China Is Getting Secretive About Its Supercomputers

The withdrawal marked the end of an era and created a divide that Western scientists say will slow the development of AI and other technologies as countries pursue separate projects.

The new secrecy also makes it harder for the U.S. government to answer a question it deems essential to national security: Does the U.S. or China have faster supercomputers? Some academics have taken it upon themselves to hunt for clues about China’s supercomputing progress, scrutinizing research papers and cornering Chinese peers at conferences.

Supercomputers have become central to the U.S.-China technological Cold War because the country with the faster supercomputers can also hold an advantage in developing nuclear weapons and other military technology.

“If the other guy can use a supercomputer to simulate and develop a fighter jet or weapon 20% or even 1% better than yours in terms of range, speed and accuracy, it’s going to target you first, and then it’s checkmate,” said Jimmy Goodrich, a senior adviser for technology analysis at Rand Corp., a think tank.

The forum that China recently stopped participating in is called the Top500, which ranks the world’s 500 fastest supercomputers. While the latest ranking, released in June, says the world’s three fastest computers are in the U.S., the reality is probably different.

“The Chinese have machines that are faster,” said Top500 co-founder Jack Dongarra. “They just haven’t submitted the results.”

Today’s fastest supercomputers are powered by tens of thousands of cutting-edge computer chips. A 2015 U.S. move curtailed Chinese supercomputer developers’ access to Intel chips and other U.S. hardware, followed by broader export restrictions four years later under the Trump administration. The Biden administration has tightened them further.

Dongarra and analysts who study China said they believed Beijing was worried the U.S. might do even more if China bragged about its supercomputing abilities.

They said it would be difficult for China to maintain its lead in supercomputing without leading-edge chips, many of them made by Silicon Valley leader Nvidia. Without those chips, China would have to use a brute-force workaround by stringing together hundreds of thousands of older-generation chips that gobble power.

In the age of artificial intelligence, limited access to high-end chips would force China to pick and choose what its supercomputers focus on, Goodrich said.

The rise of Chinese supercomputers

Supercomputing dates back to the 1960s, when U.S. government agencies started to design machines for juggling huge amounts of data simultaneously to solve problems in a way that less powerful computers working separately couldn’t match easily. The purposes were similar to today: simulating a nuclear-weapon detonation, modeling the climate and solving other big scientific problems.

The Top500 was born in 1993 when Dongarra, a University of Tennessee professor, and German colleagues distributed a math problem for supercomputers, and then ranked the machines by how long they took to solve it.

Participating in the list has always been voluntary. Dongarra estimates there are about 50 supercomputers, including those owned by intelligence agencies or private companies, that would make the Top500 if their owners submitted data.

For more than two decades, U.S. machines led the ranking, issued twice a year. But by November 2017, China had 202 machines on the list, compared with 143 for the U.S. “China dominated,” crowed Beijing’s state news agency at the time.

In 2019, the U.S. Commerce Department put five Chinese supercomputing organizations on a blacklist, saying they used supercomputers for military and nuclear purposes. The sanctions banned U.S. companies from selling components to those organizations without a license.

“It was a major turning point,” said Dongarra. Participation in the list dwindled. When he asked Chinese colleagues why, they said they weren’t allowed to submit information, Dongarra recalled. Chinese scientists also reduced how much data they shared in other scientific forums.

Chinese government agencies have led supercomputer development, with relatively few commercially operated machines. These government bodies have highlighted how the latest Chinese models use domestically produced processors, now that those from Intel and other U.S. companies are less accessible. 

Chinese government and science officials didn’t respond to requests for comment. A spokeswoman for the U.S. Commerce Department, which implements the supercomputing export controls, referred to previous statements by the agency that said the controls address national-security threats posed by China.

A new frontier

Officially, the fastest computer on the Top500 sits at the Energy Department-sponsored Oak Ridge National Laboratory, in Tennessee. Called Frontier, it is about the size of two tennis courts, cost $600 million to construct and has an electricity bill of about $20 million a year, said Dongarra, who also works at Oak Ridge. It uses tens of thousands of computer chips.

Dongarra doesn’t think Frontier is actually the world’s fastest supercomputer. Scientific papers suggest that certain Chinese machines are better. One has been referred to in state media as a prototype Tianhe-3, after a Chinese term for the Milky Way galaxy, while the other is a model in the Sunway series of supercomputers.

The Chinese papers describe the processors in the machines, what applications they run and the results they get, giving a good estimate of how fast the supercomputers are, Dongarra said.

A scientific paper submitted last year for the Gordon Bell Prize—essentially the Oscars of supercomputing—described the Sunway supercomputer as having 39 million cores, or parts of the chips that carry out processing. That is quadruple the number of cores Frontier has. Combined with other clues, that statistic indicates that the Sunway machine may be more powerful than Frontier.

China produces its own “best of” list for high-performance computing, which it calls the HPC Top100. Dongarra said he believed it left out leading Chinese supercomputers.

The No. 1 machine on the latest Top100 and some others are described in only generic terms without a name or operating institution given. Last December, a month after the latest list was released, the National Supercomputing Center in Guangzhou introduced a machine called Tianhe Xingyi, saying it achieved a manifold increase in performance over an earlier model in the Milky Way series called Tianhe-2.

Besides reading the papers, Dongarra tries to confirm the specifications in a more old-fashioned way: by attending presentations by Chinese scientists at conferences and peppering them with questions afterward.

“That’s sort of my verification: When I talk to them, it’s clear that they have such a machine,” he said, referring to the computers he believes are actually the world’s fastest.

Write to Stu Woo at Stu.Woo@wsj.com