WarmWorld at the Berlin Summit for EVE

by Hendryk Bockelmannm, Claudia Frauen and Christopher Kadow
2023-07-03

The Berlin Summit for EVE is taking place from July, 3rd - 7th, at the Harnack House in Berlin. It brings together invited participants from across the world to draft a blueprint for EVE, presently envisioned as an international digital infrastructure that exploits the latest advances in information technology (HPC and AI) to establish and maintain a global climate prediction and information system across regional partners or nodes. The leading force behind it is WarmWorld PI Bjorn Stevens and the event is co-sponsored by WarmWorld. Scientists from all of the WarmWorld modules are also participating in the event in order to shape the concept of EVE and include our experiences and concepts from WarmWorld. Over the coming days we will report here on the different aspects of EVE and the related discussions at the summit.

An important topic on day one is the changing landscape of technology. The scene was set by a vision talk by NVIDIA founder and CEO Jensen Huang, which was followed by presentations by Satoshi Matsuoka (RIKEN Center for Computational Science), Nicolas Dube (Hewlett Packard Enterprise), and Andrew Jones (Microsoft) talking about exascale computers and how to make good use of them.

With the emergence of pre-exascale HPC systems like Fugaku and LUMI and the first exascale supercomputer Frontier HPC recently crossed a threshold that now enables global km-scale climate simulations. A first example of such a simulation is the 1.25km global ICON simulation reproducing the famous Blue Marble photograph. The data from this simulation is also used by NVIDIA as the basis for their interactive visualisation tools based on Omniverse, which Jensen Huang introduced in his presentation. He also showed the potential of using the new NVIDIA Grace Hopper chip for a global coupled 2.5km ICON simulation, which is projected to achieve up to 200x the throughput compared to a CPU-based system with the same energy consumption of 1 MW, highlighting the benefits of GPU accelerated computing especially for energy efficiency.

Jensen Huang from NVIDIA presents the potential of using the new NVIDIA Grace Hopper chip for a 2.5km global ICON simulation.

The talk by Nicolas Dube also highlighted two important aspects of the exascale era: performance portability and scalable development. Both align well with WarmWorld Faster working towards the modularisation of the ICON code. It culminates in the statement that the internet of workflows will accelerate innovation in the 2nd half of the decade, which underlines the importance of WarmWorld Easier efforts.

AI/ML acceleration in science was omnipresent with examples from climate but also other disciplines. Digital twins and their calculation in flexible realms to many scales are perfect examples for dedicated applications in km-scale climate simulations. Urban or postcode scale focus and modelling systems like FourcastNet embedded in Earth2 visualization engines are glimpses of what could be possible. Even centers like ICTP showed that AI/ML moves into theoretical fields. The extraction of complex information via AI/ML will influence all climate related fields. The optimism around recent technology developments (e.g. ChatGPT) showed that smart software design can make a huge difference on interfaces potentially needed in EVE. Of course these developments could inspire WarmWorld Smarter projects soon to be kicked off.