In some of my previous entries, I mentioned a number of small computers. Here is one which performs parallel numerical computing within a small power budget. The entry level version is the Parallella Board from Adapteva.
Software and PCB have been open-sourced. Quite a number of papers have been written based upon compute projects using the platform:
Adapteva Epiphany MIMD architecture is a multicore parallel computing mesh network-on-chip (NoC). Each mesh node consists of a RISC CPU, a DMA Engine, 32 KB of shared local memory, and a mesh network interface. The entire architecture is scalable to thousands of cores in a single chip [1]. The 16-core Epiphany III coprocessor was integrated into the Parallella minicomputer platform [2]. The Parallella board combines the Epiphany coprocessor with a dual core Advanced RISC Machine (ARM) Central Processing Unit (CPU) in a Zynq 7010, and asymmetric shared-memory access to off-chip global memory.
A 64 core version should be available, but I didn't see any ready links to it. And a 1024 core is supposed to be released from the foundary at any moment. Supercomputing in a budget!