Convolutional Neural Networks on Dataflow Engines

File Description SizeFormat 
iccd18nv.pdfAccepted version196.05 kBAdobe PDFView/Open
Title: Convolutional Neural Networks on Dataflow Engines
Authors: Voss, N
Bacis, M
Mencer, O
Gaydadjiev, G
Luk, W
Item Type: Conference Paper
Abstract: In this paper we discuss a high performance implementation for Convolutional Neural Networks (CNNs) inference on the latest generation of Dataflow Engines (DFEs). We discuss the architectural choices made during the design phase taking into account the DFE chip properties. We then perform design space exploration, considering the memory bandwidth and resources utilisation constraints derived from the used DFE and the chosen architecture. Finally, we discuss the high performance implementation and compare the obtained performance against other implementations, showing that our proposed design reaches 2,450 GOPS when running VGG16 as a test case.
Issue Date: 23-Nov-2017
Date of Acceptance: 5-Nov-2017
URI: http://hdl.handle.net/10044/1/61591
DOI: https://dx.doi.org/10.1109/ICCD.2017.77
ISSN: 1063-6404
Publisher: IEEE
Start Page: 435
End Page: 438
Journal / Book Title: 2017 IEEE 35TH INTERNATIONAL CONFERENCE ON COMPUTER DESIGN (ICCD)
Copyright Statement: © 2017 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Conference Name: 35th IEEE International Conference on Computer Design (ICCD)
Keywords: Science & Technology
Technology
Computer Science, Hardware & Architecture
Engineering, Electrical & Electronic
Computer Science
Engineering
Publication Status: Published
Start Date: 2017-11-05
Finish Date: 2017-11-08
Conference Place: Boston, MA
Online Publication Date: 2017-11-23
Appears in Collections:Faculty of Engineering
Computing



Items in Spiral are protected by copyright, with all rights reserved, unless otherwise indicated.

Creative Commons