Over at Parallella.org, Adapteva Founder & CEO Andreas Olofsson writes that the organization is seeking help from its user community to “help build the most efficient virtual supercomputer in the world this week.”
Thanks to the amazing folks at resin.io there is now a containerized application deployment platform for the Parallella board enabling super easy workload scheduling for a fleet of thousands of Parallella computers across the globe with the push of a button. We intend to offer “supercomputer.io” capacity for free to academic researchers working in the field of machine learning. Researchers at Google, Facebook, Baidu, and Microsoft are making steady progress with deep learning algorithms thanks to virtually limitless resources. What if this kind of computing power could be available to all researchers for FREE? If we can get 1,000 boards connected to supercomputer.io, researchers would have a 18,000 CPU supercomputer at their disposal, putting it in the same class as some of the computers on the Top500 while consuming less than 5 KW. With 2 ARM A9 cores, the Xilinx FPGA logic, and 16 Epiphany CPU cores, this would arguably be the most efficient supercomputer on the planet!
In this slidecast, Andreas Olofsson describes the Parallella project, an open-source development platform for low-power parallel computing.