Categories
Tech

The 416 quadrillion reasons why Japan’s supercomputer is number 1

Local Fox news stations had this English captioned version of a very popular, supercomputer-generated video from Japan.

For a good part of last week, the unrivaled highlight of Japanese television—looped on news programs—was a short computer simulation of a salaryman sitting at his desk and coughing. Its primetime primacy is total: this cough is the Killing Eve of animated expectoration.

The imagined office in the video, like tens of thousands of its real-life equivalents around Japan, appears carefully prepared for the great return to work and new normal of life under COVID-19. The desks are separated by decent physical space and sensible plastic partitions cleave the landscape.

But it is all in vain. As the simulation shows in terrifying particulate detail, while most of the army of droplets released from the cougher’s mouth are blocked by the screens, a crack platoon makes it over the partition, delivering its deadly payload into the neighbouring workspace.

The reason this simulation is so compelling—and why Japan is so enchanted with it—is how it was produced: on a $1 billion-plus made-in-Japan machine called Fugaku, whose brisk operating speed of 416 quadrillion calculations per second officially makes it the world’s fastest supercomputer. Fugaku, jointly developed between the Riken institute and Fujitsu, has the same energy demand as a small city, but—at a time when analyzing such things has never been so important—it allows us to see why, with agonizing molecular accuracy, this health crisis is proving such a tough nut.

Fugaku’s processing pace, clocked at 2.8 times that of the US-built Summit machine it has now unseated, also puts a Japanese supercomputer back in the top slot of the world rankings for the first time since 2011. COVID-19 means Japan can’t host the Olympics this year, but this is a fine lockdown consolation prize.

Leaving aside my fascination with supercomputers, it is important to acknowledge the significance of this achievement and why, for subtle reasons, Japan’s incumbency of the top position (however brief it may be) stands apart from its predecessors. For 27 years, since a team of German and US scientists first began to quantify and rank the power of rival machines around the world, the battle for supercomputer supremacy has had the distinct flavor of a space or arms race—a forum for national muscle-flexing that reflects both ambition and the relative economic and industrial powers of the contestants. In this context, the steady rise of China to numerical dominance of the TOP500 list has felt inevitable.

China, the US, and many others take this extremely seriously, not least because several of the biggest supercomputers are unabashedly constructed for use by defense industries. And the TOP500, notes Fugaku’s architect-in-chief Satoshi Matsuoka, lists only the publicly disclosed supercomputers—more are lurking secretly in the private sector. But for Japan, the competition has always felt even more personal: a live index, in many ways, of its undulating global prowess and relevance.

The strength of Japan’s feelings on this—and in particular on the country’s nine-year absence from the No 1 position—are well known. In 2009, when the same Riken institute was working on Fugaku’s predecessor, K, the government of the time was on a post-financial crisis cost-saving drive. The political rising star, Renho Murata, chose to make the huge public funding of the supercomputer her target, judging—entirely wrongly—that the public would be on her side. “Isn’t it good enough to be number two?” she notoriously asked, drawing fierce and prolonged condemnation from all quarters.

But, in private, there was an acknowledgment that she had a point. The hype that preceded K, and the patriotic posturing around it, was much more about its intended occupation of the supercomputing rankings top slot than it was about what it would ultimately do. When it did eventually emerge in 2011, it was so loaded towards power over user friendliness that it never achieved its potential.

Nine years later, Riken and Fujitsu have clearly learned from that: the triumphalism around Fugaku’s launch is now explicitly about the earthquakes, weather patterns, and coughs that it will parse with its great processing might. K was explicitly designed to be number one, said Matsuoka, but Fugaku was not. It was designed to be both user friendly and good at the range of applications it will be called upon to run: it just happens that the process created a giant. Corporate Japan, whose instincts are often far more K than Fugaku, would be wise to take note.

Leo Lewis is the FT’s Tokyo correspondent

© 2020 The Financial Times Ltd. All rights reserved. Not to be redistributed, copied, or modified in any way.

Listing image by Riken Center

Credit: Source link

Leave a Reply

Your email address will not be published. Required fields are marked *