Beyond VABlock: Improving Transformer workloads through aggressive prefetching
Authors:
- Jane RheeJRJane Rhee
- Ikyoung ChoiICIkyoung Choi
- Gunjae KooGKGunjae Koo
- Yunho OhYOYunho Oh
- Myung Kuk YoonMYMyung Kuk Yoon
- 4 March 2025
- journal article
- Published by Elsevier in Journal of Systems Architecture
- Vol. 162
Abstract
No abstract availableFunding Information
- Institute for Information and Communications Technology Promotion (RS-2022-00155966)
- National Research Foundation of Korea
- Ewha Womans University
This publication has 18 references indexed in Scilit:
- Early-Adaptor: An Adaptive Framework forProactive UVM Memory Management2023
- UVM Discard: Eliminating Redundant Memory Transfers for Accelerators2022
- Raptor: Mitigating CPU-GPU False Sharing Under Unified Memory Systems2022
- The Advance of Deep Learning and Attention Mechanism2022
- Demystifying GPU UVM Cost with Deep Runtime and Workload Analysis2021
- Accelerating Towards Larger Deep Learning Models and Datasets – A System Platform View Point2020
- Coordinated Page Prefetch and Eviction for Memory Oversubscription Management in GPUs2020
- Simultaneous branch and warp interweaving for sustained GPU performance2012
- GPU Pathfinding Optimization2011
- Rodinia: A benchmark suite for heterogeneous computing2009