Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain

Michael Petry; Gabriel Wuwer; Andreas Koch; Patrick Gest; Max Ghiglione; Martin Werner

doi:10.23919/EDHPC59100.2023.10396011

Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain

Michael Petry, Gabriel Wuwer, Andreas Koch, Patrick Gest, Max Ghiglione, Martin Werner

Assistant Professorship of Big Geospatial Data Management

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

Abstract

Artificial intelligence has found its way into space and necessitates a powerful and flexible hardware platform to keep up with the fast-paced AI domain. With the space-grade variant of the Versal, AMD-Xilinx offers one of the first space-ready AI accelerators that combine multiple compute paradigms, i.e., scalar processing (CPU), adaptive engines (FPGA), and vector processing (AI-Engine array) into an adaptive System-on-Chip. This paper provides a thorough analysis of its AI capabilities with respect to throughput and power efficiency for Multi-Layer Perceptrons and CNNs, and takes a look under the hood by profiling the system's efficiency on an architectural level based on the idea of the Roofline model. We believe that the gained insights ultimately help to design optimal NN architectures for deployment on the Versal.

Original language	English
Title of host publication	Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023
Editors	Maris Tali, Max Ghiglione
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9789090379241
DOIs	https://doi.org/10.23919/EDHPC59100.2023.10396011
State	Published - 2023
Event	2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023 - Juan-Les-Pins, France Duration: 2 Oct 2023 → 6 Oct 2023

Publication series

Name	Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023

Conference

Conference	2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023
Country/Territory	France
City	Juan-Les-Pins
Period	2/10/23 → 6/10/23

Keywords

AMD-Xilinx Versal
fpga
hardware accelerator
machine learning
neural network
roofline model

Access to Document

10.23919/EDHPC59100.2023.10396011

Cite this

Petry, M., Wuwer, G., Koch, A., Gest, P., Ghiglione, M., & Werner, M. (2023). Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain. In M. Tali, & M. Ghiglione (Eds.), Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023 (Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.23919/EDHPC59100.2023.10396011

Petry, Michael ; Wuwer, Gabriel ; Koch, Andreas et al. / Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain. Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023. editor / Maris Tali ; Max Ghiglione. Institute of Electrical and Electronics Engineers Inc., 2023. (Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023).

@inproceedings{6c88e5c7bb9f4fe186e7d945eec2c748,

title = "Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain",

abstract = "Artificial intelligence has found its way into space and necessitates a powerful and flexible hardware platform to keep up with the fast-paced AI domain. With the space-grade variant of the Versal, AMD-Xilinx offers one of the first space-ready AI accelerators that combine multiple compute paradigms, i.e., scalar processing (CPU), adaptive engines (FPGA), and vector processing (AI-Engine array) into an adaptive System-on-Chip. This paper provides a thorough analysis of its AI capabilities with respect to throughput and power efficiency for Multi-Layer Perceptrons and CNNs, and takes a look under the hood by profiling the system's efficiency on an architectural level based on the idea of the Roofline model. We believe that the gained insights ultimately help to design optimal NN architectures for deployment on the Versal.",

keywords = "AMD-Xilinx Versal, fpga, hardware accelerator, machine learning, neural network, roofline model",

author = "Michael Petry and Gabriel Wuwer and Andreas Koch and Patrick Gest and Max Ghiglione and Martin Werner",

note = "Publisher Copyright: {\textcopyright} 2023 ESA.; 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023 ; Conference date: 02-10-2023 Through 06-10-2023",

year = "2023",

doi = "10.23919/EDHPC59100.2023.10396011",

language = "English",

series = "Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023",

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

editor = "Maris Tali and Max Ghiglione",

booktitle = "Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023",

}

Petry, M, Wuwer, G, Koch, A, Gest, P, Ghiglione, M & Werner, M 2023, Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain. in M Tali & M Ghiglione (eds), Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023. Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023, Institute of Electrical and Electronics Engineers Inc., 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023, Juan-Les-Pins, France, 2/10/23. https://doi.org/10.23919/EDHPC59100.2023.10396011

Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain. / Petry, Michael; Wuwer, Gabriel; Koch, Andreas et al.
Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023. ed. / Maris Tali; Max Ghiglione. Institute of Electrical and Electronics Engineers Inc., 2023. (Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023).

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

TY - GEN

T1 - Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain

AU - Petry, Michael

AU - Wuwer, Gabriel

AU - Koch, Andreas

AU - Gest, Patrick

AU - Ghiglione, Max

AU - Werner, Martin

PY - 2023

Y1 - 2023

N2 - Artificial intelligence has found its way into space and necessitates a powerful and flexible hardware platform to keep up with the fast-paced AI domain. With the space-grade variant of the Versal, AMD-Xilinx offers one of the first space-ready AI accelerators that combine multiple compute paradigms, i.e., scalar processing (CPU), adaptive engines (FPGA), and vector processing (AI-Engine array) into an adaptive System-on-Chip. This paper provides a thorough analysis of its AI capabilities with respect to throughput and power efficiency for Multi-Layer Perceptrons and CNNs, and takes a look under the hood by profiling the system's efficiency on an architectural level based on the idea of the Roofline model. We believe that the gained insights ultimately help to design optimal NN architectures for deployment on the Versal.

AB - Artificial intelligence has found its way into space and necessitates a powerful and flexible hardware platform to keep up with the fast-paced AI domain. With the space-grade variant of the Versal, AMD-Xilinx offers one of the first space-ready AI accelerators that combine multiple compute paradigms, i.e., scalar processing (CPU), adaptive engines (FPGA), and vector processing (AI-Engine array) into an adaptive System-on-Chip. This paper provides a thorough analysis of its AI capabilities with respect to throughput and power efficiency for Multi-Layer Perceptrons and CNNs, and takes a look under the hood by profiling the system's efficiency on an architectural level based on the idea of the Roofline model. We believe that the gained insights ultimately help to design optimal NN architectures for deployment on the Versal.

KW - AMD-Xilinx Versal

KW - fpga

KW - hardware accelerator

KW - machine learning

KW - neural network

KW - roofline model

UR - http://www.scopus.com/inward/record.url?scp=85184849919&partnerID=8YFLogxK

U2 - 10.23919/EDHPC59100.2023.10396011

DO - 10.23919/EDHPC59100.2023.10396011

M3 - Conference contribution

AN - SCOPUS:85184849919

T3 - Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023

BT - Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023

A2 - Tali, Maris

A2 - Ghiglione, Max

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023

Y2 - 2 October 2023 through 6 October 2023

ER -

Petry M, Wuwer G, Koch A, Gest P, Ghiglione M, Werner M. Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain. In Tali M, Ghiglione M, editors, Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023. Institute of Electrical and Electronics Engineers Inc. 2023. (Proceedings of the 2023 European Data Handling and Data Processing Conference for Space, EDHPC 2023). doi: 10.23919/EDHPC59100.2023.10396011

Accelerated Deep-Learning Inference on the Versal adaptive SoC in the Space Domain

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this