As is often stated in the field of supercomputing: “Five years from now has already happened.” That means of course that the ground work and direction for HPC systems to be deployed in five or fewer years has already been established due to the long lead times involved in the development of system hardware and software, as well as applications. Although many specifics of the detailed plans are yet to be finalized, it is possible to project at this time the likely structures and parallel semantics that will be exhibited by the first generation of exascale computers at the beginning of the next decade. The last year has seen substantial progress in planning for this follow-on to the next era of computing internationally. China, the leading nation on the Top-500 list, has announced its intention of achieving exaflops computation by 2020, the US has initiated the major Exascale Computing Project (ECP), the Japanese have put in place the Post-K project to immediately proceed and prepare for exascale, and the Europeans are heavily engaged in their own plans for this performance milestone. Even as the world HPC community is lining up to make 1 Exaflops a reality, the very nature of the computing workloads to be performed at that scale is rapidly shifting towards a sophisticated synergy of conventional numeric processing and advanced concepts in data analytics and machine learning. This emerging symbiosis of what had been entirely separate domains is creating a new paradigm for HPC-driven investigation with, as a byproduct, expanded regimes of significance into societal implications such as health and medicine, transportation and power infrastructure, agriculture, and meteorology, among other fields. Thus, HPC is rapidly permeating into vaster areas effecting humanity than the narrow esoteric fields of simulation science and engineering. Finally, even with the tightening focus on delivery of practical exascale computing at the beginning of the next decade, exploration in post-Moore’s Law technologies like quantum computing and neuromorphic architectures are gaining early advances and adherents as the challenges of possible new paradigms are anticipated. This keynote presentation, the 14th in the series, will deliver a rapid-fire summary of the major accomplishments of the last year and the challenges of the next defining the trajectory in supercomputing and its application in the immediate future. As always, audience questions and comments are welcome and appreciated. |
;){kind=link}