Understanding Performance Differences of FPGAs and GPUs

Jason Cong; Zhenman Fang; Michael Lo; Hanrui Wang; Jingxian Xu; Shaochong Zhang

doi:10.1109/FCCM.2018.00023

Understanding Performance Differences of FPGAs and GPUs

Jason Cong
, Zhenman Fang
, Michael Lo
, Hanrui Wang
, Jingxian Xu
, Shaochong Zhang

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

120 Scopus citations

Abstract

This paper aims to better understand the performance differences between FPGAs and GPUs. We intentionally begin with a widely used GPU-friendly benchmark suite, Rodinia, and port 15 of the kernels onto FPGAs using HLS C. Then we propose an analytical model to compare their performance. We find that for 6 out of the 15 ported kernels, today's FPGAs can provide comparable performance or even achieve better performance than the GPU, while consuming an average of 28% of the GPU power. Besides lower clock frequency, FPGAs usually achieve a higher number of operations per cycle in each customized deep pipeline, but lower effective parallel factor due to the far lower off-chip memory bandwidth. With 4x more memory bandwidth, 8 out of the 15 FPGA kernels are projected to achieve at least half of the GPU kernel performance.

Original language	English (US)
Title of host publication	Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	93-96
Number of pages	4
ISBN (Electronic)	9781538655221
DOIs	https://doi.org/10.1109/FCCM.2018.00023
State	Published - Sep 7 2018
Externally published	Yes
Event	26th Annual IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018 - Boulder, United States Duration: Apr 29 2018 → May 1 2018

Publication series

Name	Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018

Conference

Conference	26th Annual IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018
Country/Territory	United States
City	Boulder
Period	4/29/18 → 5/1/18

Bibliographical note

Publisher Copyright:
© 2018 IEEE.

Keywords

Analytical model
FPGA
GPU
Performance comparison

Publisher link

10.1109/FCCM.2018.00023

Find It

Find It at U of M Libraries

Cite this

Cong, J., Fang, Z., Lo, M., Wang, H., Xu, J., & Zhang, S. (2018). Understanding Performance Differences of FPGAs and GPUs. In Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018 (pp. 93-96). Article 8457638 (Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/FCCM.2018.00023

Understanding Performance Differences of FPGAs and GPUs. / Cong, Jason; Fang, Zhenman; Lo, Michael et al.
Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018. Institute of Electrical and Electronics Engineers Inc., 2018. p. 93-96 8457638 (Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Cong, J, Fang, Z, Lo, M, Wang, H, Xu, J & Zhang, S 2018, Understanding Performance Differences of FPGAs and GPUs. in Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018., 8457638, Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018, Institute of Electrical and Electronics Engineers Inc., pp. 93-96, 26th Annual IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018, Boulder, United States, 4/29/18. https://doi.org/10.1109/FCCM.2018.00023

Cong J, Fang Z, Lo M, Wang H, Xu J, Zhang S. Understanding Performance Differences of FPGAs and GPUs. In Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018. Institute of Electrical and Electronics Engineers Inc. 2018. p. 93-96. 8457638. (Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018). doi: 10.1109/FCCM.2018.00023

Cong, Jason ; Fang, Zhenman ; Lo, Michael et al. / Understanding Performance Differences of FPGAs and GPUs. Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018. Institute of Electrical and Electronics Engineers Inc., 2018. pp. 93-96 (Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018).

@inproceedings{c08a2cd2c69a4d04bb0569f00c018f46,

title = "Understanding Performance Differences of FPGAs and GPUs",

abstract = "This paper aims to better understand the performance differences between FPGAs and GPUs. We intentionally begin with a widely used GPU-friendly benchmark suite, Rodinia, and port 15 of the kernels onto FPGAs using HLS C. Then we propose an analytical model to compare their performance. We find that for 6 out of the 15 ported kernels, today's FPGAs can provide comparable performance or even achieve better performance than the GPU, while consuming an average of 28\% of the GPU power. Besides lower clock frequency, FPGAs usually achieve a higher number of operations per cycle in each customized deep pipeline, but lower effective parallel factor due to the far lower off-chip memory bandwidth. With 4x more memory bandwidth, 8 out of the 15 FPGA kernels are projected to achieve at least half of the GPU kernel performance.",

keywords = "Analytical model, FPGA, GPU, Performance comparison",

author = "Jason Cong and Zhenman Fang and Michael Lo and Hanrui Wang and Jingxian Xu and Shaochong Zhang",

note = "Publisher Copyright: {\textcopyright} 2018 IEEE.; 26th Annual IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018 ; Conference date: 29-04-2018 Through 01-05-2018",

year = "2018",

month = sep,

day = "7",

doi = "10.1109/FCCM.2018.00023",

language = "English (US)",

series = "Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "93--96",

booktitle = "Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018",

}

TY - GEN

T1 - Understanding Performance Differences of FPGAs and GPUs

AU - Cong, Jason

AU - Fang, Zhenman

AU - Lo, Michael

AU - Wang, Hanrui

AU - Xu, Jingxian

AU - Zhang, Shaochong

PY - 2018/9/7

Y1 - 2018/9/7

N2 - This paper aims to better understand the performance differences between FPGAs and GPUs. We intentionally begin with a widely used GPU-friendly benchmark suite, Rodinia, and port 15 of the kernels onto FPGAs using HLS C. Then we propose an analytical model to compare their performance. We find that for 6 out of the 15 ported kernels, today's FPGAs can provide comparable performance or even achieve better performance than the GPU, while consuming an average of 28% of the GPU power. Besides lower clock frequency, FPGAs usually achieve a higher number of operations per cycle in each customized deep pipeline, but lower effective parallel factor due to the far lower off-chip memory bandwidth. With 4x more memory bandwidth, 8 out of the 15 FPGA kernels are projected to achieve at least half of the GPU kernel performance.

AB - This paper aims to better understand the performance differences between FPGAs and GPUs. We intentionally begin with a widely used GPU-friendly benchmark suite, Rodinia, and port 15 of the kernels onto FPGAs using HLS C. Then we propose an analytical model to compare their performance. We find that for 6 out of the 15 ported kernels, today's FPGAs can provide comparable performance or even achieve better performance than the GPU, while consuming an average of 28% of the GPU power. Besides lower clock frequency, FPGAs usually achieve a higher number of operations per cycle in each customized deep pipeline, but lower effective parallel factor due to the far lower off-chip memory bandwidth. With 4x more memory bandwidth, 8 out of the 15 FPGA kernels are projected to achieve at least half of the GPU kernel performance.

KW - Analytical model

KW - FPGA

KW - GPU

KW - Performance comparison

UR - https://www.scopus.com/pages/publications/85057754695

UR - https://www.scopus.com/pages/publications/85057754695#tab=citedBy

U2 - 10.1109/FCCM.2018.00023

DO - 10.1109/FCCM.2018.00023

M3 - Conference contribution

AN - SCOPUS:85057754695

T3 - Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018

SP - 93

EP - 96

BT - Proceedings - 26th IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 26th Annual IEEE International Symposium on Field-Programmable Custom Computing Machines, FCCM 2018

Y2 - 29 April 2018 through 1 May 2018

ER -

Understanding Performance Differences of FPGAs and GPUs

Abstract

Publication series

Conference

Bibliographical note

Keywords

Publisher link

Find It

Other files and links

Fingerprint

Cite this