Project: Overkill

Geo-cooling at its best – Roger Dugans

Everyone needs a hobby or two, right?

Computers actually fulfill quite a few of my hobby needs, and one of the most enjoyable for me is Water Cooling. This article is about one way of taking water cooling a bit farther than most – perhaps a bit farther than sanity would suggest prudent! 😉

A few years back, I saw a few Overclockers.com Forum posts about relatively extreme methods of water cooling, namely reservoirs and piping buried in the ground to create a “ground cooling” system that would be silent and give excellent temperatures. This concept intrigued me and started me thinking about ways to implement my own ground cooling loop and perhaps use it to cool multiple computers. I have a number of computers since building them is another hobby of mine, and most are not portable.

Numerous methods of creating a ground cooling loop occurred to me, but my favored idea was to make a “radiator” element with 3/4″ or larger copper tube and bury it outside below the frost line, running the tube in through a window for the computers. I began to design my concept and also began to accumulate parts….

I have a major advantage in the area of parts for such a system because I am currently employed as a welder/pipe fitter installing heating systems. Much of what we do is copper pipe, and we often replace old systems with brand new ones. This leaves me with a LOT of scavenging opportunities! In fact, almost all of what I ended up with was scrounged from one scrap pile or another.

The first MAJOR score I had was a heating element used by a company that did electroplating and anodizing; this element leaked on the inlet and outlet lines and had been repaired a few times before it was considered not worth the effort and expense.

The company was quickly talked into giving me the unit, which I was easily able to repair by removing the damaged sections of tubing and TIG welding it back together (The damage was on the inlet/outlet lines which used to be about 2 feet longer.)

The element is made of 1″ stainless steel tube and approximately 3 x 5 feet in size- about 120 linear feet of 1″ tube. This is actually very similar to one of my concepts for a home-made radiator element, but much cheaper and stainless better suited to being buried than copper.

The second major score was the pump. This unit was an unused spare at a place we were installing an all new system in. The pump was incompatible with the new ones and so was to be removed. I removed it straight to my house. This unit is a Taco 0011 bronze circulation pump, rated to nearly 1800 gallons per hour max, with a max head of somewhere around 30 feet. Pretty strong pump, and it has an expected life measured in years under constant duty at pressures of 80 psi or so.

The rest of the parts I used were generally also scrounged as scrap or parts from jobs where we bought more parts than we used. Note that all the scrounging did take quite a while – I have no problem grabbing extra, unused stuff, but ethics prevent me from intentionally ordering extras. It has taken me about a year from when I first decided to do this to actual completion.
{mospagebreak}

I prefer to test my wackier ideas before I get into the final design phase due to the large amount of guesswork and trial-and-error I employ, so the first step I had to accomplish was to actually see if this element would cool a good sized heat load decently.

The first test was to simply connect one of my Eheim 1250 pumps to the radiator to cool my test bench computer. The radiator was simply leaning against the desk, where it would remain for about six months!
The radiator element passed with flying colors and I prepared to move on to Stage 2: pump and radiator testing.

Using some of my scrounged parts I soldered up a test system, as shown below:

The picture is not great, but it shows the second series of tests: radiator and pump with 1″ copper lines. Temperatures were only slightly better than with the Eheim 1250 – my theory is that the increased effectiveness of the much higher flow rate is greatly diminished by the extra heat added by the pump. The results were more than sufficient to prompt me to move on to the next step: more computers for a greater heat load and added flow restriction.

Connecting more computers left me with a question that I personally had not dealt with before: series or parallel cooling loops.

I have heard of people preferring one or the other in their systems when cooling duallies or adding VGA and chipset blocks, but I had done almost no testing of my own, and had no info at all on how to best split up the flow on the monster pump I had. So I made a manifold that would allow me to test both.

By opening and closing three valves I was able to switch between series and parallel flow paths at will.

