Impact of mixed-precision on PETSc solvers and preconditioners

Provide a benchmark framework to study the effects of mixed precision on PETSc Krylov solvers and preconditioners in terms of performance, convergence, and numerical accuracy.

Quick Facts

Type

Mini App

Status

Planned

Work Packages

WP3

Frameworks

PETSc, HPDDM, MUMPS

1. Overview

Provide a benchmark framework to study the effects of mixed precision on PETSc Krylov solvers and preconditioners in terms of performance, convergence, and numerical accuracy.

2. Technical Stack

2.1. Frameworks Used

  • PETSc

  • HPDDM

  • MUMPS

2.2. Parallel Frameworks

  • MPI

  • Multithread - OpenMP

3. Methods & Algorithms

3.1. WP3: Solvers

  • direct solver

  • multi-precision

  • domain decomposition methods

  • krylov solver

  • preconditioning

4. Data Flow

4.1. Inputs

  • PETSc binary

  • Matrix Market

4.2. Outputs

  • Residual history

  • CSV logs

5. Benchmarking

5.1. Metrics

  • Time To Solution

  • Energy To Solution

5.2. Benchmark Scope

  • Solver Scaling

6. Timeline

Milestone Date

Specification Due

N/A

Prototype Due

N/A

7. Team

Partners: Sorbonne U, Inria PARIS

Responsible: P. Jolivet