CloudDatacentreInnovationWorkspace

Japan To Build World’s Fastest AI & Big Data Supercomputer

Matt Broersma is a long standing tech freelance, who has worked for Ziff-Davis, ZDnet and other leading publications

Follow on: Google +

The AI Bridging Cloud Infrastructure is to target machine learning applications and big data analysis for academia, enterprises and start-ups

Japan’s government is planning to invest 19.5 billion yen (£14m) to construct an artificial intelligence-focused supercomputer that would be the world’s fastest, and would power machine learning and big data analytics applications.

The country’s Ministry of Economy, Trade and Industry has begun procurement for the project, with bids ending on 8 December, according to Japan’s ‎National Institute of Advanced Industrial Science and Technology (AIST).

Japan

 

AI focus

The AI Bridging Cloud Infrastructure (ABCI) system is to be built by AIST and deployed at the Kashiwa Campus of the University of Tokyo.

The aim is for it to reach 130 quadrillion calculations per second, or 130 petaflops, which would surpass the 93 petaflops attained by China’s Sunway TaihuLight, currently the fastest system.

The system does not target floating-point operations per second (FLOPS) performance as it is measured using the Linpack benchmark employed by the Top 500 rankings, however, but rather uses a “reduced precision” arithmetic better suited to large-scale data analytics and machine learning tools, said professor Satoshi Matsuoka of the Tokyo Institute of Technology, who is responsible for the system’s name and design.

In a Twitter post he said it would be more accurate to say the machine targets 130 “AI-flops”.

 

Power efficiency

The system is intended to run with a high degree of power efficiency, consuming under 3 megawatts and achieving a 1.1 PUE rating, according to Matsuoka and to AIST’s design document (PDF).

Japan’s Oakforest-PACS, the country’s current fastest supercomputer, consumes the same power to deliver 13.6 petaflops, and Sunway TaihuLight consumes more than 15 MW.

The machine learning applications planned for the system can be used to help automate complex engineering tasks and develop driverless vehicles, amongst other applications.

AIST said it wants to use the system to improve collaboration between industry and academia, and to aid the development of start-ups. It said it wants the system to be completed in 2017.

The ABCI project is part of a wider effort by the Japanese government to boost the country’s competitiveness in areas such as robotics, batteries and renewable energy.

The two top-ranked supercomputers are currently based in China, and the country has the same number of machines in the top 500 list as the US, reflecting its position as the world’s second-largest economy, a position it took over from Japan, which now ranks third.

Corporate applications

Fujitsu, which constructed Oakforest-PACS, declined to say if it would bid for ABCI, but has said it is interested in supercomputer projects.

The US Department of Energy (DOE) has said it plans to bring a supercomputer online in 2018 that will have double the computing power of Sunway Taihulight, or about 186 petaflops.

The system, called Summit, was announced in 2014 when the DOE awarded $325 million (£247m) to Oak Ridge National Laboratory in Tennessee to construct it.

What do you know about China And IT?Try our quiz!