Unrivalled functionality, increased speed and heightened performance for shared memory parallelism
June 2006. The latest release of the NAG SMP Library has been extensively enlarged to include over 1500 user-callable mathematical and statistical routines for high performance computing. As the world’s leading provider of mathematical and statistical software for developers, the Numerical Algorithms Group’s SMP Library is the largest collection of quality numerical algorithms available today for high performance computing (HPC).
With the introduction of cost-effective multi-core processors by AMD, Intel, Sun and IBM, HPC is becoming increasingly accessible to both researchers and commercial users. The NAG SMP Library helps unleash the power of these new processors by making available tuned and enhanced routines which automatically take advantage of multiple processors while maintaining NAG’s standards for accuracy and robustness.
During recent benchmark studies the NAG SMP Library outperformed all comparable products, often showing vast improvements in terms of speed, performance and accuracy, crucial for programmers in today’s highly competitive high performance computing environment. Performance results of the SMP Library, Mark 21 are available at http://www.nag.com/numeric/fl/smpbenchmarks.asp.
Professors and students at both The University of Hyderabad, India and The University of Cambridge, UK, including the National Cosmology Supercomputer project (COSMOS), have been using Mark 21 of the NAG SMP Library since its beta-release in 2005, and are delighted with the speed, performance and accuracy of the library.
Many of the world’s most prestigious HPC organizations have already recognized the benefits of using the NAG SMP Library in their numerical computation. The Hokaido University Supercomputing Center in Japan and the High Performance Computing Virtual Laboratory in Canada testify to the many benefits of using the library. Alongside these highly acclaimed centers for Supercomputing, many of the world’s most highly regarded learning institutions use the SMP Library in their research and teaching including Cornell University, Stanford University and the University of California at Berkeley in as well as the US Department of Commerce’s Geophysical Fluid Dynamics Laboratory.
Key benefits of NAG SMP Library:
Unrivalled speed, performance and accuracy over comparative products
Contains over 120 specialized routines tuned for maximum performance and speed
Delivers improved performance in over 250 other routines including key numerical areas such as optimization, statistics and partial differential equations (PDEs)
Easy to use and link to due to identical interface to NAG Fortran Library
Highly flexible - assists rapid migration from serial code, and between platforms
Interoperable – routines can be called from multiple computing languages
Based on the OpenMP Program Interface (API), which supports multi-platform shared-memory parallel programming in C/C++ and Fortran on all architectures including Unix and Windows platforms.
Specialized functionality of SMP Library includes:
Fast fourier transforms (FFTs)
Dense linear algebra (LAPACK)
Sparse iterative solvers
Sparse direct solvers (based on SuperLU)
Sparse iterative eigensolvers (based on ARPACK)
Other areas that benefit from tuned routines include:
Ordinary differential equations (ODEs)
Optimization
Multivariate statistics
Linear algebra
Commenting on the new release, Rob Meyer, CEO, NAG Group, said: “This new version of the NAG SMP Library provides a number of significant additions and complements the arrival of new hardware and operating system platforms. It is ideally suited to help those building applications in finance and other computationally intensive areas take advantage of the new generation of hardware.”
Background detail
The NAG SMP Library was produced to enable developers and programmers to make optimal use of the processing power and shared memory parallelism of the Symmetric Multi-Processor (SMP) systems. Select routines within the NAG SMP Library (http://www.nag.com/fs) have been specially developed and tuned to provide the utmost performance on SMP platforms. These tuned routines deliver levels of performance and scalability superior to many other products currently available. Indeed, NAG has uniquely developed and pioneered many parallelized computational algorithms.
A key component of this exciting new release is the inclusion of a new group of routines for large-scale eigenvalue problems. Eigenvalue and eigenvector problems are ubiquitous. Typically, for example, engineers perform such calculations to complete a vibration analysis on a structure. The early problems of the Millennium Bridge across the Thames were a fine demonstration of what can go wrong. The new suite of routines is capable of solving very large systems, and can handle both standard and generalized eigenproblems. These new, parallelized eigenproblem routines compliment the new, parallelized direct linear equation solvers and the existing parallelized iterative solvers to provide a comprehensive solution to sparse linear algebra problems.
NAG is proud to have provided two of the major contributors to the LAPACK project. While Jeremy du Croz has since retired, NAG Principal Technical Consultant, Sven Hammarling, is still active in this area. Under his guidance, NAG has completed the task of putting the whole of LAPACK 3 into the library software. Sven’s involvement has ensured that the latest LAPACK3 code, including error corrections, has been included.
Among the LAPACK computational routines optimized for use on SMP systems in Mark 21 of the NAG SMP Library are routines in the areas of band solvers, iterative refinement and error bounds estimation, band reduction and symmetric and unsymmetric eigenproblems. Additional modifications to NAG library routines increase the benefit of the underlying parallelized LAPACK routines to other NAG routines in areas such as PDEs and multivariate statistics.
Other additions to the NAG SMP Library include new optimization routines offering real performance gains over previous routines. Comparative tests have shown marked improvements in speed of computation. The Library is an essential tool for personnel with an interest in computing portfolios or in tracking indices, and will be of particular benefit to analysts in the financial industry. Also of particular benefit to the finance industry are new and improved random number generation routines, (which are heavily used when carrying out stochastic simulation), the inclusion of copulas, and improvements to the quasi-random number routine interfaces.
Expansions in linear regression software included in the latest release will benefit a wide range of industries and academic areas. A stepwise regression routine allows the user to examine a large number of models in a short period of time and is an excellent tool for hypothesis generation. When dealing with databases having a large number of potentially interesting explanatory variables, stepwise regression can be used to quickly select the most interesting subset of variables, which can then be examined in more detail using other techniques.
The statistical area of mixed effects regression, also included in the new release, can be applied in a wide range of different situations and as such is becoming common in a wide range of fields including the pharmaceutical and engineering sectors.
Covering Windows, Linux and major Unix platforms, the NAG SMP Library is available on machines ranging from PCs to supercomputers. NAG customers can benefit from the Library’s design which allows algorithms to be called from other languages such as C/C++, Fortran, Java and the .NET languages. Technical help for anyone subscribing to NAG’s Customer Support Service is available from the experts responsible for the relevant software.
About NAG
NAG (http://www.nag.com) is dedicated to making world-class cross-platform mathematical, statistical, data mining components and tools for developers as well as 3D visualization application development environments. Headquartered in the UK (Oxford) it operates worldwide with hubs in Chicago and Tokyo. Today it serves over 10,000 sites worldwide in finance, engineering, and scientific research as well as commercial software firms such as Oracle, IBM, DemandTec, and many others.