원문정보
Performance Analysis of Prefetching Techniques on Processing-in- Memory System
초록
영어
The Processing-in-Memory (PIM) paradigm is a non-von Neumann computer architecture where processing and memory functions are combined in a single location to overcome the effects of the memory wall. Prefetching is a technique used to anticipate data access patterns and proactively bring data into the processor cache to reduce memory latency. In this paper, we aim to provide insights into applicable prefetch techniques for PIM by analyzing the performance of commonly applicable prefetch techniques in a simulated PIM system. We simulate PIM systems with (i) no prefetching, (ii) next-line prefetching, (iii) stream prefetching and (iv) the Variable Length Data Prefetcher (VLDP) using a custom-built PIM simulator and the STREAM benchmark as the workload and compare their performance.
목차
1. Introduction
1.1. Processing-in-Memory
1.2. Prefetching
2. Objectives and Methodology
2.1. Dataset
2.2. Experiment setup
3. Experiment result
4. Conclusions
Acknowledgement
References