Numerical health check of industrial simulation codes from HPC environments to new hardware technologies

Authors:
Christophe Denis
Affiliations:
EDF Research and Development, SINETICS Department, Clamart Cedex, France
Venue:
PPAM'09 Proceedings of the 8th international conference on Parallel processing and applied mathematics: Part II
Year:
2009

Citing 7
Cited 0

The accuracy of floating point summation

SIAM Journal on Scientific Computing
Experiments on the evaluation of functional ranges using a random interval arithmetic

Mathematics and Computers in Simulation
Accuracy and Stability of Numerical Algorithms

Accuracy and Stability of Numerical Algorithms
Accuracy and Reliability in Scientific Computing (Software, Environments, Tools) (Software, Environments, Tools)

Accuracy and Reliability in Scientific Computing (Software, Environments, Tools) (Software, Environments, Tools)
Hydrodynamics of Free Surface Flows: Modelling with the finite element method

Hydrodynamics of Free Surface Flows: Modelling with the finite element method
Evaluation Ranges of Functions using Balanced Random Interval Arithmetic

Informatica
Accurate Floating-Point Summation Part I: Faithful Rounding

SIAM Journal on Scientific Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

The numerical health check of industrial codes is crucial to give confidence about the computed results performed by studying the round-off error propagation. This problem is exacerbated in a supercomputing environment where trillions of floating-point operations may be performed every second. A parallel program based on domain decomposition as shown in this paper could compute slightly different results depending on the number of processors. This numerical health check is also needed to verify if a numerical code (or some parts of the numerical code) could still have an acceptable accuracy when using single precision instead of double precision which is useful to run numerical codes on new hardware technologies like GPU where the double precision is unavailable or expensive. The round-off error propagation is measured with the MPFI (interval arithmetic approach) and CADNA (probabilistic approach) libraries.