There was a guy at the Makerfaire who had a Parallela based super computer and a Jetson [1] based one. He felt that by the time the Parallela stuff was running he had already surpassed it with the nVidia offering, faster and less money per Teraflop. It is an interesting space to be sure.
But how does trying to send fragments over the internet really work?
Yes, I know Brian's work well. His computers are awesome!
Counting TFLOPS is important for some, but for others it's the number of CPU cores that count. Better to compare real workloads. In the case of supercomputer.io the key will be software packages for machine learning and smartly distributing data across the network. There is a long history for this kind of work with folding@home and BOINC.
But how does trying to send fragments over the internet really work?
[1] http://www.nvidia.com/object/jetson-tk1-embedded-dev-kit.htm...