I recently gave a presentation on POP at the 17th Workshop on High Performance Computing in Meteorology, which was held at the European Centre for Medium-Range Weather Forecasts (ECMWF) in Reading, UK, between 24 and 28 October 2016. The workshop was extremely well attended, with over 100 attendees coming from Europe, America and Australia. The importance of performance optimisation for numerical weather prediction code was a topic that came up in a number of presentations; one talk described how a 40x speedup was achieved over the course of 5 years through a combination of code optimisation, use of GPUs, and hardware improvements.
I gave a presentation about POP, in which I gave an overview of the project and highlighted some of the audits that have been done on Earth Sciences codes. It was a great opportunity to meet existing users of POP, some of whom were interested in signing up more of their codes for audits. Videos of all the presentations, as well as copies of the slides, can be found on the workshop's webpage.
The workshop really brought home to me that predicting the weather requires significant HPC resources: ECMWF's Cray XC40 is number 18 on the most recent (June 2016) Top 500 list of supercomputers, with a peak performance of over 4 petaflops. ECMWF is also home to a large collection of rubber ducks (pictured) that have been donated by members of staff and visitors from all over the world. The collection recognises the important contribution rubber ducks have made to the understanding of ocean currents.
Nick Dingle (Numerical Algorithms Group)