AUDIE CORNISH, HOST:
From NPR News, this is ALL THINGS CONSIDERED. I'm Audie Cornish.
ROBERT SIEGEL, HOST:
And I'm Robert Siegel.
The world's fastest supercomputers are once again in the U.S. In June, the title was claimed by a machine named Sequoia at Lawrence Livermore Labs. Today, at the Department of Energy's Oak Ridge National Laboratory in Tennessee, what could be an even faster computer comes online. It's called Titan. As NPR's Steve Henn reports, it would not have been possible were it not for the massive market for video games.
STEVE HENN, BYLINE: The first thing you notice when you visit Titan is the noise.
(SOUNDBITE OF MACHINERY)
HENN: This room kind of reminds me of a 1980s era Kmart but much louder.
BUDDY BLAND: Yeah. Exactly. A very large room, it's about half an acre, high ceiling with fluorescent lights. We just don't have the blue light blinking in the corner.
HENN: Buddy Bland runs the Leadership Computing Facility at Oak Ridge National Lab. What are these big tubes? They look like, you know, the exhaust system on a Duisenberg or something coming off the top.
BLAND: So the big stainless steel pipes that come out of this are part of the cooling system. Cray, the company that builds this computer, is located in Chippewa Falls, Wisconsin, you know, which is a big dairy state. These stainless steel pipes actually come from the dairy industry. They're kind of off-the-shelf components that the computer manufacturer has been able to repurpose to cooling supercomputers.
HENN: Cray built special fans to cool each cabinet. The fans are so powerful, the floor in the room actually vibrates. All of this complicated engineering is in the service of speed. Titan is quite possibly the fastest computer in the world. We won't know for sure for a few weeks, but Titan is designed to do more than 20,000 trillion calculations every second. That's faster than half a million laptops. But this didn't come cheap. The machine costs $100 million. Its electric bill will be $9 million a year. So what makes all this computing worth the price?
BLAND: It gives us more fidelity. We're trying to simulate the real world, the physical world on these supercomputers. And if you think of them, they're really a time machine that let us look into the future and understand what's going to happen.
HENN: Higher fidelity and higher resolution models mean scientists can begin to predict phenomenon like climate change locally and regionally.
BLAND: At the high resolution, things like hurricanes and typhoons actually start to show up that never show up when you run the exact same simulation at just a lower resolution. That's an important outcome that policymakers need to know about.
HENN: Researchers will compete for computing time on this machine. It will be used to model everything from biochemistry to black holes to nuclear reactors.
JEREMY SMITH: My group is interested in understanding molecules of life, and we also want to understand how they wiggle about.
HENN: Jeremy Smith hopes by screening libraries of small proteins against known receptor sites his team could find new drugs for Parkinson's or prostate cancer.
SMITH: Until now, we've been able to do that with a few thousand small molecules, but with the new upgraded supercomputer, we'll be able to do tens of millions of these molecules.
HENN: And Smith fantasizes about an even faster machine, one that could screen for side effects too. But there's a problem. In the past decade, the traditional methods of making supercomputers faster hit a wall.
BLAND: Yeah. It really did hit a wall, and that's the power wall.
HENN: Buddy Bland says for years, you made microchips speedier by giving them more juice, more electricity, but Bland says there's just not enough cheap electricity to power the kinds of computers scientists like Smith dream about - 150 times faster than Titan.
BLAND: If we use today's technology, it would take two gigawatts. You know, it would take two nuclear power plants to power it.
HENN: So over the last few years, supercomputing engineers have found inspiration in an unlikely place:
(SOUNDBITE OF VIDEO GAME)
HENN: Video games. In today's games, the graphics are intense. Virtual water reflects and refracts virtual light. The rules of physics apply, and dozens of people can play these games together.
SUMIT GUPTA: Every object in that game is actually being modeled in 3-D.
HENN: Sumit Gupta runs accelerated computing at Nvidia, a company that makes chips designed for gamers. He says modern games require immensely complex calculations.
GUPTA: And you have to do billions and trillions of them per second.
HENN: Bland and Gupta say chips built for games consume much less power.
BLAND: Well, it turns out that the type of calculations that they do to make those beautiful visuals are very similar to the types of calculations we need to do to, for example, do climate simulation.
HENN: That's kind of the way conventional computer chips work: If you want to go faster, you have to step on the gas. Graphics chips are different. They work in parallel.
GUPTA: Once we get to the grocery store, my wife and I can simultaneously go through the aisles, picking up stuff.
HENN: That's how gaming chips go faster. Today, video games are a $30 billion industry, and it's pushing graphic chip designers to build faster and faster chips. Scientific supercomputers are hitching a ride, which is why Gupta says we should all just go ahead and let our kids play those video games.
GUPTA: Absolutely, except my kids.
HENN: Steve Henn, NPR News.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.