From Air to Water – Lessons Learned

Good overview in setting up watercooling — Jeremy Schmit

Like many people, I’ve been less than pleased with the amount of noise coming from my machine. I’ll never claim that it is the worst sounding computer, but it is a hell of a lot louder than my girlfriend’s HP.

I was using an OCZ Gladiator with the basic (non-Delta) fan. It was pretty good for noise/performance, but I thought I could do better. Eventually the fan developed a bit of a vibration, so it was definitely time to do something.

On a spur of the moment decision, I went to CompUSA and picked up one of their standard 92mm fans (56 cfm) and got some galvanized shingles, tin snips and assorted nuts and bolts from Home Depot. When I got home I tested the fan out, and sure enough it was dead quiet. At least I couldn’t hear it over the rest of the fans.

After making a few measurements of my case, heatsink, and new fan, I sketched out the following design on paper:


Red lines indicate folds, and black dots are where I punched holes.

I stacked five sheets of paper together and cut out the pattern. I taped four of these shapes together to see how it would fit together. After making a few modifications to the fifth shape, I traced the pattern four times onto the shingles and got out the tin snips. After cutting everything out, and punching a few holes with a hammer and a nail, I assembled the following adapter:

Adapter Side

Adapter In

The vanes on the inside are designed to straighten out the airflow from the fan. I had read that centrifugal force can prevent spiraling air from reaching the bottom of fan adapters. I don’t know how much of a problem that would be with my square design, but they were easy to add, and made it look cool.

I also wanted to make it as tall as possible so that the taper would be gradual. On the other hand, it needed enough clearance to let air get to the fan when the side of the case was closed. I figured an inch and a half should be enough.

Lesson Learned:

  • An inch and a half is not enough.

I had a huge increase in my core temp when I put the side of the case back on. I don’t remember the actual numbers, but it was definitely a bigger jump than I got with my old fan. With the side of the case off, however, it performed rather nicely. My temps were very similar to what I got with my old fan, and if there was any change in performance, it was washed out by the daily fluctuations I usually get.

Another thing I was a little worried about was the fact that the fan was now five inches away from the socket lugs that were supposed to support it. This could translate into enough torque to do several nasty things. It could pull the heatsink slightly away from the core resulting in higher temps, chip the core, or even snap the lugs off my socket.

To correct for this, I used a shim and I tied the fan to my power supply with a little bit of string. The string probably wasn’t necessary, but it was good for my peace of mind.

As for the noise, I was rather disappointed. The vibration from the old fan was gone and things were a little quieter, but not the huge improvement I hoped for. I figured the noise I was hearing was the large amount of air being forced through a small opening.

This setup kept me happy for a couple of months, but then a friend of mine got the Koolance case and I got jealous. I figured I could do better than a kit on both price and performance, so I started to do a little research on watercooling components.

Lesson Learned:
  • The forums are a tremendous resource. Just run a search for pump, radiator, or any other component, and you will get more info than you ever need to know.

After a little looking around, I felt confident that I could find what I needed. I went to Napa Auto Parts, and told them I was looking for a heater core and it needed to be small and cheap. After they got over the concept that I didn’t care what car it was for (I told them it was for a “closed circuit cooling loop”), they started digging through their book of parts.

Eventually they found a 6×7.5″ heater core and took me into the back to check it out. Anyhow, they knocked about 20% off the price for me and I walked out with it for about $20. I picked up a Rio 1100 (rated at about 300 gph) for $30 from the local aquarium store. After measuring my case, I figured a 6x6x4″ electrical box ($10) would make the perfect reservoir.

Lesson Learned:

  • The measurements used for electrical boxes refer to the internal dimensions. The external dimensions are considerably bigger.

The trickiest piece to acquire was the waterblock. I looked around at the Danger Den, Z4, Gemini Cool, and Koolance blocks, but in the end I figured it would be more fun to make my own. I work on a college campus and I have several friends who do a lot of work in the machine shop. I figured it would be no problem to ask them to grab a block of copper from the scrap pile and carve a channel in it for me.

Lessons Learned:
  1. Copper is too valuable to end up in the scrap pile;
  2. Copper takes too long to machine to ask a friend to do it for you.


Jeremy Schmit

My friends all told me that they would eventually be able to do the work but were too busy for a couple of weeks. At this point, I was already collecting the other components and a couple of weeks was longer than I was willing to wait.

Everybody has their own opinion about waterblock design, so here’s mine: As you might have noticed above, I’m fairly big on spiral designs (the Koolance block was the only non spiral I looked at.) However, I think these blocks are being misused. Obviously the center of the block is going to be somewhat warmer than the edges, and the water is going to be cooler when it enters the block than when it exits.

So, in choosing a direction for the water to flow, there are two choices:

  1. You can have the coldest water hit the hottest part of the block. This will lead to a very rapid heat transfer immediately, but in the outermost channels the water is warm and the block is relatively cooler, so not much heat is transferred.

