The Weekly Top Five features the five biggest HPC stories of the week, condensed for your reading pleasure. This week, we cover the computing power on display at SC10’s Student Cluster Competition; the University of Portsmouth’s new supercomputer; IBM Watson’s Linux platform; multicore advances at North Carolina State; and Intel’s new approach to university funding.
SC10’s Student Cluster Competition Raises the Bar
The student team from Louisiana State University was one of three teams to break the teraflop barrier at SC10’s Student Cluster Competition. This is the first year that any team has achieved that distinction, and the honor is shared with teams from the University of Texas and National Tsing Hua University (Taiwan).
In the Student Cluster Competition at SC10, which took place in New Orleans in November, eight teams gathered from around the country and from as far away as Russia and Taiwan to design and build clusters that solve real-world problems. The teams prepared for months working with their advisors and vendor partners. Winning teams were selected by a panel of experts, based on visualization output, presentations and interviews.
The LSU students received vendor support from HP and LATG, Mellanox, Portland Group and Adaptive Computing and were advised by Isaac Traxler, Unix Services Manager at LSU’s High Performance Computing (HPC) and Center for Computation & Technology. Under Traxler’s tutelage, the students spent one night a week for six months working on the project, in addition to many hours spent working on their own.
With 144 cores, the LSU cluster executed the competition’s four open source applications while staying within the 26 Amp constrained power limit.
UK-Based Supercomputer to Further Cosmic Reaserch
Scientists at the University of Portsmouth are about to get a new supercomputer, one that has the equivalent strength of approximately 1,000 desktop systems. The system will give cosmologists an edge when it comes to understanding galaxy formation and even the origin of gravity itself.
Named “SCIAMA,” the 1,008-core cluster was built by Dell and designed to process large amounts of astronomical data very quickly. Researchers at the University’s Institute of Cosmology and Gravitation (ICG) will use the cluster to solve complex cosmological problems, like simulating vast regions of the universe and exploring the properties of hundreds of millions of galaxies.
The supercomputer was named in honor of Dennis Sciama, a leading figure in the astrophysics and cosmology community. SCIAMA is also an acronym for SEPnet Computing Infrastructure for Astrophysical Modelling and Analysis.
Gary Burton, ICG’s senior specialist technician and the person who will soon be managing the supercomputer, explained that “the huge power of a supercomputer like SCIAMA is necessary to deal with the vast amount of observational data coming from satellites, telescopes and other detectors. Using it will allow us to explore the whole of cosmic history and analyse data that contains fundamental clues about the origins of the Universe.”
Watson Supercomputer Is SUSE Linux Machine
SUSE Linux is about to get its 15 minutes of fame. Novell announced this week that IBM Watson’s DeepQA software is running on SUSE Linux Enterprise Server 11. Watson is the supercomputer that will soon have the distinction of being the first non-human Jeopardy contestant. The novel tournament takes place Feb. 14-16.
Watson contains more than 200 million digital pages of information and operates at a speed of over 80 teraflops. IBM has designed Watson with a combination of deep analytics and rapid processing speeds that can make sense of the kind of “natural language” questions that are at the core of this popular primetime gameshow.
Linux has a long history of use in the field of high performance computing, and this is spelled out in the announcement:
Watson’s “Jeopardy!” appearance serves as further validation of the advantages of Linux in high-performance computing environments, as Linux has long been regarded as the operating system of choice among the fastest and most complex environments in the world. In the latest TOP500 list of the world’s most powerful supercomputers, 459 are running Linux and six of the top 10 systems are based on SUSE Linux Enterprise or a derivative of it.
NC State Research Team Speeds Chip Communication
North Carolina State University researchers have developed a hardware technology, called HAQu, that boosts software performance by enabling chip-to-chip communication. In multicore setups, the core communication is rather inefficient, with the chips using memory as the “third-party” intermediary. If the chips could communicate with one another directly, it would save a lot of time.
The computer engineers have detailed their findings in a paper, called “HAQu: Hardware-Accelerated Queueing for Fine-Grained Threading on a Chip Multiprocessor, (PDF)” which will be presented at the International Symposium on High-Performance Computer Architecture in San Antonio, Texas, on Feb. 14.
Dr. James Tuck, an assistant professor of electrical and computer engineering at NC State and co-author of the paper, explained in the university’s announcement that the “technology is more efficient because it provides a single instruction to send data to another core, which is six times faster than the best state-of-the-art software” (that the researchers could find). He went on to state that HAQu is “not hardware designed to communicate data on its own, but is hardware that expedites data-sharing using existing data paths on a computer chip.”
Even though it is a piece of hardware, HAQu is similar to software communication tools in that it is able to leverage a chip’s existing data paths. It is also reduces energy draw. Despite using more energy, it runs more quickly, resulting in a net decrease in consumption.
The same research team was responsible for a parallelization technique that could enable common computer programs to run up to 20 percent faster. The non-traditional approach works on programs that are normally difficult to parallelize, such as as word processors and Web browsers, by running memory-management functions on a separate thread. That work has also been written up as a paper (PDF).
Intel Labs Commits $100 Million to University Research
Intel Corp. announced intentions to invest $100 million into US university research over the next five years. With this new model, funding to researchers could increase five-fold.
Intel Labs will launch multiple Intel Science and Technology centers over the coming year in an effort to boost innovations in computing and communications. The centers will pursue advances in visual computing, mobility, security and embedded solutions. Stanford University will host the first center, with a focus on creating visualization solutions for both consumer and professionals.
From the release:
This first Intel Science and Technology Center, as well as those that will follow later this year, represents a new model of collaboration for the company. Until now, Intel Labs ran open collaboration centers near research universities and a substantial portion of the company’s funding focused on operating, maintaining and staffing these facilities. The new centers will be Intel-funded and jointly led by Intel and university researchers. They are designed to provide more dollars in the hands of researchers, and to encourage tighter collaboration between academic thought leaders in essential technology areas such as visual computing, security and mobile computing. For maximum flexibility, Intel will be able to tune its research agenda across the research centers over time. Intel plans to invite proposals from the academic community to continue pursuing the creation of additional Intel Science and Technology Centers.
Read more about the Stanford-based center, here.