In this issue:
New features in NAG's Algorithmic Differentiation tool for tape-free adjoints, dco/map
In January this year we announced a new release of dco/c++; an Algorithmic Differentiation (AD) tool for computing sensitivities of C++ codes. We are excited to announce that dco/map; a C++11 tape-free operator overloading AAD tool designed specifically to handle accelerators (GPUs etc), has also been updated with new features.
NAG dco/map is a revolutionary AD Tool that provides faster tape-free adjoints on CUDA and any other C++11 compiler. NAG dco/map uses meta programming to instantiate sections of the adjoint code at compile time, reusing program data on the stack. This technique is called meta adjoint programming and creates highly efficient adjoint codes. dco/map is tested, maintained, documented and supported by highly-skilled, and experienced NAG developers.
dco/map key features
- First and second order tangent and adjoint
- Produces single unified code for primal, tangent and adjoint
- Thread safe by design: high performance array and scalar types for shared input data
- Primal as fast as non dco/map primal
- Specialised high-performance array types to handle race conditions inherent in parallel adjoints
- Supports whole of C++11, cross platform
- API for storing things you don't want to recompute
- Easy integration with NAG's dco/c++ via external adjoint interface
Benchmarking as the answer to HPC performance and architecture questions
Andrew Jones, NAG VP Strategic HPC Consulting & Services, recently delivered an informative webinar on:
- When to use benchmarking
- What benchmarks can and can't tell you
- Rules, consistency and pitfalls
- Selecting the most appropriate benchmarks
- Extrapolating to larger scales or newer technologies
- Tips, tricks, and best practice
To view the webinar recording visit our YouTube channel https://youtu.be/baNYuKesbtk
If you'd like to learn more about NAG's extensive HPC work do get in touch via firstname.lastname@example.org or see the HPC area on the web.
Fortran Modernization Workshop - new dates announced
The successful series of Fortran Modernization Workshops are continuing in 2018 with new dates recently announced for sessions in Germany. The two-day practical hands-on workshops - open to all - are aimed at Fortran programmers who want to write modern code, or to modernize existing codes, to make it more readable and maintainable by encouraging good software engineering practices. This workshop will also present how to integrate tools and techniques for Fortran codes to help you develop sustainable software for your scientific and academic research particularly in a collaborative environment. If you'd like to attend a workshop pre-registration is required. See our website for more details.
- Leibniz-Rechenzentrum, Garching bei München, 4-5 June 2018
- RWTH Aachen, 7-8 June 2018
- HLRS High Performance Computing Center, Stuttgart, 12-13 June 2018
2x performance improvement for chemistry application via POP
NAG is a partner in the EU POP project - Performance Optimisation and Productivity. NAG colleagues, working within the POP project provide performance optimisation and productivity services to academia and industry.
A recent POP performance study resulted in a 2x performance improvement for code belonging to Software for Chemistry and Materials (SCM) based in The Netherlands. ADF is the flagship code from SCM. It is a computational chemistry application which uses density functional theory calculations to predict the structure and reactivity of molecules.
A POP Audit and Performance Plan were carried out and uncovered an issue with load imbalance due to an unequal distribution of work. The POP team made a recommendation to improve the load balancing algorithm, which when implemented resulted in the application running at twice its original speed.
Image above shows original timeline of load balance - communication lines in black
Read the full POP and ADF story here.
Latest NAG Student Prize Winner
We are delighted to announce the latest recipient of the NAG Prize for MSc in Mathematical Finance from the University of Manchester is Xin Xu.
Xin Xu received the Prize alongside Denis Denisov (right), Director of the MSc in Mathematical Finance at the University of Manchester. Following her excellent MSc achievement Xin has started her PhD in Actuarial Science, looking at Levy Processes in Insurance.
Congratulations Xin from all at NAG.
Maths Beyond School - NAG Schools Outreach
It was great to be invited to take part in 'Maths Beyond School' at Burford School recently. Burford School is a local secondary school close to our Oxford office for pupils aged 11-18 years. The day was held to inform Burford's year 9 and 10 students of the different opportunities open to those who choose to study maths at higher education, and hopefully inspire them to go on and study maths after their GCSEs and A' Levels. Read more
Out & About with NAG
Exhibitions, Conferences, Trade Shows and Webinars
The Trading Show Chicago - 9-10 May 2018
PyCon - 9-17 May 2018
QuantMinds International - 14-19 May 2018
We are pleased to offer our contacts a 20% discount on delegate registrations, please quote "FKN2552NAG" when registering. For more information see the event website.
OpenMP Users Conference - 21-22 May 2018
European HPC Summit Week Ljubljana - 28 May - 2 June 2018