MENU

Berlin summit to create a ‘CERN for climate modelling’

Berlin summit to create a ‘CERN for climate modelling’

Feature articles |
By Nick Flaherty

Cette publication existe aussi en Français


A summit in Berlin this week is aiming to boost the modelling of the climate of Earth, including a digital twin of the planet and AI accelerated supercomputers, in a move that has been compared to the impact of the CERN particle research centre.

The Earth Virtualization Engines (EVE) project is a digital infrastructure for climate modelling that exploits the latest advances in high performance computing (HPC) and artificial intelligence (AI) architectures, particularly from Nvidia.

This will combine the Grace Hopper combined CPU and GPU processors with new AI models and a digital twin, says Jensen Huang, CEO of Nvidia at the conference today.

“EVE is an incredible vision. Aside from the climate research community and people who can conceptualise long term impact into near term actions, very few people understand what it means until it happens to them, whether floods, forest fires or droughts,” said Huang. “It seems like California is always on fire. The vision of EVE is almost unattainable with today’s technology.”

“EVE’s vision to understand future climate impacts at kilometer scale is a grand challenge that demands simultaneous breakthroughs in modelling, computing, and AI. Nvidia is working with the EVE community to provide a path to simulate and visualize the global atmosphere at unprecedented speed and scale, using cutting-edge AI for physics simulation, accelerated scientific computing and advanced data infrastructure.”

“It has taken us almost 10 years to build the computer for this,” he said.“Grace Hopper took us a decade to develop and makes it possible for us to accelerate almost any software. The first application of Grace Hopper is our climate software and the amount of performance is fairly insane.”

“Grace Hopper can perform data analytics, graph neural networks and LLM [large language model generative AI] algorithms substantially faster with much better energy efficiency, for example we can run the Llama algorithm with 200x better energy efficiency.”

“However this is not your normal computer and requires the software to be rewritten from the ground up,” he said.  

The aim is for EVE to meet the need to produce more accurate and detailed climate predictions at regional and local scales globally, including predictions of the likelihood of extreme weather events, with the models freely accessible to everyone. No such capability exists or is foreseen from existing national climate initiatives say the campaigners.

The blueprint for EVE will be discussed and amended at the Open Science Conference of the World Climate Research Programme in Kigali, Rwanda, in October 2023, so as to better engage the global south. It will then be presented at the COP28 climate conference in November 2023.

However EVE will almost certainly not be a single physical centre. The whole point of the Berlin Summit and subsequent discussions in Kigali will be to decide how to organize the initiative.

“For the general public CERN is a symbol for how international scientific cooperation can expand human comprehension in ways that would otherwise be impossible. The evocation of a CERN for Climate Science, appeals to the idea of CERN as a symbol of international cooperation to solve pressing problems and as an engine for innovation. In recent years other modes of international scientific cooperation have emerged, but these have less public resonance, hence the frequent references to CERN,” say the researchers.

AI for climate modelling 

The key is using new generative AI algorithms for physics alongside existing HPC climate modelling to accelerate the use of the data. The generative AI can take data from the checkpoints in the climate simulation and allow more access to all kinds of data.

“There are three technologies that we are building in service of EVE and climate science. Accelerated computing, AI adapted for climate simulation. Second is new AI software that learns the laws of physics and climate. The third is a digital twin technology on a planetary scale that can ingest a massive amount of data for people to interact with at the cloud level,” said Huang.

“The physical model and AI model interact and the more one approach improves it takes the pressure off the other approach,” said Huang. “We use AI to generate more members of an ensemble for predicting extreme events to give deeper visibility into extreme weather and more time to prepare. AI gives you three orders of magnitude sped up in capability,” he added.

This is a key project for the company, says Huang.

“We have a couple of hundred people in the company leveraging the work of the rest of the company to tackle a very challenging problem I have no clue as to the business opportunity and it doesn’t matter. When we first started working on video games or AI they were also zero billion dollars.”   

eve4climate; www.nvidia.com

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News

Share:

Linked Articles
10s