TOKYO Fujitsu has decided to team up with U.K. chip-design company ARM Holdings to develop the successor to Japan's K supercomputer, signaling a departure from its emphasis on employing proprietary technologies.
The next-generation machine is being developed by the Japanese company and the government-backed Riken research institute, with the help of 130 billion yen ($1.27 billion) in funding from Tokyo. Completion is scheduled for 2020.
Under the partnership -- unveiled recently at a supercomputing conference in Frankfurt, Germany -- Fujitsu will use central processing units designed by ARM, whose products are increasingly becoming a de facto global standard.
Supercomputer technology has long been a matter of national pride for Japan, which may help explain Fujitsu's determination to design the K's CPU on its own. Hopes are high that the machine will support efforts in weather simulation, drug discovery, artificial intelligence development and other fields. But only software developed by Fujitsu or other K computer users can be run on the machine, limiting commercial applications.
Adopting ARM's processor designs would make the K readily compatible with software developed by companies and researchers worldwide. Still, Fujitsu insists that the way in which the circuit is drawn will retain the Japanese company's originality.
How efficiently computational labor is divided among a supercomputer's several hundred-thousand to several million processors is what sets the machine apart. Fujitsu has strength in communication speeds among various processing units, and this will let it "remain sufficiently competitive," said Naoki Shinjo of the company's technical computing division.
Fujitsu's strategy shift is likely aimed at avoiding the kind of overspecialization that limits the market and results in a lack of compatibility for broader applications. But unless the company can continue to innovate, it could begin to lose relevance since ARM's processor designs can be used by any company willing to pay.
CHINA'S ASCENT Competition in the supercomputing arena is more heated than ever, especially with China's rise as a major player. On June 20, China's Sunway TaihuLight supercomputer topped a ranking of the world's fastest machines, making it the first-ever Chinese homegrown computer to grab the No. 1 spot. The TaihuLight clocked in at 93 petaflops (1 petaflop equals 1 quadrillion calculations per second), making it roughly three times as fast as the second-ranked Tianhe-2, also China's, and 10 times faster than the K, which came in at No. 5. American machines took the Nos. 3 and 4 slots in the ranking, which was carried out by a team of specialists from the U.S. and elsewhere.
The TaihuLight's CPU is the culmination of China's efforts since the turn of the millennium to transform expert knowledge accumulated in the U.S. and elsewhere into polished, homegrown technology. The country's charge for the No. 1 position was in part a bid to stoke national pride -- one thought to involve massive research funding. Other machines using similar homegrown technology are expected to hit the market soon.
China "has made great strides in both hardware and software, taking its place alongside the U.S. and Japan," said Satoshi Matsuoka of the Tokyo Institute of Technology. "There's a fairly good chance of them exporting the technology in the future."
Advances in supercomputing have a large impact on a country's scientific and technological progress, corporate competitiveness and military prowess. The U.S. is slated to put three machines capable of rivaling the TaihuLight into operation as soon as 2018. But China reportedly plans to unveil a computer 10 times faster as early as 2020. The country has clearly achieved superiority over the U.S. in the field, according to Jack Dongarra, a distinguished professor at the University of Tennessee.
Japanese companies are no strangers to the pain that can result when a universal standard sweeps across an industry. U.S. companies' dominance of the markets for PC operating systems and CPUs, for example, made it difficult for Japanese players to stand out. Manufacturers in emerging nations surged into the hardware field as standard technology spread, pushing Japanese players to the sidelines.
Whether Japan's supercomputer industry can avoid the fate of the country's PC industry depends on companies' ability to continue developing unique technology on top of a set of universal parts. If they are not up to the task, Europe and the U.S. could once again dominate the design field, with China taking charge of manufacturing, leaving little room for anyone else.