I ran with both configurations for quite a while and ended up finding that it makes very little difference – at least in this case. Running parallel loops got me slightly better temps on one block, and running in series made the other block do slightly better. Either way, both blocks worked quite well and the system was easily able to handle the heat load.

In fact the temperatures increased only slightly when running two CPU blocks in the loop.{mospagebreak}

The proof of concept work was almost done: all that was left was to increase the heat load further. The small snag I hit now was that of the four water cooled machines I had running, two were LAN rigs and needed to remain easily portable. So I began messing with some ideas for quick and easy copper cap water blocks for video cards and chipsets. I ended up running the two computers with CPU, VGA and chipset blocks and found that temps still barely changed, even with the added flow restrictions and heat load.

My guess is that the pump’s flow rate and pressure are enough to keep all of the blocks flowing enough liquid to be effective for cooling, and that a combination of the large surface area of the radiator and the large volume of water – around 10 or 11 gallons – give plenty of reserve cooling capacity. My guesstimate of the maximum heat load I have applied so far is 350 or so watts (2 AMD Athlon XPs overvolted and overclocked, 2 NVidia GF4 Ti4200s overclocked, and 2 motherboard chipsets running revolter’s.)

I called the concept pretty thoroughly proven and continued to gather parts for the final phase of construction: burying the radiator element 6 feet underground, about 10 feet away from my house (had to get clear of the shrubs and flower gardens!)

I finished up all this testing in the middle of winter, which in New England is just NOT the time for excavating with a shovel, so there was more of a delay. This delay was lengthened by the need for me to have surgery performed on one of my feet with a two month recovery time. The delay was compounded still further by rain following my recovery. Fortunately for me, Massachusetts was blessed (or cursed, depending on your outlook!) with exactly ONE cool weekend after I was off of the disabled list.

I dug the hole right where it was needed and found, as always, evidence of the last ice age: a pile of rocks to make digging harder. To my dismay, I hit a very large rock about three feet down; the best I could determine was that this rock was at least two feet across and about eighteen inches wide. That much was uncovered by my hole and there was still more of the rock to be unearthed.

Although three feet is NOT under the frost line and this depth will be affected by ambient air temps, I decided that enough is enough and three feet would just have to do. I dug the slightly pitched trench for the connection piping back through the garden and bushes and soldered up the require piping to get everything above ground hidden by the bushes.

After a weekend of digging and filling, I was done with the project for the week; recovery from operating an armstrong-powered shovel was needed.

(A funny side note about the whole ditch-digging aspect of this project: The reactions of my neighbors was very interesting when they asked what I was doing and I replied “Working on my computer.” Some (who should have already known!) looked at me like I was crazy and others wanted to learn more about the whole concept of water cooling computers.)
{mospagebreak}

I de-soldered many of the fittings used for the test system during the week and scrounged around for the rest of what I needed but still fell short – I had to break loose my wallet and buy the last few pieces. My usual luck held though, and I was given a record setting high temperature Saturday to solder the final assembly.

The radiator element and underground piping is all pitched to aid in bleeding air out of the system, and the above ground pipes are also pitched, although to a lesser degree. There are valves just above ground level to isolate the radiator if needed, as well as a 1″ filling valve and a 1/2″ valve for bleeding purposes.

Inside the cellar looking out you can see the piping inside and out. There are a few items of interest located inside the room: two pressure gauges (only one visible in the first picture), a coolant thermometer and an automatic vent. The vent is at the highest point in the system and makes it very easy to remove any remaining air from the system over time.

The next picture shows the entire system as it is visible inside the house.

*The pressure gauges I hope to have soon are of the smaller size.
*The two valves that are not currently connected are my provision for future systems: with little difficulty I can add more piping to connect one or more computers to the system. The only limiting factors are the stock I can scrounge and whenever I find the limits of the system’s cooling capacity. 🙂

The gauges that are currently mounted do NOT actually work; I have my eyes on a matching pair at work that I think are available but I have not yet gotten them. I did have one working gauge on the test system and pressure was generally in the 10 to 13 psi range while running. I hope to get a pair of small valves as well as the gauges so that parts can be swapped out in the future without too much hassle.

