Why is FPGA-GPU Heterogeneity the Best Option for Embedded Deep Neural Networks? - Archive ouverte HAL Access content directly
Preprints, Working Papers, ... Year :

Why is FPGA-GPU Heterogeneity the Best Option for Embedded Deep Neural Networks?

Walther Carballo-Hernández
  • Function : Author
  • PersonId : 181326
  • IdHAL : walcarher
François Berry

Abstract

Graphics Processing Units (GPUs) are currently the dominating programmable architecture for Deep Learning (DL) accelerators. The adoption of Field Programmable Gate Arrays (FPGAs) in DL accelerators is however getting momentum. In this paper, we demonstrate that Direct Hardware Mapping (DHM) of a Convolutional Neural Network (CNN) on an embedded FPGA substantially outperforms a GPU implementation in terms of energy efficiency and execution time. However, DHM is highly resource intensive and cannot fully substitute the GPU when implementing a state-of-the-art CNN. We thus propose a hybrid FPGA-GPU DL acceleration method and demonstrate that heterogeneous acceleration outperforms GPU acceleration even including communication overheads. Experimental results are conducted on a heterogeneous multi-platform setup embedding an Nvidia(R) Jetson TX2 CPU-GPU board and an Intel(R) Cyclone10GX FPGA board. The SqueezeNet, MobileNetv2, and ShuffleNetv2 mobile-oriented CNNs are experimented. We show that heterogeneous FPG-AGPU acceleration outperforms GPU acceleration for classification inference task over MobileNetv2 (12%-30% energy reduction, 4% to 26% latency reduction), SqueezeNet (21%-28% energy reduction, same latency), and ShuffleNetv2 (25% energy reduction, 21% latency reduction).
Fichier principal
Vignette du fichier
DATE_SLOHA_2021_CameraReady.pdf (2.78 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03135114 , version 1 (08-02-2021)

Identifiers

Cite

Walther Carballo-Hernández, Maxime Pelcat, François Berry. Why is FPGA-GPU Heterogeneity the Best Option for Embedded Deep Neural Networks?. 2021. ⟨hal-03135114⟩
105 View
309 Download

Altmetric

Share

Gmail Facebook Twitter LinkedIn More