Watercooling with an aluminum waterblock

September 27, 2009 | categories: engineering, projects | View Comments

Last night, I finished assembling the waterblock for the new server in my basement. Since I'm new to this watercooling stuff, I decided that a design where the fluid path through the server has no potentially leaky fittings along it would be a good starting place. I made the waterblock from two chunks of 6061 aluminum, with a loop of copper tube sandwiched in the middle. I milled a semi-cylindrical groove in each plate of aluminum with a ball-end mill, and added relief on one side to allow space for a 180 bend in the tube. The plates are clamped together with 1/4-20 socket head cap screws. As I tightened the bolts, I could see the gaps around the tubes disappear, so thermal contact between the tube and the plates is at least decent. Unfortunately, the upper plate doesn't make good contact with the lower one. By design, there's a nominal 0.010 in gap between the two plates before the tubes are compressed. I tried adding shim stock between the two, but I couldn't easily balance the contact force between the tubes and the plates, so I left out the shims.

Half the waterblock

Half the waterblock

Disassembled view

The waterblock disassembled

I fired up the server, and it looks like watercooling is a lot more effective than I expected. The blue line is for the CPU that was air-cooled, and the purple line is for the watercooler. The blue line dives after 10 minutes or so because that's when I realized one of the fans aimed at CPU1 had stopped. If the fans all worked, I bet the watercooler would still win on absolute temperature, but the fans might win on efficiency.

watercool_chart_2009-09-26

The circulation pump, a Laing SM-909-NT-14 designed for hot tubs, I think, is rated for either 15 or 65 W, depending on which part of the label you believe. (I'll have to measure it to find out for sure.) It maxes out at a volumetric flow rate of 8 L/min, but with the small diameter copper tube attached, it drops to 2.4 L/min (40 cm3/s). The ID of the tube is 0.48 cm, which puts the mean velocity at 2.25 m/s. The Reynolds number for this flow is around 1000, which is below the transition to turbulent flow, which occurs around 2300. Looking at the jet coming out the end of the tube, it looks laminar.

Overview of the full setup

The watercooling mad scientist lair

Aluminum waterblock with copper piping in place

You'll notice that the tube is wearing a beautiful brass fitting. This is a fitting that I clamped on during the bending process to keep the tube from sliding into the pipe bender and crimping, rather than bending in a nice radius. Unfortunately, I failed to realize that the section of tube with the fitting would be trapped between bends.

Close up of waterblock

So, I call this a success. The next iteration of the waterblock will be made from copper, and I'm hoping to make two of them so I can actually run the server. I thought about trying to press tubes through reamed holes in a block, maybe with the help of a heater to expand the block, but I think I'll try a two-piece soldered design instead. My plan is to create a vaned cavity that guides a sheet of fluid across the processor, but I haven't figured out the details yet. Onward!

Read and Post Comments

Building a green server

September 13, 2009 | categories: energy, projects | View Comments

Ever since a brief stint as a network administrator in 2005, I've admired the discipline of data backup with undeserved zeal. A little more than three years ago, I decided that I should probably build my own server as a data repository; it could also serve this blog, store my email, and so forth. At the behest of my friend Mike, who once claimed that the massive data breach that Google will eventually suffer will be worse than September 11th, I also kicked my Gmail habit. (Mike has not kept up his end of the bargain, and still uses Yahoo Messenger. Also, he owes me a comic book about penguins.)

Last spring, as we were getting ready to move to Somerville, I had accumulated most of a server, but I didn't really have a good place to put it. Now that we've moved, I have an insulated attic, a moderately dry basement, and most importantly a killer garage. If the house burns down, the data is safe in the garage!

While I am definitely a sucker for an attractive rackmount server with hot-swap drives, I'm even more of a sucker for energy efficiency. I did some research and built what was, at the time, the greenest server I could build with consumer-level components. With the recent release of solid state hard drives, I don't think I can claim to be setting any records, but I think I'm doing pretty well in terms of required cooling power per processor cycle.

The rig:

  • Dual Xeon L5410 quad core processors
  • 1U rackmount case, RM1002T
  • Sparkle 80plus power supply, FSP400-601UG
  • Western Digital low power drive, WD5000AACS
  • Asus DSBV-DX motherboard
  • Two Corsair 1 GB DDR2 667 MHz FB-DIMM
  • Quiet, low power fans (Antec 40mm Ball Bearing Case Fan from ANTOnline through Amazon.com)

Total power consumption started at 131 W with the processors idle, but I was able to reduce it to 102 W by replacing the stock fans with the quieter Antec fans. I haven't actually measured the peak power consumption yet judging by the processor specs, I expect it to be at least 200 W, but probably not more than 250 W.

I tried the server in the poorly-insulated garage this summer, and it overheated pretty quickly. The revised plan is to try the aforementioned moderately dry basement, which affords the additional interesting opportunity of building a geothermal cooling system for it. (Yes, it's a ridiculous idea-- ridiculous and awesome.)

I don't know if I'll ever get around to that, but I'm starting with a water cooling system, which I think is a necessary precursor to a geothermal system. If the water cooling works, I might turn it into a water pre-heater for our domestic hot water heater. I previously calculated our hot water load at around 2 kWh per day, which is an average power of 83 W, less than the average power of the server. We'd certainly lose a lot of the heat along the way, but if I'm going to dump waste heat somewhere, it might as well be into water I want to heat anyway.

