By Don Clark
Intel Corp.'s dominant franchise in chips for the world's
largest computers faces new attacks this week, but the Silicon
Valley giant is responding with technology upgrades of its of its
own.
The latest push to loosen Intel's hold on machines called
supercomputers is being fueled by technology from ARM Holdings PLC,
whose licensees supply the vast majority of chips used in
smartphones and tablets. Applied Micro Circuits Corp., one of
several companies developing ARM chips for server systems,
announced plans Monday to market components for larger scientific
machines, too.
Applied Micro also has a new ally in Nvidia Corp., whose chips
render images for videogames and are also used to accelerate
computing work done on general-purpose processors. Nvidia is
announcing that its products, which are already used as
accelerators along with Intel-based supercomputers, are now being
used alongside Applied Micro chips in new supercomputers from at
least three vendors.
The announcements are pegged to a conference in Germany for
vendors and users of supercomputers, the room-sized machines that
have long been used for jobs like oil exploration, weather
forecasting and cracking codes. The unveilings are a sign that
companies bringing ARM technology to corporate data centers don't
plan to limit themselves to low-end applications where conserving
energy is the paramount consideration.
"Quite frankly, people have been putting ARM in a box of low
performance, while here you have Applied Micro and Nvidia coming
together for a high-performance product," said Patrick Moorhead, an
analyst at Moor Insights & Strategy.
Intel, though, is also raising the bar on performance. The
company has been selling an accelerator chip of its own to compete
with offerings from Nvidia and others, based on the x86 technology
used in both personal computers and servers. That chip, called Xeon
Phi, was designed to be used in supercomputers along with Intel's
standard Xeon server chips.
The company, at the International Supercomputing Conference in
Leipzig, is providing new details of a forthcoming version of the
Xeon Phi that can also work on its own to power an entire
supercomputer. The next version, which has been code-named Knight's
Landing, features more than 60 small calculating engines based on a
design called Silvermont. It is expected to be three times faster
than the prior version on many kinds of computing chores.
The next Xeon Phi also comes with new high-speed circuitry for
connecting chips together in a system. Most supercomputers use
connections based on either Ethernet or InfiniBand, two standard
technologies that are available from multiple vendors.
Intel's development of a custom-designed connection technology
is a new tack. But Charles Wuischpard, its vice president and
general manager for workstations and high-performance computing,
said its technical specifications will be shared so other companies
can adopt it if they choose.
The new Xeon Phi has already been picked by the National Energy
Research Scientific Computing Center, a unit of the U.S. Department
of Energy. It plans to install a Cray Inc. system in Berkeley,
Calif., that has more than 9,300 of the new Knight's Landing
processors.
Intel's current Xeon Phi is used in a system in China called
Tianhe-2 that has once again topped a twice-yearly ranking of the
500 most powerful systems in the world, which is being announced at
the Leipzig event. In all, Intel chips account for 85% of systems
on the list issued Monday.
But Paramesh Gopi, Applied Micro's chief executive, argues that
the arrival of the first supercomputers that use its ARM-based
chips is an important step in breaking Intel's hegemony in the
market--and a sign the technology is powerful enough to be used in
all kinds of servers.
"You've never seen ARM in a supercomputing context before," he
said. "This is a huge, groundbreaking step."
Write to Don Clark at don.clark@wsj.com
Subscribe to WSJ: http://online.wsj.com?mod=djnwires