Exa-MA: Methods and Algorithms for Exascale
Building the future of exascale computing through advanced numerical methods, AI-enhanced algorithms, and production-ready open-source software.
1. Our Mission
The Exa-MA project is part of France’s PEPR NumPEx initiative, dedicated to pushing the frontiers of exascale computing. We develop and deploy:
Advanced Methods
Novel discretizations, solvers, and algorithms optimized for exascale architectures
AI Integration
Scientific machine learning for simulation, model reduction, and optimization
Open Software
Production-ready libraries that abstract hardware complexity and enable seamless portability
Validated Impact
Demonstrators, benchmarks, and real-world applications across scientific domains
Exa-MA brings together leading French institutions (CEA, École Polytechnique, Inria, Sorbonne Université, Université de Strasbourg) with a budget of 6.255 M€ over the project duration.
2. Latest News & Updates
May 07, 2026 | COMET Bercy, Paris
NumPEx Operational Committee Meeting - Agenda: news and BoD feedback; mid-term evaluation covering programme KPIs, international and industrial relations, scientific production, and societal impact; training in NumPEx; software production, including the continuation of…
Jun 08-12, 2026 | IRMA, Université de Strasbourg
MAJSC 2026: ML & Autodiff in JAX - Workshop on Machine Learning and Automatic Differentiation in JAX for Scientific Computing. Five-day intensive program covering JAX fundamentals, neural networks (Equinox/Flax), automatic differentiation for PDEs (Diffrax)…
Stay informed about project milestones, publications, and community events:
-
News and Events – Latest announcements and upcoming activities
-
Deliverables – Reports, documentation, and public outputs
-
Results & Impact – Detailed metrics and scientific achievements
3. Recent Achievements
Latest Highlights
Robust Schwarz and GenEO-type preconditioners available in CPU-GPU PETSc workflows
Black-box optimization for chemistry workloads, validated on LUMI from 3 to 560 parameters
Impact at a Glance
46 Publications
16 Software Packages
18 Frameworks
55 Person-Years
4. Work Packages Overview
Exa-MA is organized into seven scientific work packages plus project management:
Numerical Methods
Discretization – Mesh generation/adaptation (AMR), finite element framework, nonconforming methods, parallel-in-time, multiphysics coupling |
|
Model Reduction & Scientific ML – Physics-driven deep learning, neural operators, data-driven ROM, reduced basis, multi-fidelity, super-resolution |
|
Solvers – Mixed precision Krylov, data compression, variable accuracy, saddle point systems, partitioned multiphysics |
Data & Decision
Inverse Problems & Data Assimilation – Deterministic/stochastic methods, observation strategies, multi-fidelity schedules, model updates |
|
Optimization – Combinatorial/continuous, surrogate-based, shape optimization, AutoML at exascale |
|
Uncertainty Quantification – Kernel-based sensitivity, UQ in PDE framework, surrogate modeling, acceleration on exascale |
Integration & Coordination
Showroom & Benchmarking – Testing/benchmarking environment, CI/CD, verification/validation, co-design coordination, training materials |
|
Project Management – Coordination, scientific/technical oversight, computing resources, communication, reporting |
5. Getting Started
Whether you’re a researcher, developer, or end user, find your entry point:
For Researchers |
Start with Results and Impact to see our latest publications and scientific contributions |
For Developers |
Visit the Software Stack to explore packages, APIs, and CI/CD status |
For Students |
Check Training & Events for workshops and educational resources |
For Partners |
Learn about Consortium Partners and collaboration opportunities |