DFG implementation on multi GPU cluster with computation-communication overlap

Author

Huet, Sylvain ; Boulos, Vincent ; Fristot, Vincent ; Salvo, Luc

Author_Institution

GIPSA-Lab., UJF, Grenoble, France

fYear

2011

fDate

2-4 Nov. 2011

Firstpage

1

Lastpage

8

Abstract

Nowadays, it is possible to build a multi-GPU supercomputer, well suited for implementation of digital signal processing algorithms, for a few thousand dollars. However, to achieve the highest performance with this kind of architecture, the programmer has to focus on inter-processor communications, tasks synchronization ... In this paper, we propose a design flow allowing an efficient implementation of a Digital Signal Processing (DSP) application specified as a Data Flow Graph (DFG) on a multi GPU computer cluster. We focus particularly on the effective implementation of communications by automating the computation-communication overlap, which can lead to significant speedups as shown in the presented benchmark. The approach is validated on a 3D granulometry application developed for research on materials.

Keywords

coprocessors; data flow graphs; digital signal processing chips; 3D granulometry application; computation-communication overlap automation; data flow graph; design flow; digital signal processing algorithm; inter-processor communication; multiGPU computer cluster; multiGPU supercomputer; tasks synchronization; Computer architecture; Computers; Digital signal processing; Graphics processing unit; Instruction sets; Kernel; Synchronization;

fLanguage

English

Publisher

ieee

Conference_Titel

Design and Architectures for Signal and Image Processing (DASIP), 2011 Conference on

Conference_Location

Tampere

Print_ISBN

978-1-4577-0620-2

Electronic_ISBN

978-1-4577-0619-6

Type

conf

DOI

10.1109/DASIP.2011.6136859

Filename

6136859