Often when I’m working a project, I’ll work through a series of ideas before settling on something that fits the bill. I try to write down these ideas and keep them around on the off chance that I could utilize them for a future project. Not long ago, I was working on a different computer project, assessing the cooling requirements, and I thought to myself, “It sure would be a cool to build a fully functional, small scale wind tunnel as a case for a computer.”
Unfortunately, I didn’t have time to investigate this wind tunnel-computer idea, and it sat around in some distant corner of my brain, until I decided to build a new computer. A medical research computer that would donate its time to cancer research.
In 2011 I became actively interested in grid computing, and specifically in World Community Grid.
The idea that I could build a computer, or use existing computer resources and donate their power so scientists and researchers could process medical and humanitarian research was extremely interesting.
By donating computer processing time, you actively contribute towards a great cause. World Community Grid has numerous projects available; finding cures and treatments for cancer, AIDS, malaria, muscular dystrophy, etc.
As I became more interested I found myself connecting all the computers I had in my home. Eventually I decided I wanted to do even more; I wanted to build a computer that could donate all of its time to processing medical and research data. That project culminated in the creation of the Lego Folding Farm. This system housed three separate computer systems, all running in one giant Lego case. The system went online in July of 2011 and has been running 24/7 ever since.
In the past year, World Community Grid moved closer to enabling a GPU (Graphics Processing Unit) based project, the Help Conquer Cancer project. GPU computing allows a project to utilize the full processing power of the GPU. In most cases, a project written to run on a GPU is significantly faster than its CPU counterpart. In the Help Conquer Cancer project, a single CPU workunit took around an hour for my other computer to run, where the GPU enabled version completed in less than 10 minutes.
While the GPU version of Help Conquer Cancer project was being developed, I felt personally challenged to donate more towards cancer research. I’ve had people in my family affected by various forms of cancer, and I’ve always felt like I wanted do more to help. But beyond donating money to cancer causes or doing cancer walks (which are also a great way to help show support) it’s hard for regular folks like myself to feel like I’m helping contribute. It seemed like perfect timing that World Community Grid was bringing onboard a project that would allow for significant increases in the speed at which cancer research could be completed. I felt very compelled to actively pursue adding more resources towards this cause.
My Lego Folding Farm is a farm of CPUs and was built to process numerous different medical and humanitarian projects. At the time it was built there was not a GPU enabled project that ran on World Community Grid, so I focused all of my efforts on CPU processing power. The GPUs that the Lego folding farm uses are nothing special. While they can process a GPU enabled project they don’t really have a lot of processing power for that task. With that in mind I decided to build a GPU processing system that would solely dedicate its time towards cancer research.
I investigated the expenses to build and run this system and realized that the component costs would be high, as were the added electricity costs required to keep the system running 24/7. I had already added a significant expense 2 years ago with the electricity required to power the Lego folding farm. I wasn’t sure I could afford the cost of the system itself, as well as another electric bill increase.
Unable to solve the funding problem on my own, I turned to my resident creative genius department @thetinnishflash and explained the issue. After some discussion, we decided to try and build this system with donations from friends, family and others that wanted to be involved in the fight against cancer.
We turned to the fundraising site Indiegogo and setup a project campaign to raise funds to defer the costs of the components as well as the electricity costs. After a couple of weeks working we were able to raise a significant sum of money, almost enough to cover the costs of the components themselves! Outstanding! I’m still so very appreciative to all those who donated and helped contribute to this project happening, without their donations it would have been a hardship to complete.
Getting the of support others, with friends and family coming together to help make something happen, was my favorite part of this project. It’s a great feeling being involved and getting to build something that others have contributed towards.
Once the fundraising was completed I began gathering up components for the build. I dubbed this project “Cancer Supercomputer” and I had fully intended on outfitting it with the components needed to be a GPU processing powerhouse. After hours of research I settled on a list of components.
CPU- Ivy Bridge 3770K
GPU- Radeon 7970 (Sapphire Dual-X cards)
RAM- Corsair Vengence 8GB 2133Mhz
SSD- Mushkin Callisto 40GB
PSU- Rosewill Fortress 650W Platinum
Motherboard- Gigabyte Sniper M3
CPU Cooler- Phanteks 140mm
My plan was to configure the computer and then overclock the hell of it. With this in mind, I started wondering what route I wanted to go as far as cooling. I debated going with water cooling, but decided instead to see how far I could push air cooling.
