China eyes 'dual-use' supercomputers
By Matthew Luce
As of November 15, the world's fastest supercomputer officially belongs to
China, according to the most recent listing of the world's Top 500 computers.
China has pushed the United States out of the top spot as well as putting a
third machine into the top ten, providing another indicator of China's rise as
a world technological power. Yet this year's results should come as a surprise
to no one.
China has been pouring investments into high performance computing for the last
decade and gradually edging the United States out of the top ten. Increased
funding and an official policy commitment have propelled China from a
technological backwater
that in 2001 did not have a single machine in the Top 500 into a supercomputing
superpower.
Supercomputers, or high-performance computers, are an enabling technology that
opens up a wide range of research frontiers previously closed to Chinese
institutions, not least in defense applications. They are an increasingly
important tool in intelligence as well as weapons design, and a crucial link in
any national innovation chain.
China emphasizes the civilian applications of its supercomputers, but a quick
glance at China's history and research and development (R&D) architecture
would indicate that they will see significant military use. At the same time,
while China's triumph in supercomputing is a milestone, it should not be seen
as a signal that China has exceeded the innovative power of the West, but
rather as a launch pad for further technological development.
A dual-use technology
By devoting national R&D resources to developing domestic supercomputing
capabilities, the Chinese government is betting on a return on its investment
in the form of heightened R&D capabilities in a wide range of fields.
Heightened commercial and civilian research capabilities are certainly among
the payoffs of building such a powerful computer, but China's military will
certainly benefit greatly as well.
Supercomputers can be put to work on one complex problem or multiple
decentralized ones, but it stands to reason that they are usually employed for
issues that require the quadrillions of calculations per second that they are
capable of. The complex mathematical analysis involved in cryptanalysis and
sensor signal processing today are problems that can only be tackled
practically by computers with these "super" capabilities. China's efforts to
develop a secure satellite communications network as well as data fusion
systems for missile tracking are critically dependent on a capability to
process encrypted data at a very high rate of calculations per second.
While China's newest and fastest supercomputer is ostensibly for civilian
research, it is highly significant that it was built by the National University
of Defense Technology (NUDT), China's premier military technology university
and one of its top research centers. The bulk of the world's supercomputer
processing power and most likely China's as well is devoted to commercial and
academic research, but any modernizing military like China's also has an
increasing need for supercomputing capabilities.
In the 1990s China was accused of diverting supposedly civilian supercomputers
purchased from the United States for military ends, so it would not be the
first time China used the fig leaf of "civilian usage" to mask military
supercomputing programs. At NUDT, supercomputer development labs like the
National Key Lab for Parallel and Distributed Processing operate on the same
campus as the respective National Key Labs for C4ISR and Automatic Target
Recognition. These are the very same kinds of research facilities that would be
expected to have need for supercomputers to support their work.
Additionally, supercomputers provide indispensable services for a nation in the
process of modernizing its nuclear and conventional armament. Since all nuclear
test explosions are precluded under the Comprehensive Nuclear Test Ban Treaty
(CTBT), nuclear blast modeling can be performed only by large computers.
Missile, jet engine, and conventional explosive design and modeling are also
increasingly done using supercomputers. With a much improved supercomputer
arsenal, China also has an increased capability for the R&D necessary to
bring its armed forces into the 21st century.
China's possession of supercomputer technology may also constitute a
proliferation risk. As a result of their military applications and in
particular their cryptanalytic functions, much of the technology in a
supercomputer is defined as "dual-use" according to the Wassenaar Arrangement
on Export Controls.
Supercomputers can be compared to any dual-use technology like a rocket engine
or a nuclear reactor: they can be used for commercial civilian projects or they
can be used to produce weapons. Since the People's Republic of China is the
only major supercomputing power that is not a signatory of the Wassenaar
Arrangement, there is an elevated risk that supercomputers could be sold to
rogue states to assist their nuclear programs.
Popular misperceptions
China now has the most powerful computer in the world, but that does not mean
that their ability to innovate has eclipsed that of the United States, nor does
it mean that China has a military or intelligence collection advantage. The
ascent of the Chinese to the number one spot is based more upon political will
to invest in technology than any significant indigenous technological
breakthrough.