  2. You can put the cold water in at the coolest part of the block. This will give you a steady heat transfer along the length of the channel, but you won’t get the massive transfer right at the core. Now, there are many situations in nature where there are similar gradient driven exchanges.

Two examples that spring to mind are oxygen exchange in fish gills, and heat transfer in the legs of birds (picture flamingos standing on one leg in cold water.) Every time, nature has chosen the counter current exchange as the most efficient. This topic has sparked a couple of exciting debates in the forums, and I suspect this article will set off another one, so let me cover a few things.

I agree that the objective is to cool the core and not the waterblock, but that should be equivalent to finding the fastest way to transfer heat from the block to the water. Also, a couple of people have tried running the water both directions and have found cooler temps putting the water in at the core. I suspect the reason for this is that commercial waterblocks have too thin of a base plate for the heat to efficiently conduct to the outer channels.

I tried to design my waterblock to take advantage of this highly efficient means of energy exchange by giving it a thick base (~1/4″) and a long, narrow channel to maximize the exchange path. I look forward to hearing people’s comments on this subject.

Now, I needed to make the waterblock. I ordered a 2x2x5/8″ block of copper from ($5+$10 shipping – Ouch!) Then, I went to the Dremel website and ran a search for bits that could cut a groove in copper. I selected one of these bits from Home Depot and rushed home to make my block.

Using the bit in my standard power drill, I made five deep holes in the copper. Soon I started noticing that it was getting harder and harder to generate the big piles of copper dust I was getting at the beginning. A quick inspection of the bit revealed that the cutting edges were becoming quite smooth:


My bit (left) and the bit before I started (from the Dremel web site).

Let’s see – I’d done about 1.5″ worth of channel with a $5 bit…for a channel over 10″ long. Damn, this was going to get expensive quick!

I went back to Home Depot and this time I picked up a standard ¼” titanium drill bit (~$5.) I wanted to drill the holes as close together as possible, so that it wouldn’t be too hard to connect them.

To do this, I drilled each hole on the path to less than an eighth of an inch deep and took them all down together a few millimeters at a time. I paused after every ten minutes of drilling in order to let the metal cool down. It really got hot fast, but holding it under the faucet cooled it back down quickly.


Each hole on the path has just been started. The holes done with the Dremel bit are the oblong holes across the top.


The work in progress. The grey object in the lower right is my thumb in a leather glove. I marked the bit at 3/8″ to let me know when the holes were deep enough.


After all the holes were drilled to 3/8″, I connected them by inserting the drill bit and gradually tilting the drill until the bit was almost horizontal. I was able to do this for each channel, so it is pretty uniformly ¼” wide, although there are some big peaks and troughs on the bottom of the channel.

I’m fairly happy with how it turned out. I think that the roughness will both add to the surface area and increase the turbulence without restricting the flow too much.



Jeremy Schmit

My original budget for the project looked roughly like this:

  • $25 radiator
  • $30 pump
  • $15 copper
  • $10 hose, plexiglass and assorted fittings.
  • $10 reservoir

Well, I was fine with the pump and copper and I even did better than I thought with the radiator, but I didn’t do so well with the other stuff.

Lessons Learned:
  1. Lots of inexpensive items add up quickly;
  2. Hose clamps are much cheaper at Kmart (4/$2) than they are at Home Depot ($0.89 each.)

By the time I got all the miscellaneous items, my bill was over $40 (including the reservoir.) And, no, I didn’t discover the hose clamps at Kmart until after I was done.

I finished the waterblock by using JBweld to attach a sheet of 1/4″ lexan over the top of the copper. Then I drilled holes in the lexan over the inlet and outlet, and used JBweld to attach the barbs. Because both my pump and radiator had 5/8″ fittings, I opted to use 5/8″ for the waterblock as well. After all, a bigger hose can’t hurt anything, right?

This wasn’t entirely true. The hose is kept on spools at the store, and the act of winding the hose on the spool will flatten the larger hoses. I figured that the water pressure would force the hose back to its original shape, but this has happened very slowly.

In the low pressure parts of my system (farthest from the pump), the hose is still very flat and I’m worried that flow might be constricted. One of these days I’ll do a major overhaul on the system and switch to silicone tubing.

Finally I lapped the block, first with 150 grit (to remove the really big dings and scratches), then with 600 grit sandpaper.


The assembled waterblock before I drilled the mounting holes in the lexan.

Now it was time to assemble my system for leak testing. I opted for a pump-waterblock-radiator arrangement, although I realize that this might not be ideal for submersible pump systems.


Everything assembled except the waterblock.

I had to drill three holes in the top of the reservoir: One for the pump outlet, one for the power cord, and one for the water return. Originally I sealed these with silicone glue – this leaked all over the place. I peeled off all the silicone and resealed the holes with JBweld. This fixed the leakage at the outlet and the power cord, but the water return still sprung the occasional leak.

