MUMPS Workshop

Linear solvers on modern computing architectures:
exploiting GPU acceleration and mixed precision

June 25, 2025, Montpellier, France

In collaboration with CINES (Computing Centre for Higher Education), in Montpellier (France), we organize a workshop on the latest advances in exploiting GPU acceleration and mixed precision arithmetic.


Registration

Registration is free of charge but mandatory. The deadline for registering is 23rd May but it would greatly help our planning if you could register before that. Please register using the registration form.
Please note that if you wish to visit the Adastra supercomputer, you need to show an identity card.


Program

13:15-14:30 Welcome
13:30-14:00 Visit of the Adastra supercomputer
14:00-14:30 Mumps Tech
MUMPS: MUltifrontal Massively Parallel Solver for the direct solution of sparse linear equations
14:30-15:15 Gabriel HAUTREUX (CINES, France)
Adastra: an exascal architecture for national research in AI and HPC
15:15-15:45 Coffee Break
15:45-16:30 Thierry GAUTIER and Pierre-Etienne POLET (Inria-LIP, ENS Lyon, France)
On the Use of APU Architectures in MUMPS / XKBlas
16:30-16:45 Théo MARY (CNRS-LIP6, Sorbonne University, France)
Mixed Precision Algorithms in Numerical Linear Algebra
16:45-17:15 Antoine JEGO (CNRS-LIP6, Sorbonne University, France)
BLAS-based Block Memory Accessors with Applications to Mixed-Precision Sparse Direct Solvers,