Photo/Illutration From right: Ian Buck, vice president of Nvidia Corp.; Makoto Gonokami, president of the Riken research institute; Takashi Fuchigami, director-general of the Research Promotion Bureau at the technology ministry; and Vivek Mahajan, vice president of Fujitsu Ltd., announce the joint development of the next-generation supercomputer FugakuNext in Tokyo on Aug. 22. (Takahiro Takenouchi)

U.S. semiconductor giant Nvidia Corp. will participate in the development of Japan’s next flagship supercomputer, which is scheduled to be operational around 2030.

The announcement was made on Aug. 22 by the Riken scientific research institute whose Kobe campus in Hyogo Prefecture is home to Fugaku, the country's current leading supercomputer.

In a first for Japan, the plan is to incorporate high-performance graphics processing units (GPUs), designed particularly for generative artificial intelligence (AI), into this new supercomputer. These would be sourced from Nvidia, the world's leading chipmaker.

The next-generation system's provisional name is "FugakuNext" and its development will be a joint project between Riken, Fujitsu Ltd. and Nvidia. 

The apparent goal is for FugakuNext to achieve 100 times the practical, real-world performance of its predecessor.

Total development cost is expected to exceed 110 billion yen ($740 million). The system will be installed on Kobe's Port Island where the campus is located and is projected to be operational around 2030.

Japan has long focused on developing supercomputers specializing in high-speed calculations.

Since 2002, three consecutive generations of its flagship supercomputers—the Earth Simulator, the K computer and Fugaku—have achieved the world's top rank for computational speed.

However, the competition has continued to intensify. Fugaku, which has been operational since 2020, has been surpassed by the latest machines from the United States, Germany and Italy. It slipped to seventh place in the most recent rankings from June.

In response, the Japanese government has decided to incorporate GPUs into FugakuNext—a first for a Japanese flagship supercomputer. 

Adoption of GPUs is becoming widespread in the realm of supercomputers across various countries as combining them with CPUs (central processing units) used in conventional computers ramps up calculation speed.

Furthermore, GPUs also excel at processing the enormous calculations required to power generative AI systems such as ChatGPT. 

After a competitive public bidding process, Fujitsu will be responsible for developing the CPU in a continuation of its role from the Fugaku project. Nvidia, with its reported 80 percent global market share in the AI sector, has been tasked with handling the GPU.

With FugakuNext, the project intends to advance the integration of CPUs and GPUs to boost hardware performance to five times that of Fugaku.

The plan is to also increase software performance 20-fold through methods like algorithm optimization, thereby raising the total practical performance by a factor of 100.

Once operational, it is expected to lead to significant improvements in accurately predicting torrential rain and contribute to developments such as high-performance materials.

The project also aims to leverage this development to strengthen Japan's domestic semiconductor technology and expand business opportunities.

Ian Buck, Nvidia's vice president, said at an Aug. 22 news conference that the company is proud to be involved with FugakuNext, which the company is confident will produce scientific breakthroughs.

The research institute also addressed the fact that a foreign company will now be making a full-scale entry into the development of Japan’s next flagship supercomputer.

Satoshi Matsuoka, director of the Riken Center for Computational Science, said, “Partnering with the American company that makes the world's best GPUs is a major strategic move. It will also lead to the global adoption of Japanese technology, including Fujitsu's CPU, which is aiming to be the best in the world.”

Related News