Building a more powerful computer is accomplished today by linking together an
incrementally larger cluster of processors and writing software that allows
them to operate together, rather than (in most cases) designing a new and
groundbreaking computing technology. The cost to build a top supercomputer has
also dropped precipitously in the past 10 years, meaning that the Chinese had
to invest a relatively smaller sum compared to the amount spent by US
institutions to develop top supercomputers just ten years ago.
This is not to say that China's newest and most powerful supercomputer, the
Tianhe-1A (which translates as "Milky Way") is not innovative. It combines its
CPUs (central processing units) with GPUs (graphics processing units) to
increase performance, consumes significantly less energy than its peers in the
United States, and possesses a Chinese-designed interconnect chip and software
that links the cluster together.
The system also contains 2048 Chinese-made Galaxy FT-1000 processors with an
undisclosed purpose, likely as a memory controller/synchronizer to boost the
speed of the system. The bulk of its hardware, however, is still designed by
Intel and NVIDIA, which are both based in California. This means that while
Chinese labs may be working hard to produce entirely indigenous supercomputer
systems, they currently still rely on foreign imports.
Just having a faster machine does not in itself provide any advantage for
Chinese researchers either. In some sense building the Tianhe-1A only provides
China with a showy muscle car: it might be able to go from 0 to 60 mph (0 to 97
km/h or 0 to 27 m/s) in 3 seconds, but that doesn't provide much added utility
if the owner is only using it to drive to work. The real test of innovative
ability will be in designing specific applications for the computer's power so
that it does not lie idle for most of the day. Building such a large cluster
will also be a waste if it is used for decentralized or cloud computing instead
of concentrating on a few very large and knotty problems, since smaller and
cheaper computer systems could be used for easier tasks.
China's supercomputer dominance cannot then be taken as a signal of US
technological inferiority. If US researchers and policymakers are to take any
lesson from the November 2010 Top 500 listing, it is that the playing field has
been leveled. By investing heavily in high performance computing and making it
a feature of the 11th 5-year Plan for Technology, Beijing has proved that it is
serious about its goals for high performance computing and is willing to devote
the necessary resources for research and development. China's supercomputing
research labs may not be superior to similar facilities in the United States,
but they currently enjoy generous funding and directives from central planners
to develop faster machines.
Now that China has demonstrated its commitment and ability to build faster
machines, it seems likely that the United States will respond to the challenge.
After Japan's surprise coup for the fastest computer, which lasted from
2002-2004, the US responded with an increase in research funding for
supercomputer projects and managed to push Japan entirely out of the top 10 by
2007.
Today US institutions have the capability to build faster machines, but
according to Jack Dongarra, the computer scientist at Oak Ridge Labs and the
University of Tennessee who oversees the judging of the Top 500, "it's a
question of will." According to an October 2010 report by the National Center
for Computational Sciences, two new supercomputers, each capable of more than
20 petaflops, are respectively under construction at Lawrence Livermore and Oak
Ridge National Labs, but the systems will not be operational until 2012 and few
details are available as to their systems.
Conclusions
This new wave of Chinese supercomputers has the potential to give Chinese
research institutions a leg up on the United States in terms of future defense
and commercial technological innovation, but the real test will be in software
and applications that may still be in development. A fast computer is a trove
of research potential, but if it lies untapped then the placement of a Chinese
machine into the number one spot will be nothing more than flag-waving.
Nevertheless, Chinese defense technology research labs now have the means,
motive, and opportunity to take advantage of high performance computing
resources. China's conventional and nuclear weapons design programs as well as
its intelligence and signal processing architecture have already progressed to
the stage where they can make efficient use of a growing supply of
supercomputers. Access to machines like the Tianhe-1A thus opens up new
horizons for Chinese defense researchers and cryptographers, and chips away at
the technological and military advantages of the United States.
Observers will have to come to terms with the fact that this is not a fluke or
a one-time effort to build a single machine and briefly upstage the Americans.
While American computing labs may be able to recover their lead in a few years,
this year's display of Chinese supercomputing power is only the latest
technology being churned out by Chinese research labs in a concerted push to
become a leading global innovator.
Head
Office: Unit B, 16/F, Li Dong Building, No. 9 Li Yuen Street East,
Central, Hong Kong Thailand Bureau:
11/13 Petchkasem Road, Hua Hin, Prachuab Kirikhan, Thailand 77110
