A heterogeneous supercomputer model for high-performance parallel computing pedagogy

To visually illustrate the impact of architecture and communication on parallel computing we created a model supercomputer in the spirit of other engineering models. Combining Raspberry Pi computers with an Nvidia TK1 the resulting machine is distinguished by it´s instructional utility, asymmetric CPU and communication channel speed, and for incorporating current trends in heterogeneous supercomputing. Running Linux, the model supports major high-performance computing software environments, including OpenMP, MPI, and Nvidia´s CUDA. The model architecture includes aspects of heterogeneous, supercomputers such as shared- and distributed-memory MIMD, and 192 CUDA core SIMD-like processors.