Will it ever be possible to improve the resolution of weather prediction to such an extent? Only time will tell, along with the success of using complex solvers tested on facilities like the University of Bath's Aquila HPC service.
Dr Eike Mueller won the prize for best presentation at the 2013 High Performance Computing (HPC) symposium at the University of Bath. Since September 2011 Eike has been a PostDoc in the Department of Mathematical Sciences at the University of Bath where he works on a HPC driven NERC funded project for improving the accuracy of global weather and climate prediction models for the Met Office.
Background to Eike Mueller's work
The UK Met Office provides global, 5 day weather forecasts. Data is fed into a computational model, this data is sourced using a grid with reference points 25 km apart. The aim is to increase the resolution of this forecast and use data from reference points 1 to 2 km apart. This requires a rethink of the computational model being used. Together with Professor Rob Scheichl, Eike is improving the performance of the solver for the global pressure correction equation, which is one of the computationally most important components of the model.
Rethinking computation models - how to make the most of Graphical Processing Units (GPU)
Traditionally users of the Aquila HPC service at the University of Bath run jobs over long periods of time, using a smaller percentage of the central processing power available from Aquila. Bath University Computing Services (BUCS) works to make sure that all requests for using the service are run as efficiently as possible.
Dr Mueller's work is different from these more traditional jobs. He is the first at the University of Bath to take advantage of the newly installed NVIDIA Fermi GPUs, which offers a total of 3,072 GPU cores. In addition, his work requires using the majority of the Aquila's capacity for very short periods of time. He is using Aquila to develop and test the performance of algorithmic and parallel scalability of solvers for elliptic partial differential equations in weather and climate prediction on novel computer architecture (in this example GPUs). Dr Steven Chapman from BUCS has provided support to assist Eike in getting the most out of the GPUs on the Aquila cluster.
What is next for Dr Mueller's work?
Eike is currently extending the work he presented in his talk at the HPC Symposium to run his solver on the EMERALD cluster, the UK's biggest NVIDIA GPU-accelerated supercomputer. In the future he also plans to test the scalability of his code on the TITAN supercomputer, which is currently the second fastest computer in the world. Having a local HPC facility allows him to test his work and solve any issues that result from the tests.