Supernovas — the spectacular explosions that occur in dying stars — may hold the answers to many questions about the origins of life.
The cataclysmic events are responsible for spreading throughout the universe the chemical elements formed within stars. Researchers at the U.S. Department of Energy’s Oak Ridge National Laboratory in Oak Ridge, Tenn., are using the Summit supercomputer to shed new light on how these elements come to be.
“We’re trying to build models of supernovae computationally to understand why there’s more of one kind of stuff than another kind of stuff,” said Bronson Messer, senior scientist at the Oak Ridge Leadership Computing Facility. “Why is gold rare? Why is there iron in our blood and not something like copper or something else? Why is there a difference between the amount of nitrogen and oxygen that we have in the atmosphere, and gases like xenon and neon?”
To answer these questions, Messer said, researchers have to gain a better understanding of what happens during a supernova, and how the phenomena disperse elements into interstellar space. And that requires computing power — a lot of it.
Enter Summit, the world’s fastest supercomputer, with more than 27,000 NVIDIA V100 Tensor Core GPUs. The system’s combination of performance and memory is enabling Messer and his team to create more powerful networks that can model the nuclear burning of more isotopes of elements, referred to as species.
Nuclear burning is the fusion process that continues after a supernova explosion. Nuclei combine, transmute one kind of matter into another, and then release that into space.
“In the past, people have typically used a baker’s dozen of species of hydrogen or helium, all the way up to iron, to be able to model nuclear burning in situ during a simulation,” said Messer. “With the power of Summit, we’re able to improve that by a factor of more than 10. We can actually go to 160 species.”
Down the line, Messer hopes to establish large enough nuclear burning networks to accommodate 2,000 species, which should be enough to analyze the yields of heavier elements such as gold and uranium.
Eventually, he foresees the work enabling researchers to better understand the matter in the interior of a star, which can be dense enough to pack the collective weight of every human on Earth into a cubic centimeter of material.
Moving Data with NVLink
One of the biggest challenges in work like this has been the movement of all the data to and from GPUs for parallel processing. Summit addresses that with NVIDIA NVLink interconnect technology, which provides lightning-fast connections between its many GPUs and its more than 9,200 IBM Power-9 NVLink-enabled CPUs.
“We no longer have to design our code so that we cover the latency of transferring data to and from the GPU,” said Messer. “NVLink changes our whole view of a GPU-enabled mode.”
When Messer and his team finish their first simulation on Summit, they expect to have a database of accurate isotopic data on the exact makeup of a supernova explosion. That data can potentially be compared to observations from optical and gamma ray telescopes, and to data on elements collected from meteorites that fall to Earth.
Once they’re armed with a better grasp of how the elements that make up life are formed, scientists will be one step closer to understanding how human beings came to be.
To further accelerate the journey to that knowledge, Messer is considering putting deep learning methods to work so that the simulations, which can run for weeks, are monitored and analyzed around the clock. Doing so would make it possible to deal with potential problems quickly, thereby preserving valuable compute time.
Such additional advances aside, scientists like Messer are already pinching themselves at seeing what a GPU-powered supercomputing beast like Summit enables them to tackle.
“We’ll be able to explore new parts of the nuclear landscape and make very precise predictions of what elements are formed and in what proportions,” said Messer. “It’s simply not possible to even conceive of trying to do this without the memory that is available on each of the GPUs in Summit.”