Intel and Microsoft Fund Multi-Threading Research

Intel and Microsoft Fund Multi-Threading Research

It's no secret that the days of procuring performance exclusively through faster clock speeds are over. The current crop of multi-core server, desktop, and mobile CPU designs are a dead giveaway that processor vendors like Intel are instead looking to increased parallelism as the facilitator of more computing horsepower. The problem, according to David Patterson, professor of computer science at UC Berkeley, becomes one of scheduling and balancing workloads across multiple sets of processing resources, so that you can truly get more performance out of those threaded designs. If an application is only running in one thread, upgrading to a dual- or quad-core chip won't help speed it up. It's a kin to revving your cars engine in neutral.  There's plenty of power there and it sounds good but you're not going anywhere.

Even today, writing software able to take advantage of multi-threading is notoriously difficult. In order to help drive the development of the tools and threading-aware applications, Microsoft and Intel are together awarding the University of Illinois at Urbana-Champaign and the University of California at Berkeley $10 million dollars over five years to fund two Universal Parallel Computing Research Centers (UPCRC). An additional $8 million will come from the University of Illinois and Berkeley has applied for $7 million in funding from a state-supported program. The aim of each center will be advancing parallel programming, architecture, and operating systems, according to Intel's release.

Professor David Patterson will step in as director of the new UPCRC at the UC Berkeley campus

Don't think that these research efforts will touch you? Think again. Dunnington, the six-core Xeon MP successor Intel recently unveiled, sets a new standard in single-socket density. That CPU will be used in enterprise environments already rife with threaded software. Intel's upcoming Nehalem architecture will scale from two to eight cores, pulling that same emphasis on parallelism down to the desktop. The difference is that a vast number of mainstream applications--including games--are still not threading-aware. As major research centers, such as the two new UPCRCs, publish papers, file for patents, and write software, desktop users stand to reap the largest gains in the form of higher performance and more sophisticated applications. Tony Hey, corporate vice president of external research at Microsoft Research, gave us an idea of what he hopes the relationship between Microsoft, Intel, and the two universities will be able to achieve.

Marc Snir, professor of computer science, will co-direct the UPCRC at the University of Illinois.

"Imagine an intelligent digital assistant that...could prioritize your incoming emails, incoming calls...can identify conflicts; it can make suggestions for priorities. Another one could be a personal health care assistant. As a third example, when we're on conferences, speech understanding and automatic translation that can recognize who is speaking and in what language, and gives you the solutions you want."

According to Hey, this is the first collaboration between major industry players and academia of such magnitude. Hopefully, it'll serve as a catalyst for ISVs who currently have to balance the extra cost of implementing threaded code with the potential performance gains of going highly parallel.

For more information, check out the release in Intel's
Press Room

Related content