Implementation and computing performances of NEMO ORCA-KM configuration - CNRS - Centre national de la recherche scientifique
Rapport (Rapport Technique) Année : 2019

Implementation and computing performances of NEMO ORCA-KM configuration

Résumé

Popularity of one kilometre scale global geophysical models is continuously growing in Europe. In this study, we increase the resolution of the NEMO ocean model, to better identify technical limitations of the present implementation and deduce, from the performance measured on one of our top supercomputers, the hardware characteristics needed to routinely use a global ocean one kilometre configuration. 24,000 Intel Broadwell cores of the Météo-France supercomputer are required to measure the performance of a simplified version of NEMO global kilometre scale. From them, assuming a perfect scaling up to 9,000,000 (nine million) computing cores, we infer that we could target a maximum speed of approximately 2 SYPD with the ORCA-Km global model. This configuration needs, at least, 4 (four) orders of magnitude more CPU power and energy than one of the most demanding coupled model of the community (CNRM-CM6-HR). It is premature to start considering the possibility of any production run based on the global one kilometre scale NEMO model, because no present day machine can fulfil the ORCA-Km memory/bandwidth/computing power requirements. And without a major breakthrough in ocean modelling science paradigms (to strongly reduce the amount of calculations and time to solution required), a simultaneous revolution affecting microprocessor industry (to reduce supercomputer energy consumption) and a code rewriting to fit the unconventional requirements of such energy-efficient chips, it is, from some point of view, not advisable to do so

Domaines

Climatologie
Fichier principal
Vignette du fichier
GLOBC-TR-Maisonnave-esiwace_gyre19-18.pdf (3.09 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04730870 , version 1 (10-10-2024)

Identifiants

  • HAL Id : hal-04730870 , version 1

Citer

Éric Maisonnave, Claire Lévy, Sébastien Masson. Implementation and computing performances of NEMO ORCA-KM configuration. [Technical Report] CECI, Université de Toulouse, CNRS, CERFACS, Toulouse, France - TR-CMGC-19-18. 2019. ⟨hal-04730870⟩
50 Consultations
4 Téléchargements

Partager

More