Woudhuysen



Chinese supercomputers

First published in China Outlook, November 2013
Associated Categories IT Tags:

If only briefly, China’s Milky Way 2 has taken the lead in the kind of machine that will be a vital to tomorrow’s heavy lifting in security, science and manufacturing

In July, China overtook the US in TOP500, a league table of the world’s fastest supercomputers, with its Tianhe-2 (Milky Way 2) machine, currently located near Changsha, Hunan province, at the National University of Defence Technology (NUDT). The news hardly hit the front pages. Yet on 5 November, president Xi Jinping visited the NUDT and lavished praise on it; and, later on that month, IBM said that it would open up Watson, its natural language supercomputer, to third-party software developers in fields such as online shopping and healthcare. It’s clear that interest in supercomputers, the heavy-lifting branch of IT, is hotting up. What’s more, China’s prowess in this fast improving market, though always subject to overtaking by other machines, was not achieved lightly.

Together with Inspur, a Chinese IT giant headquartered in Jinan, Shandong, the NUDT built Tianhe-2. It’s true that the machine uses the world’s largest number of Intel Xeon central processing units. Still, most of the rest of the package – software, front-end processors, operating system and interconnection network – is not just Chinese but also, in the words of TOP500 cofounder Jack Dongarra, ‘unique and interesting’, in part because of the density of components, and the apparent reliability and scalability of the system. Also, Tianhe-2 passed its tests nearly two years ahead of plan. From possessing no supercomputers in the TOP500 list in 2001, China is now within striking distance of building one that is home-grown in every aspect.

Like other supercomputers, Tianhe-2 is a room-sized collection of cabinets and blinking lights: right now, its 162 cabinets, each with 64 blades in them, are crammed into just 720 square metres, though it will be able to spread out more when it is moved to China’s National Supercomputer Center in Tianjin next month. Cooled by water, TH-2, as it is abbreviated, guzzles a total of no less than 24MW of power.

In its physical size and energy usage, TH-2 might seem to belong to an earlier era. Yet supercomputers will likely prove important in years to come. China’s efforts in the field have already led, in 2010, to American fears, not least because supercomputers are used to simulate the detonation of nuclear weapons, as well as in simulations of aeroplanes in wind tunnels. In the burgeoning domain of cyber-security and the trawling of masses of surveillance data, China may one day even prove a match for America’s National Security Agency (NSA).

In fact, however, strength in supercomputers will be vital to the advance of innovation in many other strategic fields, including quantum physics, the splitting of electrons, molecular modelling, fluid dynamics, materials, earth and atmospheric sciences, weather forecasting, modelling the energy use of buildings and the long-term fate of CO2 stored underground, the design of new cars and new drugs (especially those related to cancer and swine flu), and the detection of changes in the large data sets that surround payment cards. There is even speculation that supercomputers will one day be able to emulate the human brain, although at the moment it can take a supercomputer 40 minutes to mimic one second’s operation of one per cent of a brain. For its part, China has just this year published papers on using them around complex underground systems (oil, minerals, pipes, buildings and railway networks), as well as in the simulation of earthquakes, laser-propelled rockets, and particle beam accelerators.

Supercomputers have cut the time needed to juggle large data sets – for example, those used in as genome analysis – from months to seconds. They can model the noise of a jet engine before it is actually built. The basic unit of account with supercomputers is the number of floating-point operations – a calculation typically used in science – that can be executed in a second; and it is this speed that sets them apart from other computers. Here, some partly bowdlerised Greek, and a head for noughts, are relevant.

A typical processor in a laptop today can handle 10 billion flops per second, or 10 Gigaflops p/s. By contrast, a fairly mundane supercomputer can handle 100 Teraflops per second, where Tera means a factor of 1012. Top-of-the-range supercomputers currently manage more than a Petaflop p/s, where Peta means 1015. Right now the world’s total processing power in supercomputers is moving up another 1000-fold, toward an Exaflop p/s. The NSA is said to be planning a machine of that speed, India has a scheme to build one by 2017 for a cool $1bn, and China is reported to have its own answer ready in 2018. For the moment, though, Tianhe 2 can reach just 33.86 Petaflop p/s. Still, that’s nearly double the flop velocity of the machine at the US Department of Energy’s National Laboratories at Oak Ridge National Labs.

Storage is equally impressive. TH-2 can keep 1.4 Petabytes of information. Such a memory capacity is essential for the four main tasks that have been officially outlined for the machine: the simulation of fluid dynamics, and of turbulence in toroidal nuclear fusion reactors; the analysis of business opinion, and the secure operation of government cloud computing.

What is China’s next move in supercomputers? Despite it running the world’s fastest machine at the moment, the country still has a way to go before it can begin to emulate America’s aggregate clout in the field. At present, TOP500 suggests that the world’s league table for individual supercomputers looks roughly like this:

Country

Count

Share of world supercomputers

Petaflops p/s

Millions of processor cores

United States

252

50.4

107

9

China

66

13.2

47

4.8

Japan

30

6

20

1.4

United Kingdom

29

5.8

8

0.6

France

23

4.6

9

0.7

Germany

19

3.8

11

0.9

India

11

2.2

3

0.2

Canada

9

1.8

2

0.2

Russia

8

1.6

2

0.2

Sweden

7

1.4

1

0.1

Like its rivals, China will also have to deal with a number of controversies that surround supercomputers, including their demand for energy, the environmental impact of their use of semiconductors, and the degree to which their supplementing central processing units with graphics processing units is relevant beyond  playing…. computer games.

In the long term, computers that handle both 0s and 1s simultaneously, through what is known as quantum computing, will open up the ability to factor large numbers, conduct searches of unstructured information and perform cryptanalysis in ways that even supercomputers cannot. Such quantum computers do not rely on transistors, and are only in their infancy at present. However, when physicists at the University of Science and Technology of China, Hefei, Anhui province, managed to entangle eight photons in a ‘Schrodinger’s cat’ or ‘cluster’ state, they didn’t only beat their own world record for such an exercise (they’d imprisoned six photons back in 2007). Through their advance, they also showed a way in which a quantum computer could correct the errors to which it is typically prone.

Make no mistake, then. In advanced computers as in the sphere of world economy, China isn’t really on the threshold of overtaking the US. But it would be wrong to underrate its data-crunching capabilities.

Share Button

0 comments

Comments are closed.