(Before you go rushing off to Treehugger.com, note that this "heat your water with your PC" plan is not an economically sound proposition-- heat from electricity is about 4 times more expensive than heat from natural gas.)

Read and Post Comments

Building a 150 foot zipline

August 23, 2009 | categories: engineering, projects | View Comments

My brother and his wife bought a house in Michigan a few months ago, and Sharon and I went out to visit them last weekend. The house is on a wooded lot a few miles outside of town. There are other houses that you can make out through the woods, but nobody really nearby. They probably own an acre or two of woods, and they abut a 15 acre park (which previously housed a privately-owned 1/3 scale steam railroad-- I'll save that for a later post).

At some point, we were standing around in the yard when Ben pointed out an overgrown field about 30 yards through the woods. Ben had previously discovered steel posts that suggest the field was in fact a grass tennis court that hadn't been maintained for at least a decade. As soon as I saw the outline of the tennis court, I thought, "These weeds will be subdued by this man. I will burn fossil fuels in an age-old ritual: the subjugation of the wild to man's will."

A few days later, I dragged Ben's 6.5 horsepower beast of a lawnmower out of the shed and fired it up. 30 minutes later, nothing taller than three inches lived within the perimeter of the court.

My brother's newly discovered grass tennis court, neglected for decades, after I mowed it The tennis court, recovered

Once the tennis court was recovered, the landscape was changed. Instead of a house in the woods, we had in our possession a house overlooking a small athletic field. There was talk of a croquet pitch or maybe a small soccer or frisbee game. I mowed the rest of the field, and we thought about the possibilities.

Later in the day, we talked about maybe putting a zipline in the yard on the other side of the house, but it didn't really seem like the right spot. The lawn was something of a valley; I didn't see an easy way to position the zipline so the rider wouldn't crash into the tree at the bottom. As it was getting dark, I walked around the yard and considered the possibilities. Coming around the corner to the tennis court, I noticed that the court was a good ten feet lower than the house, and there was a nice tree right at the edge of the yard. Ben pointed out that there was a massive tree on the far side of the tennis court. I paced the distance off at around 150 feet.

As I was lying in bed that night, I did some calculations in my head. The potential energy of a person at the top of the hill would be converted into kinetic energy at the bottom of the hill. If we ignore wind resistance and friction in the pulleys, we can estimate an upper limit of the speed.

<img src="http://chart.apis.google.com/chart?cht=tx&chl=mgh = frac{1}{2}mv^2">

The m cancels out, so either we all die, or nobody does. When you reach the bottom, the maximum possible speed is

<img src="http://chart.apis.google.com/chart?cht=tx&chl=v = sqrt{2gh}">

h is around 10 feet, and g is 32 ft/s^2. This means that roughly

<img src="http://chart.apis.google.com/chart?cht=tx&chl=v = sqrt{640}approxsqrt{625} = 25 frac{ft}{s}">

I went to sleep thinking that a zipline where you landed at a 25 ft/s sprint was about as dangerous as I wanted.

The zipline site, looking back up the hill The zipline site, looking up the hill

The next day, we went to Home Depot and picked up 200 feet of 7x19 1/4" steel cable, a 24" piece of schedule 40, 1" PVC pipe, and some cable clamps. A little internet research suggested that REI might have some cheap pulleys that could bear the load. They didn't have exactly the pulleys we were hoping for, but we found some that were good enough. We also got two carabiners. The advertised breaking strength of everything involved was at least 2000 lbs, more than 10x the weight of the heaviest rider. (I was expecting the cable to be slightly slack, so I wasn't worried about the 1/sin force amplifying effect if the cable were taut.)

All the supplies needed for a badass zipline The supplies

In the afternoon, we laid out the cable. We didn't have any good way of tightening the cable other than Ben pulling on it as hard as he could while I clamped down the fittings real quick-like, but the terrain was such that a slightly slack line worked great. You skim about 2 feet off the ground most of the way, before running to a stop as you intersect the flat tennis court. (The videos below suggest an alternative "crash-and-burn" method of landing, but we were going for distance then. It's relatively easy to land on your feet if you do it when the cable is at shoulder height. Cable at thigh height-- not so much.)

Zipline trolley closeup The trolley

The pipe-trolley forcing alignment with the cable worked well; I think a trapeze mounted at a single point would make it harder to launch and land safely. After a few runs, I noticed that the axles in the pulleys were too hot to touch, due to the friction from the nylon sheaves. That was mitigated with a few squirts of Boeshield T-9, everyone's favorite waxy lubricant. In the future, if the zipline proves popular, we might raise the start a foot or two and switch to a Petzl "Tandem Speed" pulley, which has ball bearings instead of a nylon sheave riding on a fixed axle. There's also talk of adding a slip-n-slide in the landing zone.

I knew that the zipline was a success when my brother's wife went out and rode it three times in a row by herself the next morning.

Videos and more photos below.

Sarah rides the zipline

Ben, method air on the zipline

Read and Post Comments