Visit additional Tabor Communication Publications
May 23, 2011
Researchers at the PCS and HP Labs achieve unprecedented speedup in a key machine-learning algorithm.
PITTSBURGH, May 23 -- Computational scientists at the Pittsburgh Supercomputing Center (PSC) and HP Labs are achieving speedups of nearly 10 times with GPUs (graphic processing units) versus CPU-only code (and more than 1000 times versus an implementation in a high-level language) in k-means clustering, a critical operation for data analysis and machine learning.
A branch of artificial intelligence, machine learning enables computers to process and learn from vast amounts of empirical data through algorithms that can recognize complex patterns and make intelligent decisions based on them. For many machine-learning applications, a first step is identifying how data can be partitioned into related groups or "clustered."
Ren Wu, principal investigator of the CUDA Research Center at HP Labs, developed advanced clustering algorithms that run on GPUs, which have advantages for many data-intensive applications. PSC scientific specialist Joel Welling recently applied Wu's innovations to tackle a real-world machine-learning problem. Using data from Google's "Books N-gram" dataset and working together, Wu and Welling were able to cluster all five-word sets of the one thousand most common words ("5-grams") occurring in all books published in 2005. With this project, representative of many research efforts in natural-language processing and culturomics, the researchers demonstrated an extremely high-performance, scalable GPU implementation of k-means clustering, one of the most used approaches to clustering.
Wu and Welling ran on the latest "Fermi" generation of NVIDIA GPUs. Using MPI between nodes (three nodes, with three GPUs and two CPUs per node), they observed a speedup of 9.8 times relative to running an identical distributed k-means algorithm (written in C+MPI) on all CPU cores in the cluster, and thousands of times faster than the purely high-level language implementation commonly used in machine-learning research. Using their GPU implementation, the entire dataset with more than 15 million data points and 1000 dimensions can be clustered in less than nine seconds. This breakthrough in execution speed will enable researchers to explore new ideas and develop more complex algorithms layered atop k-means clustering.
"K-means is one of the most frequently used clustering methods in machine learning," says William Cohen, professor of machine learning at Carnegie Mellon University. "It is often used as a subroutine in spectral clustering and other unsupervised or semi-supervised learning methods. Because some of these applications involve many clustering passes with different numbers of means or different randomized starting points a greatly accelerated k-means clustering method will be useful in many machine learning settings." Cohen co-leads the Never-Ending Language Learning (NELL) and Read the Web projects (http://rtw.ml.cmu.edu/rtw/). The goal of NELL is to automate inferences based on continually "reading" natural-language text from the Web.
Machine learning is just one example of the exploding field of data analytics, notes PSC scientist Nick Nystrom. Other data-analytic applications range from understanding the results of traditional high-performance computing (HPC) simulations of global climate, engineering, and protein dynamics to emerging fields that need HPC such as genomics, social network analysis, and mining extensive datasets in the humanities.
"A substantial body of major application codes is already being developed specifically for NVIDIA GPUs," notes Nystrom, PSC director of strategic applications. "Because NVIDIA GPUs are so widespread, those codes can run well on anything from a supercomputer to a netbook." Nystrom has been instrumental in PSC's work with advanced technologies for scientifically important, data-intensive problems. This application of NVIDIA GPUs to k-means clustering, he notes, is one example of how a pervasive technology that leverages broad markets can benefit important algorithms in science and data analysis.
This advanced clustering algorithm, notes Wu, also has the advantage of being easy to use, which facilitated rapid implementation with Welling. "I think that the CUDA programming model is a very nice framework," says Wu, "well balanced on abstraction and expressing power, easy to learn but with enough control for advanced algorithm designers, and supported by hardware with exceptional performance (compared to other alternatives). The key for any high-performance algorithm on modern multi/many-core architecture is to minimize the data movement and to optimize against memory hierarchy. Keeping this in mind, CUDA is as easy, if not easier, than any other alternatives."
Nystrom concurs and sees an exciting future for software developers: "There's a rich software ecosystem supporting NVIDIA's GPUs, ranging from easy-to-use compiler directives to explicit memory management to powerful performance tools. Add to that integration of general-purpose processors in this successful line of architectures, and the potential for developing transformative software architectures is extraordinary."
The Pittsburgh Supercomputing Center is a joint effort of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Company. Established in 1986, PSC is supported by several federal agencies, the Commonwealth of Pennsylvania and private industry, and is a partner in the National Science Foundation TeraGrid program. http://www.psc.edu.
Source: Pittsburgh Supercomputing Center
In quieter times, sounding the bell of funding big science with big systems tends to resonate further than when ears are already burning with sour economic and national security news. For exascale's future, however, the time could be ripe to instill some sense of urgency....
In a recent solicitation, the NSF laid out needs for furthering its scientific and engineering infrastructure with new tools to go beyond top performance, Having already delivered systems like Stampede and Blue Waters, they're turning an eye to solving data-intensive challenges. We spoke with the agency's Irene Qualters and Barry Schneider about..
Large-scale, worldwide scientific initiatives rely on some cloud-based system to both coordinate efforts and manage computational efforts at peak times that cannot be contained within the combined in-house HPC resources. Last week at Google I/O, Brookhaven National Lab’s Sergey Panitkin discussed the role of the Google Compute Engine in providing computational support to ATLAS, a detector of high-energy particles at the Large Hadron Collider (LHC).
May 23, 2013 |
The study of climate change is one of those scientific problems where it is almost essential to model the entire Earth to attain accurate results and make worthwhile predictions. In an attempt to make climate science more accessible to smaller research facilities, NASA introduced what they call ‘Climate in a Box,’ a system they note acts as a desktop supercomputer.
May 22, 2013 |
At some point in the not-too-distant future, building powerful, miniature computing systems will be considered a hobby for high schoolers, just as robotics or even Lego-building are today. That could be made possible through recent advancements made with the Raspberry Pi computers.
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 15, 2013 |
Supercomputers at the Department of Energy’s National Energy Research Scientific Computing Center (NERSC) have worked on important computational problems such as collapse of the atomic state, the optimization of chemical catalysts, and now modeling popping bubbles.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/15/2013 | Bull | “50% of HPC users say their largest jobs scale to 120 cores or less.” How about yours? Are your codes ready to take advantage of today’s and tomorrow’s ultra-parallel HPC systems? Download this White Paper by Analysts Intersect360 Research to see what Bull and Intel’s Center for Excellence in Parallel Programming can do for your codes.
In this demonstration of SGI DMF ZeroWatt disk solution, Dr. Eng Lim Goh, SGI CTO, discusses a function of SGI DMF software to reduce costs and power consumption in an exascale (Big Data) storage datacenter.
The Cray CS300-AC cluster supercomputer offers energy efficient, air-cooled design based on modular, industry-standard platforms featuring the latest processor and network technologies and a wide range of datacenter cooling requirements.