Tokyo — The creators of the world’s fastest supercomputer want to make one thing perfectly clear: They were not trying to be number one.
“Of course I was happy and relieved,” Satoshi Matsuoka told CBS News about the moment he learned his team’s “Fugaku” supercomputer had taken the crown for the fastest calculating speed on earth. But for Matsuoka and his team, the honor merely vindicated a strategy of focusing on practical applications.
Fugaku (pronounced foo-gah-koo and named after Japan’s tallest mountain, Fuji) culminates a $1 billion, decade-long mission by several thousand developers from the government-run RIKEN Center for Computational Science and computer maker Fujitsu.
The machine scored top marks in four supercomputing benchmarks, including processing speed, artificial intelligence (AI) applications and big data analytics, earning it the number-one spot on the Top500 list of global supercomputers this week.
User-friendliness was the mantra for Fugaku’s developers. For all its world-beating speed — 415 quadrillion computations per second — Fugaku was designed to run ordinary applications like Word or Powerpoint, making it widely accessible to researchers from sectors as diverse as medicine, automotive design and civil engineering.
Rather than create something like, say, a Formula One car (fast, but not particularly useful), Matsuoka said his team “wanted something fast, but that your grandmother could drive — to the supermarket!”
Fugaku runs 2.8 times faster than IBM’s Summit supercomputer, the previous record-holder that runs at the U.S. Oak Ridge National Laboratory, and it marks the first such victory for Japan in nine years in a supercomputing field that has come to be dominated by the U.S. and China.
With processing units occupying a room that could comfortably fit two high school gyms at Riken’s offices in the western city of Kobe, Fugaku is akin to, “a very beefed-up version of the ordinary smartphone.” It has the computing brawn, however, of 20 million smartphones.
Although it won’t be fully installed until next year, Fugaku has already started to earn its keep, churning out simulations demonstrating the frightening ease with whichspreads between human hosts on trains and in other indoor settings.
The videos, shown widely on TV here, underscored the risks of congregating in poorly ventilated and crowded spaces, how droplets are emitted from the edges of face masks, and what happens when a hospitalized patient coughs. Despite their brevity, the simulations were deceptively complex to create, Matsuoka said, incorporating vast amounts of data about air flow and how virus-laden droplets travel through air.
“If you have a very dry environment, these droplets evaporate and turn into aerosols and thus carry much farther in distance,” he said. “You really need a ventilation apparatus to get rid of these aerosols as quickly as possible.” In humid spaces, he said, drops don’t aerosolize, but tend to fall on desks, which need to be disinfected more diligently.
Fugaku, which can crunch through three months’ worth of typical computing before lunchtime, is also combing through a database of 2,000 drugs to ferret out which formulations might be used to mitigate the symptoms of COVID-19. The search has already turned up some promising potential treatments that haven’t yet been considered by medical scientists, Matsuoka said.
The team is also working on a molecular simulation of how spiky coronaviruses attach to human cells — a valuable tool for diagnostics and drug discovery.
“This is not possible without supercomputers,” Matsuoka said. “There is just no way you can take movies of how a virus attaches to a human cell.”
Winning is nice, “useful” is key
Besides researching bio-medicine, pre-testing cars for collision safety, disaster-proofing design for buildings, designing the automobile engines of tomorrow and developing entirely new materials, Fugaku is also being deployed for weather and climate forecasting, especially disaster warnings.
That work is a particular priority for Japan, one of the world’s most vulnerable countries to earthquakes, typhoons, tsunamis and flooding.
“Sudden torrential rainfalls are becoming more common in Japan and the southern U.S. because of global warming, but these are very localized phenomena,” Matsuoka said. Among the concepts Fugaku is studying are highly targeted emergency forecasts, which could give residents a 30-minute warning on their smartphones when danger is approaching.
Japan’s high-performance computer scientists faced scrutiny in 2009, when an earlier supercomputer project known as “K” fell into the crosshairs of a crusading Democratic Party of Japan lawmaker. She demanded to know why taxpayers should be on the hook for what seemed, to the uninitiated, like a budget-busting national vanity project.
“Is there any reason why Japan needs to be No. 1?” she declaimed, playing to the TV audience. “Is No. 2 unacceptable?” The science establishment learned its lesson.
“Absolutely we cared about” the disastrous attempt to justify Japan’s supercomputer mission, Matsuoka said, calling the hearing “a fiasco.”
Bureaucrats were “never able to explain why it was important. So we set out to build a machine that would excel in applications and be useful.”
Being named number one, he said, was just icing on the cake.