World's fastest computer, Tianhe-2, might get very little use

The Tianhe-2 has been hailed as an example of China's tech muscle, but some experts say it may be too powerful for most tasks

PUBLISHED : Thursday, 20 June, 2013, 12:00am
UPDATED : Thursday, 20 June, 2013, 5:07am

At the Daya Bay nuclear power plant in Shenzhen, hopes that a neutrino experiment can lead to a Nobel Prize in physics are buried in a mountain of data.

Almost every day, an antineutrino detector at the reactor generates very large volumes of raw data that need to be kept, processed and analysed for traces of evidence that could alter the direction of modern physics.

The job can only be done with a computer, and Tianhe-2 in nearby Guangzhou, recently crowned the world's fastest supercomputer, would, at first glance, seem perfect for the job.

But Dr Cao Jun, a team leader at Daya Bay, said they had no interest in Tianhe-2 for a number of reasons: a huge amount of data could choke the computational monster; data flow from one calculating unit to another could be slow; and a slower machine at the Chinese Academy of Sciences' Institute of High Energy Physics could get the same job done more quickly and cheaply.

"Tianhe-2 cannot give us what we need," he said. "It is not quite suitable for high energy physics. I am not sure what it can do."

Unlike home computers that can handle various tasks, ranging from word processing to gaming and web browsing, supercomputers are built for very specific purposes. To exploit their full computational capabilities, researchers have to spend months, if not years, writing or rewriting software codes to train the machine to do a job efficiently.

For instance, dethroned former champion Titan, built by Cray at the Oak Ridge National Laboratory in Tennessee, has served in only six, quite similar major projects dealing with molecular physics, such as simulating engine combustion to improve fuel efficiency and modelling the movement of air molecules for climate change estimates.

However, most supercomputers built on the mainland in recent years have been given long to-do lists. Tianhe-2 is expected to control traffic lights, predict earthquakes, develop new drugs, design cars and create movie special effects, among a dozen other things, Xinhua reported.

Computer scientists say such a multipurpose computer is not especially good at anything and, like most expensive supercomputers in big cities, is likely to be idle for long periods.

They said Tianhe-2, to be officially launched at the end of this year, would not have much chance to run at full working capacity.

Using Intel Ivy Bridge and Xeon Phi chips with a total of 3.12 million processor cores, Tianhe-2 can perform 33,860 trillion calculations a second, nearly double the speed of Titan and leaving the K computer, the fastest machine in Japan, in the dust.

Tianhe-2 was built by the National University of Defence Technology and is being run by Guangzhou's city government, which met most of its 2.4 billion yuan (HK$3 billion) cost.

Li Nan, its deputy chief designer, told China National Radio it showcased China's hi-tech muscle.

"The birth of Tianhe-2 is a proof that China can lead the world in scientific innovation," he said. "The superb performance of Tianhe-2 has attracted government and enterprise users from around the world."

But Li's claim has been challenged by many scientists and industry experts. Less than six months before Tianhe-2's official launch, they said, it was still unclear who would be its first batch of users.

Officials at the supercomputer centre in Guangzhou declined an interview request, but according to state media reports Tianhe-2's top priority will be to improve China's car industry, in particular Guangzhou Automobile Group.

Professor Bian Mingyuan, from the department of automobile engineering at Tsinghua University, said that the biggest headache for China's car industry was not a lack of computational power but a lack of craftsmanship and ideas.

"I have never heard of Toyota or Daimler or any major carmaker using a supercomputer to design their cars," he said. "It is like running after a chicken with an axe. It is quite unnecessary."

Computers were playing an increasingly important role in car design, such as simulating crash tests, Bian said, but the computing power needed could easily be provided by a commercial industrial design computer costing just a small fraction of Tianhe-2's price, and the software used on commercial computers might not run on the supercomputer.

Even in the study of climate change, where demand for supercomputers is huge, Tianhe-2's role remains uncertain.

Professor Lin Jintai, a climatologist at Tsinghua University's school of physics, said he was interested in Tianhe-2 but after using several supercomputers for his research in the United States he was concerned about many problems and might be forced to use smaller but more specialised systems.

For instance, his team was unlikely to have full access to Tianhe-2's computational resources, and whether it could secure enough processing cores for the job would depend on lengthy negotiations with the computing centre.

Like Cao, Lin was also concerned about Tianhe-2's ability to handle huge amounts of data, because climate change study often requires simulation of the global atmospheric system.

"Supercomputers are only good for certain kinds of jobs," he said. "For other jobs, the desktop computer in my office could be more efficient."

A senior scientist at the Beijing Computing Centre said China had much more computing power than it actually needed, and the excess power was not only a waste of taxpayers' money but would also become obsolete in a few years due to the rapid upgrading of CPUs and information technology.

"Supercomputers have become a face project in many cities," he said. "Many government officials believed that their city would become hi-tech overnight with a supercomputer.

"I have visited many computing centres. In some places dust has built up on the computer because the entire facility had been shut down for months without a single user. Even in big cities such as Beijing and Shanghai only a small fraction of the total capacity was used.

"The supercomputer bubble is worse than a real estate bubble. A building will stand for decades after it is built, but a computer, no matter how fast it is today, will become garbage in five years."