메뉴 건너뛰기
소속 기관 / 학교 인증
인증하면 논문, 학술자료 등을  무료로 열람할 수 있어요.
한국대학교, 누리자동차, 시립도서관 등 나의 기관을 확인해보세요
(국내 대학 90% 이상 구독 중)
고객센터 ENG
주제분류

논문 기본 정보

저자정보
(Kyungpook National University)
저널정보
한국통신학회 한국통신학회논문지 한국통신학회논문지 제50권 제3호
오류 신고하기

피인용 0

검색

    초록·키워드

    Edge computing offers a promising solution to the latency issues inherent in centralized cloud processing, particularly for industrial Internet of Things (IIoT) applications. However, the limited computational capabilities of edge devices pose challenges to optimal artificial intelligence (AI) workload performance. This study provides a comparative performance analysis of several edge devices, focusing on evaluating the impact of hardware accelerators like graphics processing units (GPUs) on AI application processing. We employ YOLOv8, a popular object detection model, to evaluate five tasks―image classification, object detection, pose estimation, instance segmentation, and oriented bounding box detection―by measuring job completion time (JCT), GPU utilization, and memory usage. Our findings indicate that expensive high-end devices do not always provide a proportionate performance boost, with mid-range devices frequently offering comparable inference performance for less computationally demanding tasks. These results underscore the need for a careful balance between hardware specifications and application requirements to achieve efficient and cost-effective AI deployment. Additionally, we observe that multi-threading does not consistently yield performance improvements on edge devices due to Python’ s Global Interpreter Lock (GIL) overhead. This limitation highlights the need for innovative solutions, such as simultaneous task management and GPU scheduling, to improve parallelism and optimize resource utilization in edge environments.

    최근 본 자료 전체보기

      UCI(KEPA) : I410-151-25-02-092459463