Core Concepts
A3PIM proposes an automated, analytic, and accurate offloading framework for CPU-PIM systems, achieving significant speedups compared to CPU-only and PIM-only executions.
Abstract
The performance gap between memory and processor is a major concern in modern computer systems.
Processing-in-Memory (PIM) architecture aims to reduce data movements by integrating computational units close to memory.
A3PIM utilizes static code analysis to identify PIM-friendly segments and achieve speedups in real-world workloads.
Evaluation shows A3PIM outperforms traditional methods and approaches theoretical peak performance.
A3PIM addresses data movement overhead and context switch costs to optimize task distribution in CPU-PIM systems.
Stats
A3PIM은 CPU-only 및 PIM-only 실행과 비교하여 평균 속도 향상을 달성합니다.
A3PIM은 실제 워크로드에서 속도 향상을 실현합니다.
A3PIM은 데이터 이동 오버헤드와 컨텍스트 스위치 비용을 해결하여 CPU-PIM 시스템에서 작업 분배를 최적화합니다.
Quotes
"A3PIM proposes an automated, analytic, and accurate offloading framework for CPU-PIM systems."
"Evaluation shows A3PIM outperforms traditional methods and approaches theoretical peak performance."