Benjamin Barsdell
Scientists, especially astronomers, are always searching for more computing power with which to run their latest simulations or reduce their latest data sets. While the speeds of traditional computers (i.e., CPUs) continue to increase, some researchers have begun to exploit alternative computing architectures. One such architecture, which we can thank the booming video-gaming industry for, is the Graphics Processing Unit (GPU). Relative to current CPU hardware, GPUs provide theoretical speed-ups of over 100x, and for certain problems such numbers have already been measured. But exactly which problems are likely to see these kind of performance boosts? The architecture of a GPU is quite different to that of a traditional CPU, and translating a problem for the GPU can be a challenging task. My work with David Barnes and Chris Fluke takes a step back from ad-hoc GPU implementations of common problems in astronomy and analyses the underlying algorithms, with the goal of being able to categorise problems for which GPU-efficient algorithms exist. Once we have an understanding of which problems are suitable for GPU implementation, the task of making efficient use of GPUs can be generalised. Further, the work will apply not only to GPUs, but to all future computer architectures. It is my hope that many areas of astronomy and astrophysics will benefit from both this analysis and the increased computing power offered by advanced computing architectures.
| Email |
|
|
| Phone |
|
+61 3 9214 4399 |
| Fax |
|
+61 3 9214 8797 |
| Office |
|
AR 206 |
