Performance and Numerical Aspects of Decompositional Factorizations with FP64 Floating-Point Emulation in INT8

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Luszczek, Piotr, Gadepally, Vijay, Anderson, LaToya, Arcand, William, Bestor, David, Bergeron, William, Bonn, Alex, Burrill, Daniel J., Byun, Chansup, Houle, Michael, Hubbell, Matthew, Jananthan, Hayden, Jones, Michael, Michaleas, Peter, Morales, Guillermo, Mullen, Julia, Prout, Andrew, Reuther, Albert, Rosa, Antonio, Yee, Charles, Kepner, Jeremy
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911181993672704
author Luszczek, Piotr
Gadepally, Vijay
Anderson, LaToya
Arcand, William
Bestor, David
Bergeron, William
Bonn, Alex
Burrill, Daniel J.
Byun, Chansup
Houle, Michael
Hubbell, Matthew
Jananthan, Hayden
Jones, Michael
Michaleas, Peter
Morales, Guillermo
Mullen, Julia
Prout, Andrew
Reuther, Albert
Rosa, Antonio
Yee, Charles
Kepner, Jeremy
author_facet Luszczek, Piotr
Gadepally, Vijay
Anderson, LaToya
Arcand, William
Bestor, David
Bergeron, William
Bonn, Alex
Burrill, Daniel J.
Byun, Chansup
Houle, Michael
Hubbell, Matthew
Jananthan, Hayden
Jones, Michael
Michaleas, Peter
Morales, Guillermo
Mullen, Julia
Prout, Andrew
Reuther, Albert
Rosa, Antonio
Yee, Charles
Kepner, Jeremy
contents Mixing precisions for performance has been an ongoing trend as the modern hardware accelerators started including new, and mostly lower-precision, data formats. The advantage of using them is a great potential of performance gain and energy savings. The disadvantage are the numerical issues not present in the standard-mandated floating-point formats. Split integer emulation of FP64 takes this to an extreme with the computation performed only by fixed-point tensor core units. We present the new issues the emulation faces for practical cases involving dense linear solver. We show extensive numerical tests indicating the effect of extended numerical range of matrix entries. We also scaled the input sizes to study the performance and numerical profiles on the NVIDIA Hopper GPUs.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23565
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Performance and Numerical Aspects of Decompositional Factorizations with FP64 Floating-Point Emulation in INT8
Luszczek, Piotr
Gadepally, Vijay
Anderson, LaToya
Arcand, William
Bestor, David
Bergeron, William
Bonn, Alex
Burrill, Daniel J.
Byun, Chansup
Houle, Michael
Hubbell, Matthew
Jananthan, Hayden
Jones, Michael
Michaleas, Peter
Morales, Guillermo
Mullen, Julia
Prout, Andrew
Reuther, Albert
Rosa, Antonio
Yee, Charles
Kepner, Jeremy
Numerical Analysis
Mixing precisions for performance has been an ongoing trend as the modern hardware accelerators started including new, and mostly lower-precision, data formats. The advantage of using them is a great potential of performance gain and energy savings. The disadvantage are the numerical issues not present in the standard-mandated floating-point formats. Split integer emulation of FP64 takes this to an extreme with the computation performed only by fixed-point tensor core units. We present the new issues the emulation faces for practical cases involving dense linear solver. We show extensive numerical tests indicating the effect of extended numerical range of matrix entries. We also scaled the input sizes to study the performance and numerical profiles on the NVIDIA Hopper GPUs.
title Performance and Numerical Aspects of Decompositional Factorizations with FP64 Floating-Point Emulation in INT8
topic Numerical Analysis
url https://arxiv.org/abs/2509.23565