EUPEX Webinar – Manual and Automated Vectorisation Techniques for the Integrated Forecasting System (IFS)

We are organising monthly webinars on the results of EUPEX (full programme here). Our eighth webinar will focus on IFS, the weather forecasting system developed by ECMWF and NEMO an ocean model.

Registration link here : 

https://us06web.zoom.us/meeting/register/15n0wwfQTNaVw3O4NSTFqg

Abstract: What’s better, a manual or automated approach to optimisation? As always, the answer is somewhere in the middle. In this webinar we will explore automated optimisation techniques applied to CloudSC (a component of the IFS) via Loki, and the effects they have on performance on EUPEX’s SDV platform. We also look at the NEMO ocean model and the way it runs on GPUs, particularly NVIDIA Grace Hopper GPUs of the EUPEX SDV. Since NEMO was originally based on MPI for CPUs, we are porting it to GPUs to evaluate the potential speedup that can be achieved. This work provides insights to help others evaluate which approach may be most suitable for different scenarios when preparing for exascale systems.

Speakers:

Andrew Beggs, ECMWF
Andrew works as a Computational Scientist in the Computing Department at ECMWF. Originally from just north of Belfast in Northern Ireland, he studied computing at Imperial College London. At ECMWF, his work focuses primarily on hardware acceleration and optimisation techniques. Before joining ECMWF, he worked at ARM in a similar field.

Laleh Pouyan, ECMWF
Laleh works as a Computational Scientist in the Computing Department at ECMWF. Originally from Iran, she moved to Germany to pursue a PhD in High-Performance Computing (HPC). She joined the ECMWF Computing Department in August 2022. Her research mainly involves profiling, benchmarking, and optimising parts of the IFS application. She is currently working on porting the NEMO ocean model to GPUs using OpenACC.

Registration link here :

https://us06web.zoom.us/meeting/register/15n0wwfQTNaVw3O4NSTFqg