Joshua Kiepert’s Raspberry Pi-based desktop supercomputer
This next board really needs no introduction. Some even consider it the father of the 21st century's modern-era single-board computers. Of course, we’re talking about the versatile Raspberry Pi, which has been used as the driving platform in everything from robotics to gaming systems. The board was released back in 2012 and has since been used as a cheap solution for makers and developers to incorporate into their projects. Even I have used one in a few projects.
Joshua Kiepert’s Raspberry Pi-based desktop supercomputer -- a.k.a., the Beowulf cluster research project.
The Pi has also been used to build relatively cheap and powerful supercomputers, such as Joshua Kiepert’s “Beowulf cluster research project.” A Beowulf cluster is a group of identical computers that are strung together or run parallel with each other in order to tackle complex problems. For Joshua’s research on data-sharing for wireless sensor networks, he needed access to a supercomputer to run his simulations.
His school (Boise State) had its own, however gaining access to it was sketchy at best due to the erratic times it would become available. His solution to the issue was to build his own supercomputer with the help of Raspberry Pi, which is not only cheap, but features a myriad of interfaces for embedded-systems use. These include the ever useful GPIO connector, as well as I2C, SPI, and UART. Joshua took advantage of the interfaces and strung together 32 of the boards to create his supercomputer. The boards were housed in an acrylic case outfitted with red LED fans to keep the Pis cool while under load.
With the board costing a mere $45 (model B with an 8GB SD card), the grand total for his supercomputer was only $1,440, which is cheap if you consider that a single node at his school’s supercomputer costs anywhere from $1,000 to $1,500. Of course, this isn’t the first supercomputer built entirely out of the Raspberry Pi SBC. Engineers from the University of Southampton built one using 64 of the boards back when they hit the market in 2012. The Raspberry Pi is certainly an attractive option for those looking to build their own data-crunching powerhouse, even more so than Nvidia’s TK1, since it’s significantly cheaper, albeit with less computational power.