Earth system and space science applications on current and emerging high performance computing architectures
|Convener: Xavier Lapillonne | Co-Conveners: Ebru Bozdag , Valentine Anantharaj , Oliver Fuhrer , R. S. Nanjundiah , Peter Düben|
Numerical simulations have become the third pillar of science and are of great importance for advancing our understanding of geophysical systems such as the atmosphere, the ocean and the earth’s crust. In the past years intra-node parallelism on high performance computers has continuously been increasing, either because of the increasing number of cores on CPUs or because of accelerators such as GPUs or Intel Xeon Phi. This trend is particularly reflected in the latest TOP500 ranking where many of the top-ranking supercomputers are heterogeneous systems equipped with accelerators. In order to fully benefit from this hardware evolution applications often require substantial modifications.
This session will bring together experts in high performance computing and experts in computing models of geophysical systems to discuss tools, approaches and issues related to the adaption of earth science codes to new and emerging supercomputers.
The session particularly encourages presentations and discussions related to the following topics:
- Analysis of application performance and identification of crucial bottlenecks which may be relevant to the wider community.
- Experience in adapting codes to emerging high performance computing architectures.
- Implementation approaches: re-write using new programming paradigms (CUDA, OpenCL, CHAPEL, ...), Domain Specific Languages (DSL) or incremental adaptation using compiler directives.
- Achieving performance portability: Is it possible to achieve good performance with a single code across various platforms ?
- Achieving high degree of parallelism and scalability.
- Novel hardware architectures of high relevance for geophysical applications.
Solicited speakers :
Dimitri Komatitsch, CNRS, Laboratory of Mechanics and Acoustics, Marseille, France