March 18, 2008, 4:03 PM — Imagine a man you know but whose name you can't remember approaches you, and
your mobile phone uses face-recognition capability to give you his name and
information about him before he says hello. This is the kind of application
that researchers hope will be developed from US$20 million Microsoft
and Intel are giving two
U.S. universities for research on parallel computing.
The companies are donating the money to Universal Parallel Computing Research
Centers (UPCRCs) at the University of California Berkeley and the University
of Illinois at Urbana-Champaign, they announced at a news conference on Tuesday.
The centers are aimed at tackling the challenges of programming for processors
that have more than one core and so can carry out more than one set of program
instructions at a time, a scenario known as parallel computing.
In addition to the $20 million, the University of Illinois will provide $8
million to fund its center, and UC Berkeley has applied for $7 million in grants
for its research.
UC Berkeley quietly opened its Parallel Computing Lab in January, according
to a UC Berkeley
Web site. The lab was born out of research done there and published in a
white paper by researchers at Berkeley's Electrical Engineering and Computer
Sciences department in 2006.
In the paper, they said the current evolution of programming models from single-core
to the dual-core and quad-core processors available today from Intel and AMD
won't work for a future where processors could have as many as 16, 32 or hundreds
of processors. They set out to find a better way to develop programming models
to meet the challenges of multi-core chips.
UC Berkeley's David Patterson, a professor of computer science and director
of the UPCRC, described the problem as one of designing programs to take advantage
of parallel computing's ability to divvy up workloads across different processors.
On Tuesday's conference call, he compared the scenario to dividing the work
of writing one story between 16, or even hundreds, of reporters. While the work
could potentially be done 16 -- or even hundreds of times -- faster, "we
won't get to deliver on that performance without balancing the work well,"
he said.
Microsoft's and Intel's interest in parallel computing is not merely altruistic
-- both companies already are doing their own research so they can take advantage
of the computing power that comes with multi-core technology, and thus gain
a competitive advantage in their respective software and processor markets.
The agendas of the research centers will align closely with Intel’s Tera-scale
Computing Research Program and Microsoft’s Technical Computing Initiative,
the companies said.
As for some of the real-world applications of parallel computing, Patterson
and Marc Snir, professor of computer science at the University of Illinois,
said if researchers can use programming to harness the capabilities of multi-core
machines, it will give mobile devices the computing performance that today comes
only from supercomputers.
Patterson described the scenario in which a mobile phone might use face-recognition
technology to save someone -- he used himself as an example -- from an embarrassing
situation of not knowing a person's name. "I'd personally be excited to
buy a cell phone that has that technology," because this is a situation
he often encounters as a university professor, Patterson said.
Fourteen members from the UC Berkeley faculty, as well as 50 doctoral students
and postdoctoral researchers, will staff the UPCRC, while the center at the
University of Illinois will be led by Snir and Wen-Mei Hwu, professor of electrical
and computer engineering. Twenty additional faculty members and 26 graduate
students and researchers also will participate in research at the Illinois center.
Both centers will make software available to the technology community for additional
development.
While there are only dual-core and quad-core processors available today, Intel
plans to release a six-core processor, code-named Dunnington, in the second
half of this year, and an eight-core processor, called Nehalem, at some point
in the future.
