3/8″ grommet in the hole to run the power cord for the pump to the outlet. Some people choose to use a vacant slot in the computer case, but I wanted this to look professional.


I then decided on a position for the rheostat that controls the radiator’s cooling fan. Some people mount them in spare knockouts in the rear of the case, but I wanted mine more accessible so I decided on the front. I marked a position and drilled a 3/8″ hole to mount it, put the knob on, and made a label.


Finally, I traced a 120mm fan guard on the front bezel of the computer case after lining it up with the cut out for the radiator. I then used a 4″ hole saw to cut through the bezel and mounted a 120mm filter and fan guard to the front of the case for a clean appearance as well as cut down on dust.


Once all holes were cut and or drilled, I thoroughly vacuumed the case out with a shopvac and then blew it out with an air compressor. Make sure to clean the case out very well at this point. A small shard of metal can wreak havoc on computer components

With everything mounted and put together, I filled the system with tap water to flush it out and do a preliminary leak test. After about 2 hours, I was satisfied and went on to stage two testing.

I pumped out all the tap water, then flushed the entire system with ½ gallon of distilled water, then filled the system with 3 cups of distilled water (Kroger Grocery Store). Distilled water was used due to it not being quite as conductive as tap water. I don’t know this for a fact but have read it multiple places. At 89 cents a gallon it was cheap insurance – let’s hope I never have to test the conductivity of distilled water.

I then let the system run for 48 hours to test for leaks. I also played with the rheostat controlled fan at different settings to ensure it was working properly and wasn’t generating too much heat. The rheostat gets warm to the touch but nothing major.

You’ll notice a “loop” in the vinyl tubing coming out of the reservoir going to the radiator. I did this to slow the water down entering the radiator hoping to give the radiator more time to cool the water. It worked as I realized a 2 C drop in temps vs. having the tube go straight to the radiator. If you have a slow pump (under 200 GPH) then this won’t be necessary.


No leaks after 48 hours so I felt I was relatively safe. I mounted everything into the case and put the system back together. I checked to make sure the water block mount wasn’t going to damage the core – my rule is to check everything three times to make sure which I did throughout this project.

I put a very thin layer of Arctic Silver II thermal grease on the core using a thin piece of plastic, making sure it was very thin and perfectly smooth by dragging the plastic over the core in both directions. I then mounted up the block and tightened up my mount. I turned it until it was very snug. I then loosened it up and checked the block to make sure it was seating correctly. A perfectly square print of Arctic Silver II was on the block letting me know it was mounting correctly.

I remounted it, said a prayer, and fired up my system. I quickly went into the BIOS to check the temp – steady at 28 C. Room temp was 26 C. I rebooted and let it go into Windows. MBM5 reported a temp of 29 C after booting. This was a Blue Die Duron 650 @ 650 1.6V.

So I rebooted, went into the BIOS and set it to 6×133 = 800Mhz at 1.625V. Rebooted and MBM5 reported a temp of 31 C. I tinkered around on the net, and wrote some e-mail watching the temp hover back and forth between 30 C and 31 C. I turned the fan back half way to lessen the noise and monitor temps. Temps went up to 33 C. I let it run this way over night and idle temp was 30 C. Room temp was 25.6 C and ambient temp 1″ from the processor was 28.9 C.

Room and ambient temps were measured with a Radio Shack digital indoor/outdoor thermometer – not the best testing equipment, but good enough. Processor temps were monitored via MBM5 referencing the in socket thermistor. The in socket thermistor was bent up slightly to contact the bottom of the CPU. MBM5 temps mirrored temps in Via Hardware Manager.

CONTINUED – The Final Product and Performance Results…

J. Orrand

Here’s the Final Product:

Final Product

I let the system run a couple of days to make sure everything was in order.


Then I decided to run some Prime 95 tests in “Torture Test” mode; with priority set at 10 after 1 hour and stable temp, I had the following results:

  • Duron 650 Blue Die @ 800 1.625 V
  • Room Temp = 25.8 C
  • Ambient = 29.9 C
  • Processor Temp @ Idle = 30 C
  • Processor Temp under Prime 95 = 35 C

This equates to a C/W of 0.15

Note: Prime 95 was ran at priority 10 torture test until stable temp was reached. To determine C/W use Radiate available HERE to determine wattage of the processor. Then use the following formulae:

Delta = CPU temp – ambient temp

C/W = Delta / CPU Watts (determined from Radiate)

More on C/W is available at

When I was air cooled with a Globalwin FEP32 shown below, I had the following results:

Global Win FEP32
  • Duron 650 Blue Die @ 800 1.625 V
  • Room Temp = 23.9 C
  • Ambient = 29.9 C
  • Processor Temp @ Idle = 29 C
  • Processor Temp under Prime 95 = 44 C

This equates to a C/W of 0.32

*note: Prime 95 was ran at priority 10 torture test until stable temp was reached

The numbers show a 53% increase in performance with this water cooled setup over the previous Globalwin FEP32 cooling solution. Granted I made some small case modifications, but I firmly believe the water is the main performance enhancing addition. I’m very pleased with these results. Keep in mind this is on a Duron core, which is more challenging to cool than a T-Bird due to the die size.

After comparing the results using the same system settings as air cooled, I wanted to see how far I could push this Duron 650. My final stable speed was:

  • 952Mhz (136X7) @ 1.85V
  • Room Temp = 25.4 C
  • Ambient = 29.6 C
  • Processor Temp @ Idle = 30 C
  • Processor Temp under Prime 95 = 40 C

Windows and Sisoft Sandra 2001se would run at 966 (138X7) @ 1.85V but Prime 95 reported errors during testing. I don’t consider an overclock stable unless Prime 95 completes, so I dropped it back to 952 Mhz. The system would POST at 1000 (133X7.5) but I got the blue screen after windows started.

I feel I’m limited by the voltage as the temp has never gone above 40 C. I have information on how to wire a resistor onto the voltage regulator of the motherboard to push the voltage past the maximum 1.85V, but I’m leery of doing so. I’ve put the information aside in case I get more gutsy down the road.

I’m very happy wit the final product. Before going to this water cooled system ,I couldn’t run above 850Mhz stably. Now it’s rock solid at 952Mhz. If I could raise the voltage to 1.90, I’m confident I could run this chip at 1000 Mhz if not more. For those of you who like percentages, 952 Mhz is a 46% increase over the default of 650Mhz.


Sisoft Sandra 2001se Benchmarks





I frequent the Forum. regularly (nick AMDGuy) or you can e-mail me with any questions about my setup.

J. Orrand


Leave a Reply