Seminar
The Future of High Performance Computing for Astrophysicists

Dr. Lars Koesterke

Abstract

With the advent of GPU accelerators the landscape of High Performance Computing has started to change rapidly. While this is in principle good news, the increased compute power comes with a steep price tag in that new languages (CUDA, OpenCL) must be used. Recently Intel has announced their own coprocessor Many Integrated Cores (MIC) technology which will deliver competitive performance but will be programmed through familiar languages (Fortran, C/C++ and OpenMP). In my talk I will introduce Intel's MIC architecture and will discuss the ongoing efforts at the Texas Advanced Computing Center to build a 10 PetaFlop cluster with MIC coprocessors in early 2013. Coprocessors (MIC) and accelerators (GPU) are here to stay and the changing hardware will spur considerable changes in general software design. Astrophysics codes of all varieties (for example highly parallel simulations, data-intensive software pipelines for large surveys, and even data reduction software on desktops) will have to adapt to the new environment. I will discuss software design, performance considerations, and optimizations in general and specifically with respect to the MIC technology. In the second part of my talk I will introduce the software package ASSET (Advanced Spectral Synthesis 3D Tool) that allows for the fast and efficient calculation of spectra from 3D hydrodynamical models and will highlight recent projects that have employed high-resolution (> 1,000,000), wide-range (1000's of Angstroem) synthetic spectra derived from 3D radiation transfer.

About the talk