Gates Outlines Microsoft's High-Performance Computing Plans

Microsoft's entry into the scientific computing market could yield less expensive, easier-to-use supercomputing systems and pave the way for advances in business computing, chairman Bill Gates said during a keynote speech at a supercomputing conference in Seattle Tuesday.

Microsoft next year plans to release a version of Windows for small supercomputers, and is funding 10 universities in the United States, Europe, and Asia to help develop a set of high-performance computing products that includes operating systems, middleware, and development tools. As computer technology becomes integral to advances in biology, physics, medicine, and earth science, there's demand for simpler tools that can speed up scientists' "time to insight," Gates said during the speech at the SC05 conference.

High-performance computing techniques are also finding their way into industrial applications, including consumer product design and automobile crash testing. Microsoft's upcoming high-performance computing software, its first designed for the market, could help tie together the desktop and powerful clusters of computer servers in new ways, said Gates.

"Computation has become a tool for all the sciences," Gates said. "We need an approach that scales from the smallest supercomputer up to the largest."

Microsoft Tuesday released a second beta version of Windows Compute Cluster Server 2003, a version of its operating system designed to run on clusters of perhaps a few dozen machines. The product is due in the first half of next year, and will compete with the open-source Linux operating system that dominates cluster installations today. Using clustering technology, users can chain together dozens or hundreds of inexpensive PC servers with special cabling to run software that distributes work among the processors. The approach has let universities, supercomputing research centers, and corporate IT departments access supercomputing power at much lower costs than specialized supercomputing architectures have allowed. Microsoft is also designing new capabilities for its graphical development tools that could help scientists program clusters.As part of its development effort, Microsoft has funded 10 "institutes for high-performance computing," including the universities of Washington, Virginia, Tennessee, Utah, and Cornell University. Professors there will provide Microsoft with product development assistance as part of the new program. One challenge Microsoft and its partners will face is the need to develop new software algorithms that can run more efficiently across multiple processors. As chip designers become unable to significantly raise microprocessors' clock speeds to boost performance, the computer industry is turning to integrating multiple processing cores on a chip and extracting more parallel processing capability from software code to fuel performance gains. That means techniques used in high-performance computing today could in five or 10 years become commonplace on the desktop.

In an interview after his speech, Gates said Microsoft's research and development of high-performance computing technologies could help usher in speech and computer vision as user interfaces for PCs, improve the performance of software that can prioritize information for users, and run security algorithms more robustly. "We’ll be able to use all the extra power," says Gates. "In a certain sense, you could say it's been a Holy Grail of computer science" to distribute code that runs well in one machine across many.

During Gates' speech, Microsoft high-performance computing director Kyril Faenov demonstrated a prototype "personal cluster" of four Windows computers running at 25 billion computations per second, on which departments or workgroups could run preliminary calculations before submitting them to a larger system for more detailed modeling of data. The prototype could be priced at less than $10,000, he said. During the demonstration, Faenov ran a version of The MathWorks' Matlab software on the personal cluster to analyze proteomics data, and then transferred the job to a 64-node cluster run by Intel Corp. at a different location in Washington.

Gates said that as part of its high-performance computing effort, Microsoft researchers are entering into new types of collaborations with scientists and medical researchers.