High-Performance Mixed-Precision Matrix Multiplication via Tile-Centric Design on Modern Architectures

Qiao Zhang; Rabab Alomairy; Dali Wang; Zhuowei Gu; Qinglei Cao

doi:10.1007/s42979-025-04575-0

High-Performance Mixed-Precision Matrix Multiplication via Tile-Centric Design on Modern Architectures

Qiao Zhang
, Rabab Alomairy
, Dali Wang
, Zhuowei Gu
, Qinglei Cao

Earth Systems Modeling

Research output: Contribution to journal › Article › peer-review

Abstract

General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research presents an adaptive mixed-precision GEMM framework that enables support for various precision formats at fine-grained tile and block levels, offering a reliable foundation for trustworthy mixed-precision computations. Furthermore, we leverage the PaRSEC runtime system to effectively balance workloads across diverse architectures. The performance exhibits strong scalability across both homogeneous platforms (Intel CPU-based systems and the ARM CPU-based Fugaku supercomputer) and heterogeneous systems (Nvidia V100, A100, and H100 GPU-based platforms, as well as the AMD GPU-based Frontier supercomputer). This work aims to improve computational efficiency and accuracy by bridging algorithmic innovations with hardware capabilities, fostering transformative advancements across a wide range of applications.

Original language	English
Article number	24
Journal	SN Computer Science
Volume	7
Issue number	1
DOIs	https://doi.org/10.1007/s42979-025-04575-0
State	Published - Jan 2026

Funding

This work was supported by the U.S. National Science Foundation under Award OAC-2451577. This research was supported by internal awards from Saint Louis University (Grant-0001651 and PROJ-000498), as well as by the U.S. National Science Foundation under Award OAC-2451577. For computational resources, this work utilized the compute node at the Innovative Computing Laboratory of the University of Tennessee, Knoxville, the Fugaku supercomputer at RIKEN, the Polaris supercomputer at Argonne National Laboratory, and the Frontier supercomputer at Oak Ridge National Laboratory.

Keywords

General matrix multiply
High-performance computing
Mixed precision
Task-based runtime

Access to Document

10.1007/s42979-025-04575-0

Cite this

@article{0673a7f5a6f14fe4b9547b04f7058199,

title = "High-Performance Mixed-Precision Matrix Multiplication via Tile-Centric Design on Modern Architectures",

abstract = "General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research presents an adaptive mixed-precision GEMM framework that enables support for various precision formats at fine-grained tile and block levels, offering a reliable foundation for trustworthy mixed-precision computations. Furthermore, we leverage the PaRSEC runtime system to effectively balance workloads across diverse architectures. The performance exhibits strong scalability across both homogeneous platforms (Intel CPU-based systems and the ARM CPU-based Fugaku supercomputer) and heterogeneous systems (Nvidia V100, A100, and H100 GPU-based platforms, as well as the AMD GPU-based Frontier supercomputer). This work aims to improve computational efficiency and accuracy by bridging algorithmic innovations with hardware capabilities, fostering transformative advancements across a wide range of applications.",

keywords = "General matrix multiply, High-performance computing, Mixed precision, Task-based runtime",

author = "Qiao Zhang and Rabab Alomairy and Dali Wang and Zhuowei Gu and Qinglei Cao",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.",

year = "2026",

month = jan,

doi = "10.1007/s42979-025-04575-0",

language = "English",

volume = "7",

journal = "SN Computer Science",

issn = "2662-995X",

publisher = "Springer",

number = "1",

}

TY - JOUR

T1 - High-Performance Mixed-Precision Matrix Multiplication via Tile-Centric Design on Modern Architectures

AU - Zhang, Qiao

AU - Alomairy, Rabab

AU - Wang, Dali

AU - Gu, Zhuowei

AU - Cao, Qinglei

N1 - Publisher Copyright: © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

PY - 2026/1

Y1 - 2026/1

N2 - General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research presents an adaptive mixed-precision GEMM framework that enables support for various precision formats at fine-grained tile and block levels, offering a reliable foundation for trustworthy mixed-precision computations. Furthermore, we leverage the PaRSEC runtime system to effectively balance workloads across diverse architectures. The performance exhibits strong scalability across both homogeneous platforms (Intel CPU-based systems and the ARM CPU-based Fugaku supercomputer) and heterogeneous systems (Nvidia V100, A100, and H100 GPU-based platforms, as well as the AMD GPU-based Frontier supercomputer). This work aims to improve computational efficiency and accuracy by bridging algorithmic innovations with hardware capabilities, fostering transformative advancements across a wide range of applications.

AB - General Matrix Multiplication (GEMM) is a critical operation underpinning a wide range of applications in high-performance computing (HPC) and artificial intelligence (AI). The emergence of hardware optimized for low-precision arithmetic necessitates a reevaluation of numerical algorithms to leverage mixed-precision computations, achieving improved performance and energy efficiency. This research presents an adaptive mixed-precision GEMM framework that enables support for various precision formats at fine-grained tile and block levels, offering a reliable foundation for trustworthy mixed-precision computations. Furthermore, we leverage the PaRSEC runtime system to effectively balance workloads across diverse architectures. The performance exhibits strong scalability across both homogeneous platforms (Intel CPU-based systems and the ARM CPU-based Fugaku supercomputer) and heterogeneous systems (Nvidia V100, A100, and H100 GPU-based platforms, as well as the AMD GPU-based Frontier supercomputer). This work aims to improve computational efficiency and accuracy by bridging algorithmic innovations with hardware capabilities, fostering transformative advancements across a wide range of applications.

KW - General matrix multiply

KW - High-performance computing

KW - Mixed precision

KW - Task-based runtime

UR - https://www.scopus.com/pages/publications/105025355009

U2 - 10.1007/s42979-025-04575-0

DO - 10.1007/s42979-025-04575-0

M3 - Article

AN - SCOPUS:105025355009

SN - 2662-995X

VL - 7

JO - SN Computer Science

JF - SN Computer Science

IS - 1

M1 - 24

ER -

High-Performance Mixed-Precision Matrix Multiplication via Tile-Centric Design on Modern Architectures

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this