Winter 2009

Every year when Cross Country ends, I find myself in a strange position. All of a sudden, six straight months of intense training and competition are over and I'm left with an abundance of free time. It usually takes a week or so to get out of the "GO GO GO!" mood that creeps in during a long season, but once that's taken care of I catch the bug like nothing else. I first started building computers around this time Sophomore year, and last year I took on a 250-hour science fair project (see 'The Liquid Nitrogen Experiment'). This year, I'm being even more ambitious. Here are a few likely projects for this winter:

I4 Go-Kart:

Normally I'm not a four cylinder kind of guy, but when a friend suggested that we build a go-kart, my first thought was how great a high-revving, mid-mounted four cylinder could be on a modified kart. It'll take loads of work, but the end result should be something to see. I want to join the Formula SAE team at whichever college I attend next year, so this project would give me some invaluable experience. It's also Mario approved.

Modified A/C Water Chiller:

I've been thinking about this one for a long time. The plan is to buy a window air conditioning unit and modify it to run as an auto-cascade. Auto-cascades take all the benefits of multi-stage cooling systems and leave their problems behind; with the one exception being complexity. Though this makes it the perfect long-term project, it also makes it somewhat unlikely. The cost of all of the tools and materials would run upwards of $700, and the end product would be too expensive to run continuously. I've been researching alternate means of reaching cryogenic temperatures via sound waves, magnets and lasers, but though these options may be more effective in the long term, they are all but impossible to create with my limited resources and know-how.

  

New Computer Builds:

Perhaps the most likely project for this winter is a new computer case. I'll be doing a full re-build for college this summer and I want a new case to go along with it. I've been working on a few different designs, the best of which are posted below. The 3-tiered case is 6.7" x 6.7" excluding the curves, and the larger case is 11" x 12". They will both be made out of carbon fiber. Here's the kicker: the smaller computer has the potential to be just as powerful as the larger one, minus the liquid cooling.

 

 

Project: Alpha Dog

Meet my first real graphics card. His name is Alpha Dog (given by the manufacturer, not me). He's a 'superclocked' version of XFX's 8800GT. He was designed to be a thoroughbred that overclocked like mad; some users reported gains of up to 200mhz over the stock 670mhz. Unfortunately, mine only scaled up to 715mhz without any extra voltage. I let it go for a while, luck of the draw and everything, but I eventually decided to take matters into my own hands.

Meet Alpha Dog's PCB. It's pretty nice as far as PCBs go. Most of it's unimportant to us right now, though. We're interested in the area in the red circle. Therein lies the chip that regulates the GPU core voltage. The idea is to solder a 1kΩ variable resistor onto the feedback pin of this chip and a ground. Decreasing the resistance increases the core voltage, allowing a greater overclock. (NOTE: I did not come up with this mod)

 Here's a closeup of the same area on another card, courtesy of Dinos22 of Xtremesystems.com.

The sixth pin from the left is the one we want. It's official name is VSENSE. The red line follows the trace from this pin. Since all of the solder joints on this trace are effectively the same point, I went for the last one because it seemed to be the easiest. Now, the mod calls for a 1kohm variable resistor, so I picked one up at Radioshack and gave it a go one boring day during the winter of Sophomore year. Young and inexperienced as I was, I rushed the job and ended up taking off one of the SMRs surrounding the solder point. Whoops.

I posted a few pictures on Xtremesystems' forums and the members there proposed an elaborate fix. I was to desolder the remaining 3 SMRs in the chain (red X's) and solder on two 16ohm linear resistors in their place. (Thanks to Largon and Jason4207)

I bought the two resistors and spent a good 3 or 4 hours attempting the fix, but it proved to be too much for me. I sold the card to an Xtremesystems member for parts and he ended up fixing it and using it in his own computer. Much respect to him for pulling it off. As for me, I'm just glad my Alpha Dog is still alive and kickin'.

Robotics and Beyond

As part of my intership with Robotics and Beyond Inc., I was tasked with the creation of a Computer Science curriculum to teach to interested students this past summer. The idea was to create a computer that would be easy enough to take apart and put back together that, after some quick instruction from me, the students could successfully assemble it by themselves. I knew from experience that a store-bought case wouldn't fit the bill, so I decided to build one myself. I decided on an open case design so the students could see how the individual components work while the computer is turned on (plus I just think it looks better).

 I wanted the students to have something to relate the hardware to. They knew that 'this' was a graphics card and 'that' was a CPU, but they didn't know what each did. I wanted to show them what programs stressed each component individually, and that you have to keep the intended use of the computer in mind when selecting its components. I offered to teach the basics of Solidworks (a CAD program) to illustrate this point. My supervisors liked the idea and asked me to build another computer for a student to use Solidworks on. The catch: the total cost of the parts could not exceed $350. The cost would certainly limit things, but I still wanted to make a statement to the students. I wanted to show them that a small, cheap computer designed with efficiency in mind could be just as powerful as a desktop. I built the mini-ITX PC below for $312 and tweaked it until it was as fast as the open machine above.

In Solidworks, I showed the students how to build complex 3D arrangements by combining singular parts in an assembly file. I decided the iconic Lego brick was the perfect tool for this, so I walked each student through the making of an individual brick and showed them the mechanism by which Solidworks locks two 'part' files together. I then let the students experiment, encouraging them to create models of the robots they had made in the other projects.

 

The Progression of Design

It all started with 'The Box'. Everyone knows The Box. Every family has one. The Box represents the boring, monolithic tower that everyone believes the computer has to be. I inherited The Box four years ago when my brother left for college and ever since I've been searching for ways to put an artistic spin on the traditional computer.

