NAG: available libraries for all applications
Starting with François Cassier, Account Manager Europe of Numerical Algorithms Group Ltd (NAG) explained what this structure is about and how it works. “Collaboration has been in NAG's genes since its creation in 1970, as it is based on a collaborative project between several British universities in the field of numerical computation. We became an independent commercial company in 1976 and notably a non-profit organization, living off income from its products and services. We achieved a turnover of £8.4m in our last financial year and now employ 70 people”.
With an HPC entity in Manchester, this British structure based in Oxford has subsidiaries in the USA and Japan, as well as offices in France and Germany. NAG has 3,000 customer sites worldwide operating in industrial, financial and oil markets. NAG also works with processor manufacturers (AMD, ARM, Intel...) and software publishers.
"Being a non-profit structure protects us from acquisitions, which also allows us to have a long-term vision of our missions. They have been of two kinds for 50 years: to offer our customers robust and efficient digital solutions to solve their technical problems on the one hand and, on the other hand, to help them bring the next generation of numerical computing users within their communities”.
NAG is well known for its library which includes hundreds of mathematical and statistical components, that developers integrate into their own computing and data analysis applications to improve performance in Fortran, C, C, C++, Python, Matlab and other environments. "Our library is widely used by the CEA's TGCC hosting us for this General Assembly”.
In addition to this library, NAG offers: high-performance Fortran compilers for error detection; DCO tools dedicated to algorithmic differentiation (calculation of code derivatives) as one key alternative to finite difference; Origami tool, providing a development and execution environment for Grid and Cloud Computing.
NAG also offers a number of digital services such as development-oriented algorithms or other HPC-based code aiming at modernization and parallelization, technology assessment and benchmarking, as well as consulting in the implementation of HPC systems or for supply assistance.
"We have a long-standing collaboration history with scientists and engineers around the world in the private or public, academic or industrial sectors. For example, we are currently working with the Universities of Lancaster, Manchester, Oxford, Sheffield and Warwick to enrich the NAG library with new functions, or with Stanford University on mathematical optimization. We also have a partnership with the University of Aachen in Germany on algorithmic differentiation for industrial and financial applications. Finally, we are working on European projects such as POP 2 which as we know aims to improve codes, or POEMA with INT Marie Curie for polynomial optimization."
NAG is very much involved in the digital community, since it participates in many normative bodies (Fortran and MathML languages, Lustre OpenSFS file system, Blas & Lapack libraries). We also collaborate with many universities to promote careers in science, technology, engineering and mathematics, and we contribute to the Women in HPC initiative.
Addressing a question from the audience, François Cassier said: "We do indeed work mainly around CPUs, but we also have a consulting activity around GPU accelerators".
|
