New report outlines exascale computing research priorities for the Met Office
August 2016 - This research news article looks back at the main findings from a technical report published in May 2016, which sets out long term research priorities for exploiting new computing technologies and how they might be used at the Met Office.
Use of supercomputers
The Met Office uses supercomputers to efficiently solve the mathematical models which describe the atmosphere so that predictions of the weather and the climate can be made.
Exploiting new technologies is necessary because the processor speed (i.e. the clock speed of the processor) has not increased for successive generations of processor since the year 2005.
Moore's Law, which says that the number of transistors in an area of silicon doubles every 18 months or so, used to mean that each successive generation of processor was faster than the previous one. However, it has not been possible to reduce the electrical power consumed for these smaller circuits, so to prevent the silicon overheating, the processor clock speed cannot be increased.
Instead, more transistors are used to create silicon chips that have many central processing unit (CPU) 'cores'. For example the previous Met Office supercomputer, based on IMB Power 7, had around 40 thousand cores across two machines. The current supercomputer, a Cray XC40 with Intel processors, will have a combined total approaching half a million cores. The Met Office will deliver around £2 billion of socio-economic benefit to the UK as a result of this supercomputer.
Met Office technical report - May 2016
A Met Office technical report published in May 2016 looks beyond the time-frame of the current supercomputer, where other types of processors besides CPUs could be used to prevent the total power consumption of the supercomputer becoming too large. These include:
- low power processors, such as those based on the ARM processor design used in mobile devices;
- Graphical Processor Units (GPUs), used for processing the graphics in games consoles and other devices;
- and Intel's Xeon Phi processor.
More exotic devices such as Field Programmable Gate Arrays (FPGAs) and Quantum devices are considered. The technical report also examines the impact this would have on how scientific software is developed and what research needs to be done into how to program such devices effectively.
Research into developing new mathematical algorithms that can exploit the large number of cores to solve the mathematical model of weather and climate is also required.
Next generation modelling
The Met Office is already developing a next generation weather and climate model with a time-frame of beyond the current supercomputer. The technical report concludes that these disruptive technology changes require research into computational science techniques and methods so that scientific advances of the new models can effectively exploit these new computing technologies.