Communication-Avoiding SpGEMM via Trident Partitioning on Hierarchical GPU Interconnects Conference: 2026 International Conference on Supercomputing; July 06–09, 2026; Belfast, United Kingdom 2026 International Conference on Supercomputing (ICS ’26), July 06–09, 2026, Belfast, United Kingdom DOI: 10.1145/3797905.3800543 ISBN: 979-8-4007-2522-7/2026/07

Julian Bellavita, Lorenzo Pichetti, Thomas Pasquali, Flavio Vella, Giulia Guidi · 2026

Read Communication-Avoiding SpGEMM via Trident Partitioning on Hierarchical GPU Interconnects Conference: 2026 International Conference on Supercomputing; July 06–09, 2026; Belfast, United Kingdom 2026 International Conference on Supercomputing (ICS ’26), July 06–09, 2026, Belfast, United Kingdom DOI: 10.1145/3797905.3800543 ISBN: 979-8-4007-2522-7/2026/07 by Julian Bellavita, Lorenzo Pichetti, Thomas Pasquali, Flavio Vella, Giulia Guidi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-21444.txt.

Description

Bellavita and co-authors speed up sparse matrix multiplication, a core kernel in science and data work, on GPU supercomputers. In distributed runs, moving and merging partial products across nodes eats much of the time. Their Trident algorithm splits work with a hierarchy-aware scheme that leans on fast links inside a node. It also overlaps communication with computation. On the Perlmutter supercomputer, Trident ran up to 2.38 times faster than a standard 2D method and halved internode traffic.

Catalog record

CREATORJulian Bellavita, Lorenzo Pichetti, Thomas Pasquali, Flavio Vella, Giulia Guidi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61956
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.21444

Browse the full library · About the Tower of Babel