BOF Heterogeneous Systems in Computational Sciences
Tomasz Bednarz - CSIRO Mathematics Informatics and Statistics, Sydney, Australia
John A Taylor - CSIRO Mathematics Informatics and Statistics, Canberra, Australia
Luke Domanski - CSIRO IM&T, Sydney, Australia
Alex Khassapov - CSIRO IM&T, Melbourne, Australia
The Heterogeneous Systems in Computational Sciences will provide attendees with an intensive forum, which covers both a general introduction as well more advanced topics, showcasing CUDA, OpenCL, OpenACC as modern massively-parallel programming environments. Invited guests from both industry and academia will discuss a range of subjects, including core fundamentals, hardware architectures, parallel programming, workload scheduling as well as various scientific applications utilizing CUDA, OpenACC and OpenCL in Computational Simulation Sciences including: Computational Fluid Dynamics, image analysis and image processing techniques, environmental modeling, biomedical modeling, etc. It will be also great opportunity to meet in person members of Australia GPU community [1-4] and possibly join remotely with groups from the USA.
SCHEDULE, MONDAY 29TH OCTOBER 2012
Welcome, Australian GPU Meetups and Khronos Chapters (Dr Tomasz Bednarz/CSIRO)
Computational Simulation Sciences and eResearch at CSIRO (Dr John Taylor/CSIRO)
Supporting eResearch-ers in a heterogeneous HPC world” (Dr Luke Domanski/CSIRO)
CT and Imaging tools for Windows HPC clusters and Azure Cloud (Alex Khassapov/CSIRO)
Supporting eResearch-ers in a heterogeneos HPC world – Dr Luke Domanski
One challenge for large multidiscipline science organizations in the uptake of eResearch services, particularly in the face of modern heterogeneous and massively multi-core technologies, is the porting of complex serial science codes to high performance computing (HPC) platforms, and the ongoing training and support of scientists using and developing these codes. EResearch offers the tools for scientists to make new and significant discoveries on problems once consider infeasible to solve, however, encouraging and coordinating HPC activities amongst an army of scientists with different disciplinary backgrounds, goals, and computing skills, using a broad range of software and programming languages, can be a daunting task. Some recent efforts to address this challenge will be discussed, and pros and cons of our approaches will be highlighted.