GPGPU Solutions for Astronomical Radio Interferometry

Proposer: Michelle Kuttel

Abbreviation: ASTRO

Brief Description: Radio interferometry requires considerable computing power to extract information of interest (e.g. images) from raw astronomical data. The hugely increased size and complexity of the new South African radio telescope – MeerKAT – that will come online in 2012 will raise these computational requirements by orders of magnitude. There is considerable potential for the development of parallel algorithms to take advantage of the emerging low-cost, highly parallel hybrid architectures – such as multi-core CPUs in combination with GPUs. This project aims to develop highly multithreaded software for aperture synthesis deconvolution and interactive 3D image data visualization.

Computer Science Content: Parallel algorithms, multi-threading, numerical algorithm development, scientific visualization, software engineering

Specific Learning Outcomes: You will learn some basics of radio interferometry, general purpose GPU computing and open source scientific software engineering. You will have experience with direct contact with the “clients” (Astronomy department) .

Skills Required by Team as a Whole:

• Theory: Numerical algorithms, parallel algorithm design and effciency, basics of radiointerferometry

• Implementation: This will be challenging, but interesting. Good C/C++ coding ability is required and students will have to learn general purpose GPU programming with Cuda.

Facilities needed: PC equiped with recent GPGPU-capable Nvidia GPU

Supervision: This will be a first project in a new collboration between the Astronomy and Computer Science departments and we are all very enthusiastic about it, “we” being Michelle Kuttel (advice on GPU/parallel solutions), Patrick Marais (Graphics, image processing and compression), Kurt van der Heyden (Radio interferometry) and Ian Stewart (Radio interferometry).

Number of Students: 3 or 2