Climate simulations accurate right down to the neighborhood level. Personalized cancer treatment. An accurate map of the billions of neurons in the human brain. These are just some of the possible applications of the new Aurora supercomputer that’s being developed at Argonne National Laboratory in Lemont.
It’s set to be the fastest supercomputer in the world, capable of exascale computing: one quintillion operations per second. The project has been a dream for more than 10 years.
“We weren’t even sure it was possible to build machines this fast,” said Rick Stevens, Argonne’s associate lab director for computing, environment and life sciences. “These things have literally hundreds and hundreds of billions of transistors, they have millions of processing elements, they have many, many tens of thousands of memory chips.”
Stevens said it took years of progress to advance to a point where such a supercomputer would work for more than a few minutes at a time, not to mention operate without needing unsustainable levels of power.
The Aurora supercomputer is set to be operational in 2021, but it’s not just hardware and software engineers working on the project today. The researchers who will be putting the Aurora’s supercomputing powers into action are also a part of the design process. Argonne neuroscience researcher Bobby Kasthuri is one of them. He’s excited about the prospect of working with data sets so massive that only a supercomputer can handle them, with the ultimate goal of making neuroscience more exact and less ad hoc.
“I think big data does two things. One, it’s a place for hypotheses to go to die. There are tons of hypotheses about how brains work that aren’t going to be compatible with the actual description of the brain,” Kasthuri said. “The thing I’m actually interested in is … figuring out ways to surprise myself with the data that I’ve collected. One way to do that is to collect a ton of data in an objective, arbitrary, unbiased fashion. And in that data set, if you did it right, should be a bunch of surprises for how we think about the brain.”
Stevens thinks supercomputers with capabilities like the Aurora are also key to preserving the country’s “leadership role in science and engineering.” And, he says, they’re so powerful they won’t just be for big data – or for any single demanding task.
“What’s happening is more and more we’re doing simulation and data analysis and artificial intelligence, all on the same hardware,” Stevens said. “It’s the beginning of a new class of supercomputers.”