|Name:||BoF 10: Emerging Trends in HPC Systems & Application Evolution|
|Time:||Tuesday, June 20, 2017
01:45 pm - 02:45 pm
|Speaker:||Hans-Christian Hoppe, Intel|
|Marie-Christine Sawley, Intel|
|Abstract:||The development of HPC platforms in recent years shows three clear trends: addition of memory layers (with fast limited-size memory, and/or large capacity non-volatile memory), substantial increase in core counts and tighter integration of high-performance interconnects with compute nodes. All of these are driven by fundamental technology properties and application requirements and will thus continue into the foreseeable future. This BoF discusses the impact on HPC applications and the evolution of application codes that will be necessary to take best advantage of new HPC architectures, achieve high end-to-end performance and energy efficiency and at the same time allow sustainable code development and maintenance. For the latter objective, it will be key to limit the complexity of codes and programming models. Developers of significant HPC applications will present their approaches and experiences in code evolution, setting the scene for the follow-on panel--style discussion with the audience. We plan to include amongst others speakers from TU Munich (SEISOL and EXAHYPE), T-Systems/DLR (TAU code), ECMWF (IFS forecasting system) and CEA (material science, CFD). The overarching objective is to foster a global community of practitioners across funding organizations and schemes and to combine academic and industrial participants.
Targeted AudienceThis BoF will be of prime interest to developers of HPC and HPDA applications and programming models/systems. This audience will learn about successful approaches to code evolution/modernization, contribute to the discussion, and make their point of view heard. The secondary audience are system designers and architects, who will gain a better understanding of how applications can take advantage of new system features, of the challenges of doing so, besides being able to steer the discussion to the most relevant properties of future systems.