Intel, Microsoft Collaborate with Universities on Parallel Computing Initiative
For Microsoft and Intel, the computing universe is going parallel. Representatives from the companies met with reporters Tuesday morning to talk about a new parallel computing initiative involving the launch of two multi-million-dollar Universal Parallel Computing Research Centers--one at University of California, Berkeley and one at University of Illinois at Urbana-Champaign, which were selected from 25 universities evaluated for the initiative. The companies have committed $20 million over the next five years to fund the centers, which will focus on applications, architecture, and operating systems.
In addition to funding from Microsoft and Intel, IUUC will contribute $8 million toward the research center on its campus, and UC Berkeley has applied for $7 million from a state-supported program to match industry grants.
At UC Berkeley, the UPCRC, headed up by David Patterson, professor of computer science, will be staffed by 14 faculty members and will include 50 doctoral and post-doctoral researchers. At UIUC, the UPCRC will be headed up by Marc Snir, professor of computer science, and Wen-Mei Hwu, professor of electrical and computer engineering. Its team will include 20 faculty members and 26 graduate students and researchers.
"This is the first joint industry-university research collaboration in parallel computing of this magnitude," said Tony Hey, corporate vice president of External Research at Microsoft Research, "and it's really focused on achieving the long-term [advances] that are needed to make parallel computing accessible to mainstream developers."
Foundations for Future Computing
This goal of making parallel programming more accessible to the wider programming population, according to those involved with the initiative, is going to be critical in a world in which advances in computing performance will be coming by way of increases in the number of processing cores built into generic CPUs, which could conceivably double every couple of years--with no upper limit.
"We want to democratize parallelism," said UIUC's Snir in a telephone conference. "We want each programmer to be a parallel programmer. We want parallel programming to be synonymous with programming really. And, in fact, parallel programming can be simple. We do believe that with the right technology, we can make most parallel programming easy. We believe we have a very promising roadmap for these technologies and how to make the parallel programmers' lives simple."
The move represents not just another parallel computing initiative, according Berkeley's Patterson; it's setting the foundation for the future of computing in which the drive for higher-performing systems turns to energy-efficient, expanded multi-core architectures and away from the seeming dead end of single-core clock speed increases.
"No one's building the 15 GHz processor right now that can compete with a parallel processor," he said. "So programmers have no choice. If they want to write fast programs, they're going to have to write parallel programs. That's never been an obstacle. That's always been what I call the lazy-boy approach to programming: 'Why rewrite my program to be parallel? I'll just sit here and wait for Intel ... to make my programs faster without me doing anything.' Well, that era's over. So programmers are going to have to do something new if they want to run faster. That's one of the reasons I think why this time we have a better shot at revolutionizing programming around parallelism."
He added: "If we do this, I think we'll be able to set the foundation of computing for the next 30 or 40 years. For me--and I think for all the researchers--I think this is a once-in-a-career opportunity to set the foundation that people will be able to take advantage of for decades. If we solve this, the future of performance is rosy. And if we don't, it's not rosy."
A 'Wake Up Call'
Microsoft's Hey acknowledged that the age of parallel computing is upon us and said that this is a wake-up call for the industry and for academia.
"We're aware that the sea change is going to be on us," he said, "and we have been pursuing--within Microsoft Research, within other parts of the company--research into parallel programming. We have been looking into that. But we do think there needs to be a wake-up call to both the funding agencies and to academia that this really is an important for the computer science community and the IT industry to focus on. And that's why we think it's a very significant time to invest in two major centers at Berkeley and Illinois."
"This really is a change in how the whole industry is going to scale and deliver performance going forward, and that change represents ... a challenge, but it also represents an opportunity," said Intel's Andrew Chien, vice president, Corporate Technology Group and director, Intel Research. "The companies that are successfully advocating this transition to parallelism will probably be the leaders of the IT industry 10 or 15 years from today. The challenge for Intel, the challenge I'll speak to, is really to create the best opportunities for people to build applications that deliver compelling value on our parallel platforms...."
Who Benefits from the Research?
Research conducted through the centers will generally be made available openly. But Microsoft and Intel have certain options for negotiating exclusivity to exploit the technologies.
Said Microsoft's Hey, "Intel and Microsoft have non-exclusive, royalty-free access to patents filed on research supported by the centers. And we also have a right to negotiate an exclusive license based on this research, if that's deemed appropriate. But otherwise, the software that we will be producing in the centers we anticipate will be made freely available under Illinois and Berkeley open source licenses and be a benefit to the whole of the IT industry and to the whole of academia. So we have a non-exclusive, royalty-free license to intellectual property with an option to negotiate an exclusive [one], but otherwise we'll make the software available to everybody."
However, Berkeley's Patterson downplayed the significance of any patents coming out of the research and focused instead on papers and open source software that will be produced. "I've never filed a patent. I've been at Berkeley 30 years, and it's not what I see as one of the critical pieces for the success of technology transfer. The important thing, what we really will do, is do papers, and those we get to publish without restriction. And of course software, which is [going to be made available] as open source: We're going to use the BSD license. So I think that's going to really make it possible to have a real impact with what we're doing, and that's we're excited to be working on it."