Dear colleagues, 

We are delighted to send you this first newsletter from the Digital for Exascale research program (PEPR), also known as NumPEx (Numérique pour l’Exascale in French). Launched on June 24, 2023, this ambitious program aims to contribute to the design and development of numerical methods and software components that will equip future European Exascale and post-Exascale machines. NumPEx also aims to help scientific and industrial applications exploit their full potential. The program is supported by the investment plan France 2030 to the tune of €41 million. Managed by the CEA, the CNRS and Inria research organizations, NumPEx involves 80 research teams and bringing together over 500 researchers.

One of the main drivers of the NumPEx program is the current paradigm shift in high performance computing (HPC) system architectures, with the rapid emergence of new technologies and uses (e.g., the digital continuum and AI), which means that the HPC software stack must be developed and adapted for future Exascale supercomputers. Scientific and industrial applications also need to prepare for post-Exascale systems and uses. The utilization of such supercomputers by academic and industrial researchers, with its potential for innovation and competitiveness, requires considerable work in adapting and porting codes, both for simulation and data analysis. Adapting applications to these new architectures is a major scientific and technological challenge, requiring major research efforts.

In the era of Exascale, mastering the software stack of supercomputers is a scientific and technological challenge, as well as a matter of national and European sovereignty. To achieve this goal, NumPEx will contribute to the construction of a software stack that will enable full exploitation of the capabilities of the future Exascale machine “Alice Recoque” of the Jules Verne project.

As you will discover in this newsletter, the first year of the program saw the launch of his six projects, which are now fully operational. Their news is already rich in terms of results and initiatives both nationally and internationally, demonstrating the vitality of the community. In this respect, NumPEx is forging strong links with our European, American and Japanese colleagues through the InPEx conference series launched last June in Sietges, Spain. The key role played by application co-design in NumPEx is reflected in the organization of a series of thematic workshops, which will involve the scientific communities in the program and enable them to co-develop the software building blocks that will meet their needs.

In the coming weeks, the NumPEx agenda will be rich with the organization of the NumPEx General Assembly in March 2025. To complete NumPEx's activities and extend the community to new academic players, a call for projects will be published between now and the beginning of 2025, with selected themes on Artificial Intelligence, programming models for accelerated architectures and the development of workflows for scientific data analysis. These events and initiatives are unique opportunities to join us and contribute to NumPEx.

In the meantime, happy reading!

The program directors

Exa-MA: a key player of the SciML 2024 Integrating Machine Learning with Scientific Computing Conference

SciML 2024, held in Strasbourg, was a key event for the scientific computing and machine learning communities, with Exa-MA as a funding partner. The conference focused on themes central to Exa-MA’s Work Package 2 (WP2), including scientific machine learning and reduced-order modeling. Organized by Emmanuel Franck and Victor Michel-Dansac, both key contributors to WP2, the event aligned with Exa-MA's goal of advancing scientific computing for exascale systems.

Exa-SofT: light-weight prediction for improving energy consumption in HPC platforms

As computing power demand grows, managing energy use is a challenge for the HPC community. With his co-authors, the member of Exa-SofT Georges Da Costa propose a method to predict task power consumption on supercomputers to schedule jobs more efficiently, reducing energy use without impacting performance.

Exa-DoST: a second meeting to prepare the future

The second annual meeting of the NumPEx Exa-DoST project took place at Inria, in Rennes, on 18-19 September 2024. This two days were an opportunity to share the first outcomes of the projet and discuss the upcoming research.

Exa-AtoW: a collaboration with Data Terra

The Exa-AToW project focuses on large-scale workflows deployed across multiple infrastructures, including data centers, HPC facilities, and cloud environments. In collaboration with the Gaia Data project, the initiative emphasizes the development of Machine Actionable Data Project Plans (MADPP). These plans are designed to capture and manage the complex set of constraints related to data logistics and the computational resource requirements needed to ensure efficient workflow execution.

Exa-DI: the Computational and data science Team (CDT) is growing

The goal of the CDT is to implement the Exa-DI co-design and co-development process, working with scientific and industrial application teams to define mini-apps that address exascale application challenges. These mini-apps will be developed by bringing together software components provided by other NumPEx research teams and package them to deliver software development kits. In the first semester, two engineers and a program manager were recruited, followed by two more engineers this quarter, with one recruitment still ongoing and more interviews to come.

Software

Feel++ Release V111 preview.10
Christophe Prud'homme, Vincent Chabannes, Thomas Saigre, et al.
This software release introduces significant improvements and extended platform support for Feel++, an open-source computational library widely used in scientific simulations.

Publications

Nonlinear compressive reduced basis approximation for multi-parameter elliptic problems
Hassan Ballout, Yvon Maday, Christophe Prud'homme. 2024

Error analysis of matric multiplication with narrow range floating-point arithmetic
Théo Mary, Mantas Mikaitis. 2024

Prediction and interpretability of HPC I/O ResourcesNonlinear compressive reduced basis approximation for multi-parameter elliptic problems 
Alexis Bandet, Francieli Boito, Guillaume Pallez. 2024

Volume-preserving geometric shape optimization of the Dirichlet energy using variational neural network
Amaury Bélières--Frendo, Emmanuel Franck, Victor Michel-Dansac, Yannick Privat. 2024

This email has been sent by the Centre national de la recherche scientifique (CNRS), one of the three pilots of the Digital for Exascale (NumPEx) research program (PEPR).
Contact: https://numpex.naospowered1.wpenginepowered.com/contact-us/
 The NumPEx research program is committed to ensure the protection of the collected personal data. Find out more about how NumPEx processes your data here.