Abstract:
Less than three years have passed since the hybrid GPGPU-based systems occupied the first positions in the Top-500 list [1]. Again, the world of supercomputing changed unexpectedly and rapidly. What conclusions from this technology change can be done right now? What is the nearest future forecast? What should the supercomputer designers do, and what the supercomputer users should get ready for?
Keywords:supercomputer, high-performance computing, supercomputer architecture, GPGPU, light and heavy compute nodes, MVS-Express interconnect, hybrid applications development problems, heterogeneous communications, FPGA computations.