Novel Approaches to Accelerate Chemistry for Climate Models
DescriptionThe increase in Earth System Models (ESMs) capabilities is strongly linked to the amount of computing power and data storage capacity available. Exascale ESMs are expected to open up a new range of opportunities, from increasing model resolution and larger ensemble simulations to the integration of new components to increase the accuracy of our predictions. To make this possible, exascale supercomputers will rely on accelerators and specialised hardware. One critical goal to take advantage of the new exascale era is to exploit the new hardware and accelerate the models through kernels and data structures suitable for the new GPU partitions. This should be done by 1) focusing on computationally intensive parts and 2) optimizing the new kernels to take into account the particularities of the accelerators. If we think about computationally intensive parts, chemistry is usually one of the most time-expensive component of climate models, something not trivial because they were written exclusively for CPUs in the past. However, we aim to achieve an efficient and heterogeneous implementation that exploits both CPU and GPU hardware. This talk will cover new solutions to accelerate climate models and the porting of a chemistry module will be used as an example.
TimeMonday, June 2617:00 - 17:30 CEST
Climate, Weather and Earth Sciences