• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

Çмú´ëȸ ÇÁ·Î½Ãµù

Ȩ Ȩ > ¿¬±¸¹®Çå > Çмú´ëȸ ÇÁ·Î½Ãµù > Çѱ¹Á¤º¸°úÇÐȸ Çмú´ëȸ > KSC 2020

KSC 2020

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) Acceleration of Deep Learning Applications by Pipelining on NVIDIA Jetson AGX Xavier
¿µ¹®Á¦¸ñ(English Title) Acceleration of Deep Learning Applications by Pipelining on NVIDIA Jetson AGX Xavier
ÀúÀÚ(Author) Samnieng Tan   EunJin Jeong   Jangryul Kim   Jaeseong Lee   Soonhoi Ha  
¿ø¹®¼ö·Ïó(Citation) VOL 47 NO. 02 PP. 0982 ~ 0984 (2020. 12)
Çѱ۳»¿ë
(Korean Abstract)
¿µ¹®³»¿ë
(English Abstract)
Since the need of deep learning applications is increasing in embedded systems, non-CPU processing elements are equipped on an embedded device to accelerate the applications. NVIDIA Jetson AGX Xavier (Xavier) is a representative example which has not only an octa-core CPU but also one powerful GPU and two deep learning accelerators to enhance the performance of deep learning inference on resource-constrained environments. Even though an embedded device provides heterogeneous processing elements, utilizing diverse computation units is burdensome to increase performance. In this paper, we propose a technique that combines multiple existing methods and our proposed network pipelining methods to utilize heterogeneous processing elements on Xavier to maximize the throughput of deep learning applications. We experimented with image classification and object detection examples and observed up to 490% FPS improvement compared to the baseline FPS with a single GPU.
Å°¿öµå(Keyword)
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå