Analysis on Coherent Memory Systems Within AI Training Workloads

Ziang Zhao

2024

Abstract

Within larger groups of multiprocessors and heterogeneous computing clusters, the problem of data coherence has been of increasing concern. As the computational devices become increasingly complex, so do the methods that were being used to ensure the data integrity. With the recent rise in the training requirements of Artificial intelligence systems, the demand for larger coherent data systems has risen significantly. This paper aimed to provide an overview and analysis of different data coherence methods and analyze their potential performance within an AI training workload, and concluded that the Artificial intelligence (AI) training models would often require memory systems to be efficient over access of training parameters over extended periods of time and ensure its reliability across the extended training process. The paper would mostly contain a theoretical analysis of different coherence methods and would aim at providing an upper limit of the performance of these different methods. Further research regarding the physical implementation of such coherent systems might be still required.

Paper Citation

in Harvard Style

Zhao Z. (2024). Analysis on Coherent Memory Systems Within AI Training Workloads. In Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-754-2, SciTePress, pages 227-231. DOI: 10.5220/0013514800004619

in Bibtex Style

@conference{daml24,
author={Ziang Zhao},
title={Analysis on Coherent Memory Systems Within AI Training Workloads},
booktitle={Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2024},
pages={227-231},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013514800004619},
isbn={978-989-758-754-2},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - Analysis on Coherent Memory Systems Within AI Training Workloads
SN - 978-989-758-754-2
AU - Zhao Z.
PY - 2024
SP - 227
EP - 231
DO - 10.5220/0013514800004619
PB - SciTePress