But what are the TEMPS?!?!

OK – some temperature data for those who are interested in RESULTS!

You won’t be in awe of my temps – idle around 36C and load around 39C for the CPU in my single connected system as pictured below. These temps are with a daytime high around 85 F. But the system is very nearly silent with some low-pitched noise from the pump, which will soon be enclosed in a vented case, and very low noise from the power supply fan and the single Mechatronics 120 mm fan I have blowing over the video card RAM, which is still air-cooled……for now.

This compares with my previous, more traditional setup of dual 120 mm fan heater core and Eheim 1250 cooling the CPU to a load temp of 38 to 40C. Not much temperature difference but considerably quieter than even THAT quiet system: I hear the keys on my keyboard “click” now;
I can literally drop a piece of paper on my desk and hear it land.

The Computer:

Nothing special by today’s standards:

• Abit NF-7s
• AMD Athlon XP 2200 or something @ 2270
• 1024 MB DDR (2 x 512 MB Dual Channel)
• NVidia GF4 Ti4200 128 MB
• Creative Labs sound card (emu10k1 Linux driver)
• Western Digital 8 GB and 120 GB IDE HDDs
• DVD reader/CD-RW driver
• Cascade CPU water blocks
• Home made copper cap video and chipset water blocks
• 10′ Clearflex 1/2″ ID x 1/8″ wall tubing
• Temporary case while I redo my tower
• Gentoo Linux 2005.0

Financial Nitty Gritty:

This is always a subject that is of great importance to me – I am a cheap bastard. 🙂

The whole project has been time and effort intensive, but it has cost me very little in terms of cash: about $135 for various fittings that I couldn’t wait for, coolant and additives, paint and grass seed to help fix my lawn.

The approximate cost of the items used:

• Custom stainless steel heating element: $2000
• Taco 0011-BF4 pump: $350
• 50 feet @ 1″ copper pipe: $175
• 6 @ 1″ brass valves: $120 ($20 each)
• Vent: $15
• 1/2″ brass valve: $8
• 3 @ 1″ sweat/IPS unions: $27 ($9 each)
• 2 @ 1″ copper Tees: $7 ($3.50 each)
• 3 @ 1″ x 1/2″ copper Tees: $10.50 ($3.50 each)
• 9 @ 1″ copper Elbows: $27 ($3 each)
• 1″ copper x IPS Tee: $4.50
• 1″ x 1/8″ brass Drip Tee: $6
• Bronze Pump flanges: $20
• Assorted brass nipples: $20
• 2 @ 1/2″ brass copper x IPS elbow: $7 (3.50 each)
• 2 @ Gauges: $40 ($20 each)
• Thermometer: $20
• Paint: $6
• Solder: $10
• Threaded rod, copper clamps and plates: $15
• Distilled water: $10
• Antifreeze: $18
• Water Wetter: $8

For a total around $2900 US Dollars for parts alone.

Also include the insane amount of work for just the FINAL system:

• 1 day digging hole
• 1 day soldering pipe and radiator in the ground, then filling hole
• 1 day soldering and filling system with coolant

I list the approximate cost of all this stuff for a few reasons:

• To prepare anyone who might think of trying something like this for the costs
• Assert the value of Scrounging
• Brag about my scrounging ability, of course

To sum the whole thing up:

• The results are marginal at best for cooling a single computer: an equally low-noise system can be built for much less money and with much less difficulty
• For cooling multiple computers with virtually no noise, the test system leads me to believe it will be very effective and should be able to support a heat load of 600 watts or more while maintaining good temps
• It has been an enjoyable project so far even though parts of it have been fairly backbreaking work
• Yes, I admit that my mind may be missing

Roger Dugans