GPU Computing: Where Science Meets Games

by The Computing Committee | UW-Madison Astronomy Department
Posted Aug 31, 2011

It's not often that video games are linked with scientific progress.

However, there is growing interest in tapping an unused computational resource found in every personal computer: the video card. This is the hardware module responsible for displaying text, images, movies and 3-D graphics on the screen. The sophistication of video cards has grown exponentially over the past decade, driven by the increasing graphics demands of video games. The most significant advances have occurred in the computing power of the graphics processing unit (GPU), the chip at the heart of a video card which is responsible for the many arithmetic, geometric and shading operations required to draw 3-D photo-realistic scenes at cinematic frame rates.

Today, a mid-range laptop computer typically contains a GPU capable of over 100 billion operations per second — around ten times the computing power of the central processing unit (CPU) which comprises the 'brain' of the computer. It's little surprise, therefore, that interest has arisen in applying GPUs to number-crunching tasks that are scientific, as opposed to graphical, in nature.

"Once the major chip manufacturers started producing fully programmable GPUs, back in 2005, it became a no-brainer to become involved in so-called 'GPU computing'", remarks professor Rich Townsend, who leads the Massive Stars group within the Department of Astronomy. "Many of the computational problems we face during our day-to-day research involve solving the same problem again and again (maybe millions or even billions of times) with slightly different parameters. This sort of task is ideally suited to GPUs, which offer an unparalleled bang-for-the-buck.", he adds.

The Massive Stars group is currently involved in a number of different GPU computing projects, all supported under the aegis of an ongoing 3-year 'Advanced Technologies and Instrumentation' grant from the National Science Foundation. Last year, Townsend worked in collaboration with professor Karu Sankaralingam (Computer Science) and graduate student Matt Sinclair (Electrical Engineering) to develop GPU software for fast analysis of oscillating stars. These stars are undergoing motions analogous to terrestrial earthquakes, and by studying the light variations caused by the motions it is possible to figure out the stars' internal structure — a technique known as 'asteroseismology'. "My interest in asteroseismology extends all the way back to my original PhD thesis; but this GPU work is particularly well timed to meet the steep data analysis challenges presented by the Kepler space mission, which is looking both for Earth-like planets and for stellar oscillations", says Townsend.

At the beginning of this year, Townsend invested in a new server computer hosting 6 NVIDIA GPUs. This computer has a throughput approaching 6,000 billion operations per second, and is the fastest machine in the Department. It was purchased primarily to support the asteroseismic analysis research, but since then a number of side projects have sprung up involving graduate students within Townsend's group. Nick Hill is investigating whether the core graphical functionality of GPUs can be put to use in modeling oddball stars which show, e.g., departures from spherical symmetry due to rapid rotation, or surface inhomogeneities due to magnetic fields. Brittin Borland is using GPUs to calculate light curves for eclipsing binary systems, with the ultimate goal of speeding up the planet-hunting component of the Kepler data analysis. And Chris Bard has recently been awarded a prestigious NASA Graduate Student Research Fellowship, to develop GPU-based magnetohydrodynamical simulations of the magnetospheres of the Earth and of massive stars.

But for Townsend and his team, it's not always about the science. "I've been a video game enthusiast since I was six, and I continue to play online on a regular basis", confesses Townsend. And in the Massive Stars group, he's in like-minded company. "All of us get a huge buzz from knowing that the GPUs responsible for our immersive gaming experiences can also push our scientific research way beyond the limitations of conventional, CPU-based computing".


UW-Madison Astronomy Home