When on topic, astrophysicists tend to think—and speak—rather sweepingly. Where we normal folk “count the minutes” until we meet again and deliberate long and hard over the extra distance we’re forced to travel to get to the really good restaurant, those who earnestly study the cosmos tend to see a million years as a few moments in time and a million miles as a mere jaunt.

However, that perspective shifts dramatically when it comes to the efficiency of their research. Suddenly, hours and weeks and months are a Really Big Deal. And that’s why the trend of video cards evolving into full-fledged general purpose GPUs has made an astrophysicist by the name of Patrik Jonsson, working at the Harvard-Smithsonian Center for Astrophysics in Cambridge, MA, one very happy astral investigator.
Jonsson, you see, is a man on a mission—a mission that incorporates not only our own world and solar system but  also all the other solar systems, all the other galaxies, and essentially all the important stuff that’s happened since the Big Bang, too. The end goal: to better understand how the universe, and in particular our own Milky Way, came to be what it is today.

Dust—it’s everywhere!
Yet many of the answers lie not in massive celestial events and inconceivably enormous star systems, but in something infinitesimally smaller. Something as small as…dust—space dust, to be exact.

According to Jonsson, a better understanding of space dust, which in itself is a byproduct of past star explosions, is crucial to a better understanding of the current condition of the cosmos. Scientists, it seems, have a pretty good grasp on what they’re able to see through today’s highly advanced optical telescopes. But it’s all the stuff they can’t see that’s so critical. And one of the primary reasons they can’t see beyond our immediate celestial neighborhood is the presence of all that nasty space dust.

Crazily prevalent in the universe, space dust hovers around virtually every galaxy out there and effectively blocks most of them from our inspection. To say it’s hampered serious exploration is an understatement. Yet if Jonsson and his colleagues have their way, it won’t prevent it.

Space dust, as it turns out, not only blocks light but also absorbs it. It then becomes heated by it, and in turn emits long-wavelength infrared radiation. And that’s a factor Jonsson is using to his advantage. His concept is called “Sunrise,” a computer simulation he developed that calculates radiation transfer through the dust and theoretically should help determine what lies beyond that dusty veil in real-life outer space. (You can see Sunrise in action at http://tinyurl.com/yd25lzj.)

GPU: The cosmic Swiffer?
There’s “only” one problem—certain elements of Sunrise take a very, very long time to compute. And that’s where GPUs come in—or at least Jonsson hopes they’ll come in soon. Sunrise currently runs on NASA’s immense CPU-server farm called Columbia, which consists of thousands of CPUs. However, GPUs specialize in parallel computing, wherein multiple “similar but different” calculations are carried out simultaneously. Because parallel calculations are such an essential part of the Sunrise simulation—in particular the time-consuming calculation of dust grain temperatures—a potential switch from CPU to GPU computing seemed like a great idea. So when NVIDIA provided Jonsson’s team with Tesla product for conducting initial tests, the answer was an emphatic “thank you!”

And the results of Jonsson’s tests with Sunrise have been extremely encouraging. In fact, in a recent paper entitled “Accelerating Dust Temperature Calculations with Graphics Processing Units,” Jonsson and associate Joel Primack, a professor of physics and astrophysics at the University of California, Santa Cruz, describe in exceptionally complex language and equations the benefits they’ve seen when GPUs are applied to Sunrise. The conclusions are startling: A single NVIDIA Tesla c1060 GPU can perform the dust grain temperature simulations 69 times faster than an eight-core CPU.

Despite that, Jonsson’s GPU experiment remains, for the time being at least, just that—an experiment. Although he’s proven that the Tesla GPU is much more
efficient to use than a comparable CPU when performing the parallel calculations Sunrise requires, it turns out that a machine with thousands of CPUs is still a better bet. Or rather, remains the only bet, as NASA does not yet have a GPU render farm to compete with NASA’s massive Columbia supercomputer, which is where Jonsson’s simulation currently whirrs away.

A GPU Future?
But that hasn’t stopped him from looking to the future—a future where GPU-equipped supercomputers are commonplace and easily accessible to the top-tier scientific community. According to Jonsson, “A big cluster of GPUs would make it possible to use the GPUs for what they excel at…overall, once we factor in the portions of our code that don’t take advantage of GPUs, we could run our simulations in about one quarter of the time it takes now. That would be a huge boost.”
Granted, Jonsson’s meticulous space dust project—be it CPU- or GPU-powered—is just part of man’s current investigatory trends into the mysteries of the universe. As Jonsson himself says, “Our research must be linked to the findings derived from direct observation.” Yet it would seem that any concept that helps man see beyond his normal visual range must be considered a key piece of a very complex puzzle.