The problem is that the hose it too flexible and will peel away from the adhesive if it is moved. The outlet does not have this problem because the pump outlet actually protrudes slightly through the hole. Currently, the problem is fixed with a ton of silicone glue, but if I were to start over from scratch, I would insert a barb into the hose where it enters the reservoir. This would give the hose enough stiffness for the glue to make a permanent bond.

After letting all the glue dry, I turned the system on for leak testing. When I got back from work I found a large puddle. It appeared that the leak was coming from the JBweld-copper joint. This was surprising, because it was the only joint in the system that I had already double glued. I applied another generous layer of JBweld, and when it was dried I found that it still leaked.

I was getting really frustrated at this point, so I applied a layer of silicone and found that this stopped the leakage. I only tested the system for about 12 hours because I was very anxious to try it out. I hooked up the water system and, after a quick test, I took off to the local bar to congratulate myself for a job well done (it was Friday night, and I was meeting a friend.)

When I got back, the beer inside me made the decision to turn the machine back on and leave it on overnight. I realize this wasn’t too smart, but when I woke up everything was fine, so I left it on when I left the apartment.

When I got back that evening, I turned my monitor on and found that the system had frozen. Figuring that I had pushed the overclock too far, I shut it down and waited a few minutes for things to cool down. Still nothing. After several attempts to start the computer, I noticed a shiny spot on my video card.

Lesson Learned:

  • Video cards don’t work when they are wet.

This was definitely reason for panic. I wouldn’t mind replacing just about any part of my computer, but my video card was a Radeon 8500 and my latest frivolous purchase. It was my most expensive component by a factor of two. I dried out my AGP and PCI slot by wrapping a credit card in paper towel and inserting it into the slot (many, many times.)

After letting things dry overnight, I noticed a white residue on the parts of my video card and motherboard that had gotten wet. I scrubbed this off with a Q-tip and water, and after allowing another couple of hours to dry, I put everything back together. It worked!

Now to fix that damn waterblock! I chipped off as much of the old JBweld as I could, then I attacked the remaining JBweld, copper and lexan with 100 grit sandpaper. Then I cleaned off the surface with rubbing alcohol and applied a generous layer of JBweld. After this dried for 12 hours, I applied two coats of silicone.

After this treatment, my waterblock has not leaked in over a month.

Lesson Learned:

  • When the instructions on the glue say “surface must be clean and dry” they mean microchip-fabrication-plant-clean, not wiped-off-with-a-paper-towel-clean.


Jeremy Schmit

So how does it work?

All I have to go on is my socket thermister, so I can’t give any accurate temps. But it is safe to say that it works much better than my old air cooler. Because I am a member of the folding team, I don’t have any idle temps, so all of the temps here are at 100% utilization after the water reaches equilibrium.

With air cooling, I ran my 1.2 Thunderbird at 1.41 GHz (10*141, 1.80 vcore) and it averaged about 51C.

With my water setup I jumped up to 1.48 GHz (10.5*141, 1.85v) and my temps dropped to about a 41C average.

Recently I upgraded my radiator fan from a single 92mm to two 120mm Deltas. This dropped the temps to 35C and allowed me to bump the FSB to 143 (although with the weather warming up recently, I’ve dropped back to 141 and my temps are 37C.)

I make sure that all my overclocks can run Prime95 for several hours before I leave the computer at that setting. I’m a big fan of the torture test and I’ve started swearing at my computer a lot less since I started using it.

Of course the excuse for the entire project was to cut the noise of the system, so how did that go?

Well, I realized early on that it was going to be a disappointment in that respect. After I purchased all the parts, I took a close look at my case to see how they would fit together. When I got my head down right next to the case, I realized that the vast majority of the noise was coming from the power supply, not the CPU. DOH!

I think the next step is to disconnect the PSU fan and replace it with my 92mm (attached to the outside of the PSU.) The 120mm Deltas were a little too noisy for my taste, so I got a 25 ohm rheostat from Radio Shack ($4, item# 271-265) which allows me to turn them down just to the point where I can’t notice them anymore.

I’d also like to thank the members of the forum for suggesting a Radio Shack relay to turn the pump on ($8, item# 275-218.)

Here’s some pics of my complete system. As I mentioned before, the reservoir is a little too big to fit in my case, and when I try to make it fit, the hose kinks because I cut it too long. Rather than force the issue, I think I’ll just keep an eye out for a full tower that will fix both issues at once.


The final setup. The pink color come from the water wetter I added. The tube sticking straight up from the T-joint is for filling and bleeding the system.


The radiator. The 120mm fans are mounted in a staggered push-pull arrangement. i.e. The other fan is on the other side of the radiator on the lower left corner (as viewed from this direction).

I have to say that I am very satisfied with how it turned out, but I think it is clear that I would do some things differently if I started over. I hope I have spared at least one person from making the same mistakes.

Jeremy Schmit

Be the first to comment

Leave a Reply