During the design phase of the Lego Folding Farm, I decided that the best method of cooling many hot components in a tight space was a very short path for airflow. Conventional computer cases generally intake air from the front which then travels through the case and rises upwards. Depending on the configuration there are numerous paths for the hot air to exit. There is almost always a rear exhaust fan, and in some cases a top mounted exhaust fan, as well. Additionally, some air will exit through the power supply fan. The problem I see with this method of air cooling is that the path that the airflow must take is long and winding.
To work around this, I built the Lego Folding Farm with a very short path from intake to exhaust. The intake fans sit directly in front of the components and blow air across the CPU cooler, the GPU cooler, the motherboards on both top and bottom, the heatsinks, the power regulation circuitry, the power supply. These are directly cooled by unobstructed airflow. The exhaust side is a mirror of the intake path, where air is directly exhausted out of the case. This method allowed for a much shorter path to get fresh, cool air in and blowing directly on the components.
However, with the Cancer Supercomputer I knew the graphics cards were very long and had very big coolers on them, so I needed to change my approach. I also knew from testing that under full load the GPUs could output a lot of heat, especially with multiple GPUs in a tight space. I needed a solution that allowed for more space, while also allowing me to move a lot of air.
Additionally, I wanted to pay close attention to air pressure. A conventional computer case has a lot of large “dead-space” areas. These areas don’t direct or force air to flow towards the components that are being cooled. I wanted to avoid dead space. I wanted the case to closely mirror the shape or profile of the finished motherboard with components. This way all the air flow would be forced to travel across the components surfaces, as opposed to flowing past large empty areas of case.
(My drawing skills are clearly unmatched)
Wind Tunnel Concept
Knowing I needed to maximize cooling, I explored a concept I’d tinkered with previously; a wind tunnel computer case. I had done research on wind tunnels back when I first investigated using a wind tunnel to cool components. There was a lot of information, much of it only helpful to those with a physics background. I’m not a physicist, but given enough time I can usually understand theories well enough to apply them in the real world and test them out, and that was the case here.
I used an anemometer (velocity/airspeed meter) for testing, as well as different fans and basic shapes made out of cardboard. I was able to conclude that I could increase air velocity through a scale wind tunnel.
There are two types of wind tunnels; the one I’ve built is a subsonic wind tunnel. This type of design involves a contraction section which is used to increase velocity (airspeed) through the test section. This increase in airspeed was what I looking for, a way to increase airspeed over the computer components.
There are numerous factors to consider when it comes to wind tunnel design and testing, things like Reynolds numbers, turbulence, boundary layer air, and heat from surface friction. I did extensive testing with various designs, materials and configurations until I settled on a final design.
In a perfect world I would have been able to build a larger wind tunnel. There were many constraints on this project to make it workable. The biggest of these was space. I’ve slowly converted my basement to a lab with toys, computers, Lego, robots and all kinds of various geekery. Therefore, I only had one area with enough space to construct a wind tunnel. The space itself was about 72″ long by about 26″ wide, and whatever I was going to build had to fit within those confines.
Next, I needed to fit computer components in the test section of the wind tunnel. I needed the test/contraction section to be big enough to accommodate the motherboard/CPU cooler/GPUs and also fit the power supply and a SSD. After tinkering with several designs I settled on a shape and size that allowed room for further growth. The contraction section of this wind tunnel has enough space for two full mATX motherboards plus the PSU. This allows for the possibility of adding another system in the future and cooling two systems inside one wind tunnel.
Next I had to contend with fan noise, as I would with any other air cooled pc. I debated using various sized PC fans and configuring them in a grid pattern at the inlet. Using fans around 220mm would’ve allowed for multiple fans at the inlet without too much noise. But when testing those fans I found that the airflow and CFM (cubic feet minute) produced weren’t adequate. At that point I changed direction and decided that I would use a single large fan with multiple speeds. This would produce a much higher airspeed as well much more CFM. I tested a couple of different fans and settled in on a box fan. It only c