The 17th Workshop on High Performance Computing in Meteorology took place at ECMWF from 24 to 28 October 2016. The theme of this biennially organised workshop was the scalability of numerical weather prediction (NWP) systems. Over the four and a half days, more than 40 speakers presented their vision of what kind of systems and programming paradigms we might be using in a few years’ time or even in the next decade. The event was attended by 100 external participants from ECMWF’s Member and Co-operating States as well as from the USA, Canada, Japan, South Korea and Australia. Computer vendors and software providers were also present.
Keynote talks were given by Dr Jack Dongarra from the University of Tennessee, who developed the LINPACK test used to estimate the performance of supercomputers, and by Dr Thomas Schulthess, the Director of the Swiss National Supercomputing Centre (CSCS), who has been in charge of installing the largest GPU-cluster in Europe for operational weather prediction. Dr Dongarra gave an update on the latest HPC developments and set out the ‘changing rules at exascale’, while Dr Schulthess asked whether exascale computing represented the ‘endgame’ or a ‘new beginning for climate modelling’. In addition, a talk by Professor Tim Palmer from the University of Oxford on using lower precision in meteorology triggered a lot of discussion. The decision to distribute and mix the talks across the week paid off as this kept the momentum going and there were no sessions without interesting topics. We also introduced a new kind of panel discussion, which took place on Thursday afternoon and gave rise to a lot of debate and exchanges of ideas on how to develop NWP coding to make it fit for future systems.
The presentations and video recordings of the talks are available online at www.ecmwf.int/en/learning/workshops-and-seminars/17th-workshop-high-performance-computing-meteorology. The next HPC workshop will take place in the autumn of 2018.