The first step was a case. I was tired of the over-done, generic, beige tower. I had heard of people that build their own cases, but at the time I thought that was crazy and impossible. I ended up buying the CM Stacker 830 and modding it slightly to fit what I wanted. The design was much better, but still lacking quite a bit.

 After a few months I grew tired of leaving everything the same. I got the itch and I had to run with it. There weren't any new components available in my very limited price range, so I started looking around for something else to try. I soon learned about liquid cooling from lurking on a few forums and I was fascinated by the idea. It was definitely a way to add expression to my computer. You can actually give a computer a certain measure of personality by playing with color schemes and tubing runs.

 I was content with my new creation for a while, but I was annoyed at how big the whole unit was and how difficult it was to swap out components. Even pulling out a hard drive took some serious time. I looked and looked for a store-bought case that would solve these problems, but there was really nothing. I finally realized that if I wanted the perfect case I would have to make it myself. I planned the case out in my head on and off for about a month, and once I had it perfected I modeled it in Google's Sketchup.

 The design was, in my mind, the perfect blend of form and function. It was small and efficient, and even made sense thermally. The bottom chamber separated the power supply, hard drives and radiators from the heat-sensitive components up top. The side radiator draws in cool air from the outside while the power supply and front rad expell it. This provides more than adequate air-flow to the hard drives and pump while ensuring maximum cooling power for the CPU, graphics card and northbridge. And all this with the added benefit of being seriously attractive.

  I believe this computer makes an important statement. I once asked my brother, an art history major, if a computer could ever be art. He told me, unequivocally, no. I can only hope that after seeing this piece, the true answer is obvious enough. Art is anything that inspires passion and emotion in the viewer, and it is my sincere hope that I can achieve this with my work.

The Liquid Nitrogen Experiment

So I've been wanting to try liquid nitrogen cooling for a long time. I've seen it around here and there and it's always seemed like the most extreme and fun thing to do with a home-made computer. The only problem is it's expensive and complicated, so I really needed a good excuse to try it out. Well, the opportunity came in Chem class Junior year. All non-AP science classes have to do a year long project and I figured this would be the perfect overkill project. I ended up spending well over 200 hours on it, but it ranks pretty high as one of my best High School experiences.

The design and planning stage took the longest at somewhere around 150-170 hours. I needed a wide range of equipment for the project, and since essentially none of it is sold in stores I ended up having to make most of it myself. First on the list was a liquid nitrogen pot for the graphics card. This would mount directly to the gpu core and serve as the interface between it and the liquid nitrogen bubbling away on the inside. The final design can be seen in action above. Due to inefficiencies in the design the final temperature was measured to be -147C (LN2 boils at -196C). One of the other revisions I designed can be seen below. It was too costly to machine and had to be simplified.

After many hours of thought, I came up with a design that would prove to be much cheaper and much more efficient. It would be milled from a solid block of copper and be fitted with an aluminum extension tube to allow for a higher volume of liquid nitrogen.

 

At this point I was still in contact with a machine shop to mill the pot for me, but a week before my experiment was scheduled to begin they decided they just didn't have time for a small job. Understandable, of course, but a little more notice would have been greatly appreciated. I had to get the pot made somehow, so I called up a friend and we decided to give it a go with his dad's drill press. After 13 straight hours and many, many problems we had a finished product.

 

 

One of the many complexities of liquid nitrogen cooling is how to insulate the sensitive electronics surrounding the pots from condensation and extreme temperature swings. I thought this would be a relatively easy problem to solve; I bought some Armaflex pipe insulation from Home Depot and used it for my first benching session. Well, the session lasted for 10 hours or so and I ended up having to throw all the data out because the Armaflex was preventing contact between the pots and the processor cores.

 

Needless to say, I went to sleep very frustrated that night, but the next day I found a solution that would fix everything. As I soon learned, artist's eraser, though soft and moldable, is a much more effective insulator than Armaflex. I used it to create a thin layer of insulation ontop of the PCB that wouldn't extend past the processor die.

 

 Though I didn't use Armaflex on the PCBs, I used it extensively on the pots themselves to keep them from freezing up too badly.

 

Another issue is how to accurately guage the temperature of the processors once the liquid nitrogen is added. Below -40C or so the onboard temperature sensors are useless, so I used a type-k thermocouple with a probe embedded in the base of the pots. Unfortunately I only had one probe so I had to switch it back and forth between the CPU and GPU during the tests. It's not the easiest thing to keep two pots with just the right amount of ln2 in them without knowing exactly what the temperature is. CPUs are especially sensitive to temperature swings so I kept the probe there the most, but GPUs put out much more heat and so eat through liquid nitrogen much faster. There were several occasions where the GPU pot ran dry because I was so busy tweaking in the BIOS and watching the CPU temperature.

 

I made extensive use of the computer's BIOS (Basic Input-Output Service) to tweak the processors. This controls the interactions between every major part on the motherboard and so there are pages upon pages of settings. You would think that only the CPU settings would make a difference, but I found that completely unrelated settings sometimes raised performance significantly. It was my job to hunt those settings down and find the board's sweet spot in order to get the maximum possible overclock.

 

The picture above shows a 100% overclock. That is, I was able to run the processor at twice the manufacturer's rated speed. That's quite a big boost, but it doesn't stop there. By the end of the day I had the same CPU running at 5.3 GHz. That's a whopping 239% increase over manufacturer spec.

I spent another few dozen hours running benchmarks and recording all the data, and then a few dozen more writing up a 19-page report on my findings. At the end I provided a concept for the large-scale supercooling of servers using an autocascade refrigeration system (as shown below, unit and photo courtesy of Michael St. Pierre).

 

This device would be able to cool multiple processors down to -155C while using no more power than the AC units that already cool server rooms. I hope to build one of these units as an independent project this winter.

For more information you can contact me at chaykak2@gmail.com