By the end of the decade, exaFLOP computers are predicted to go online heralding a new chapter in scientific discovery.
The United States, China,
Japan, the European Union and Russia are all investing millions of
dollars in supercomputer research. In February, the EU announced it was
doubling investment in research to €1.2 billion ($1.6 billion).
What is an exaFLOP?
Computer scientists
measure a supercomputer's performance in FLOPS, an acronym for FLoating
Point Operations per Second, while "exa" is a metric prefix which stands
for quintillion (or a billion billion). An exascale computer could
perform approximately as many operations per second as 50 million
laptops.
"It is the next frontier for high-performance computing," says Dimitrios Nikolopoulos, professor at the School of Electronics at the UK's Queen's University of Belfast.
How fast are today's supercomputers?
Today, the fastest
supercomputers operate at the petaFLOP level says Nikolopoulos,
performing in excess of one quadrillion (or a million billion)
operations per second.
The first computer to
break through the petaFLOP barrier was IBM's Roadrunner in 2008. But its
reign as the fastest computer in the world didn't last long, with the
Cray Jaguar installed at Oak Ridge National Laboratory in the United States becoming the quickest with a performance of 1.75 petaFLOPS in 2009.
Today, the crown is held by is Japan's K computer developed by RIKEN and Fujitsu, according to TOP500 -- a project that tracks trends in high-performance computing.
The machine, installed at the RIKEN Advanced Institute for Computational Science, in Kobe, Japan, currently operates at over 10 petaFLOPS. It is more than four times faster than its nearest rival, China's NUDT YH MPP computer (2.57 petaflops).
How big are they?
"The kind of space that
you need is similar to that of a football field. You're talking about
many, many lanes of computer racks and thousands of processors," says
Nikolopoulos.
The K computer contains a mind-boggling 88,128 computer processors and is made up of 864 refrigerator-sized cabinets.
Physically, exascale
computing won't get any bigger, says Nikolopoulos, and might even get a
little smaller. But the amount of processors will rise substantially to
anywhere between one million and 100 million.
What are the challenges of reaching exascale?
Nikolopoulos says
"severe technology barriers" remain, the most important being power.
"Power consumption of supercomputers in general is not sustainable," he
says.
"The current projections
suggest that power consumption of exascale computers will be 100
megawatts. It's impossible to build a suitable facility and have enough
power."
Historically, a
computer's processor has used the most power (around 40-50% of the
total) Nikolopoulos says, but memory is rapidly catching it up.
"Changing materials and also the architecture of processors and memories is critical to exascale's success," he says.
"We are beginning to
understand the challenges of exascale in terms of hardware, software and
applications. We are at the stage where we can make mental projections
and set up directions for research."
What benefits could exascale computing bring?
It will enable discovery in many areas of science, says Nikolopoulos. "Aerospace engineering, astrophysics, biology, climate modeling and national security all have applications with extreme computing requirements," Nikolopoulos said.
Bill Cabbage, public
information officer at Oak Ridge National Laboratory, says exascale will
attempt to tackle very serious challenges in energy supply and
sustainability.
"These are very
difficult problems and will require the development of new
forward-thinking technologies to deal with them," Cabbage said.
"We are bringing all our resources to bear on these problems," he added.
Social sciences could also profit, says Nikolopoulos.
"More and more people
are interested in understanding the behaviors of societies as a whole.
These require simulations -- how people interact, communicate, how they
move. That will require exascale computing," he said.
Source: CNN News
0 comments:
Post a Comment