Light-water nuclear reactors—those that use water as a coolant—only use 5% of their fuel before it has to be replaced; 95% of the energy remains in the fuel rods. Scientists have been trying design reactors that burn more of the fuel efficiently, to lead to less toxic waste and longer burning reactors.
But rather than experiment them on the ground or in labs, scientists test reactor and fuel rod arrangements in elaborate computer simulations. This saves time and money during the experimental and design phase. Using software that plows through the expected life cycle of a reaction, scientists can use modify reactor design to see how it changes.
But reactors being used today were modeled and designed years ago using comparatively primitive computing power.
"When they first started building the reactors, there was a lot of conservatism built into it," Tom Evans, a researcher at Oak Ridge National Laboratories tells Co.Exist "The calculations they could perform to figure out how long you could burn were limited by the computing power," Evans says, which means fuel is replaced way on the early side, just to be safe. That’s a process that’s expensive and takes time.
Titan, a new supercomputer installed at the Oak Ridge National Labs, is powerful enough to up the speed at which we can watch these reactions take place, and offer a better view of how those reactions pan out.
But better computing power gives reactor designers and operators a more confident idea of when fuel is spent. Say needed to plan an oil change for your car, Evans explains. "Before, they might have been able to tell you [to change it in] three years, plus or minus a year and a half," so of course, to be sure you didn’t do any damage, you might change it in a year. But better computing power would tell you "three years plus or minus three months."
Titan is 3.5 times faster than Jaguar, its predecessor at Oak Ridge, when it comes to locating the millions of particles bouncing around in a simulated nuclear reaction: A task that took Jaguar 60 hours to do, Titan will crush in just 13.
Titan comfortably clears 20 petaflops, which means it can process 20,000 trillion calculations each second. Part of that speed comes from the way it’s built. It’s been loaded up with GPUs—graphics processing units—the device that powers video games or other graphics-intesive displays. Rather than crunching commands one after the other as conventional central processing units (CPUs) do, GPUs can chew through several simple tasks all at the same time, in parallel. This makes them uniquely suited to certain kinds of computing work—like climate modeling, combustion simulations, and nuclear reactions— all of which they can burn through faster, and using less energy than CPUs would. Safer and longer-lasting nuclear power, then, may be just a few supercomputing cycles away.
Note: This piece formerly stated that nuclear plants use 95% of their energy, not that 95% of the energy remains in the spent fuel rods. We regret the error.