Systematic Literature Review: GPU, TPU, FPGA dalam Akselerasi AI
Abstract
The increasing complexity of artificial intelligence (AI) models has raised the demand for efficient hardware accelerators. A key challenge is selecting an accelerator that aligns with application needs, as mismatches can affect energy efficiency, inference speed, and system scalability. GPU, TPU, and FPGA are the most commonly used accelerators in AI deployment, each with specific advantages and limitations. This study aims to systematically evaluate the utilization of these three accelerators across various AI domains. A Systematic Literature Review (SLR) was conducted using the Kitchenham framework, analyzing 20 scientific articles. Results show that GPUs are used in 90% of studies, TPUs in 50%, and FPGAs in 70%. In terms of energy efficiency, FPGAs are superior in 78% of the relevant articles, while GPUs dominate inference performance in 85% of cases. This study concludes that selecting an AI accelerator should be guided by power efficiency, system architecture, and domain-specific requirements. The findings offer practical implications for graduate students in selecting appropriate accelerators that align with their research topics, experimental goals, and resource constraints in AI-driven thesis projects.
Keywords: Artificial Intelligence Accelerator; Graphics Processing Unit; Tensor Processing Unit; Field Programmable Gate Array; Systematic Literature Review
Abstrak
Perkembangan model kecerdasan buatan (AI) yang semakin kompleks menimbulkan kebutuhan akan akselerator perangkat keras yang efisien. Masalah utama yang sering dihadapi adalah pemilihan akselerator yang tidak sesuai dengan kebutuhan aplikasi, yang berdampak pada efisiensi daya, kecepatan inferensi, dan skalabilitas sistem. GPU, TPU, dan FPGA merupakan tiga jenis akselerator yang paling banyak digunakan dalam implementasi AI, Penelitian ini bertujuan mengevaluasi pemanfaatan ketiga akselerator dalam berbagai domain AI menggunakan Systematic Literature Review (SLR) berbasis pendekatan Kitchenham. Sebanyak 20 artikel diseleksi dari lima basis data ilmiah terkemuka. Hasil menunjukkan GPU digunakan dalam 90% studi, FPGA dalam 70%, dan TPU dalam 50%. FPGA unggul dalam efisiensi energi (78% studi), sementara GPU dominan dalam performa inferensi (85% kasus). Penelitian menyimpulkan pemilihan akselerator AI harus mempertimbangkan efisiensi daya, arsitektur sistem, dan kebutuhan domain. Temuan ini memberikan panduan praktis bagi mahasiswa magister dalam memilih akselerator sesuai topik, tujuan eksperimen, dan keterbatasan sumber daya.
Kata Kunci: Akselerator Artificial Intelligence; Graphics Processing Unit; Tensor Processing Unit; Field Programmable Gate Array; Tinjauan Sistematik
References
B. Kitchenham, L. Madeyski, & P. Brereton, "Meta-analysis for families of experiments in software engineering: A systematic review and reproducibility and validity assessment. Empirical Software Engineering, Vol. 25, no. 1, pp. 353–401, 2020. https://doi.org/10.1007/s10664-019-09747-0
S. Bertazzoni et al., “Design Space Exploration for Edge Machine Learning Featured by MathWorks FPGA DL Processor : A Survey,” vol. 12, no. 11, pp. 157482–157504, 2024, doi: 10.1109/ACCESS.2024.3349128.
A. Martín-martín et al., “Hardware Implementations of a Deep Learning Approach to Optimal Configuration of Reconfigurable Intelligence Surfaces,” IEEE Trans. Wireless Commun., vol. 23, no. 8, pp. 4521–4541, 2024, doi: 10.1109/TWC.2024.3372610.
L. Alzubaidi et al., Review of deep learning : concepts , CNN architectures , challenges , applications , future directions. Springer International Publishing, 2021. doi: 10.1186/s40537-021-00444-8.
M. Blott et al., “Evaluation of Optimized CNNs on Heterogeneous Accelerators Using a Novel Benchmarking Approach,” IEEE Trans. Comput., vol. 70, no. 10, pp. 1654–1669, 2021, doi: 10.1109/TC.2020.3022318.
S. Alam, C. Yakopcic, Q. Wu, M. Barnell, S. Khan, and T. M. Taha, “Survey of Deep Learning Accelerators for Edge and Emerging Computing,” Electronics, vol. 13, 2024, doi: 10.3390/electronics13152988.
P. Rech, “Artificial Neural Networks for Space and Safety-Critical Applications : Reliability Issues and Potential Solutions,” IEEE Trans. Nucl. Sci., vol. 71, no. 1, pp. 377–404, 2024, doi: 10.1109/TNS.2024.3349956.
M. Vaithianathan, S. Udkar, and A. Micro, “Comparative Study of FPGA and GPU for High-Performance Computing and AI,” Int. J. Adv. Comput. Technol., vol. 1, no. 1, pp. 107–115, 2024, doi: 10.56472/25838628/IJACT-V1I1P107.
I. Papoutsis, N. Ioannis, A. Zavras, D. Michail, and C. Tryfonopoulos, “ISPRS Journal of Photogrammetry and Remote Sensing Benchmarking and scaling of deep learning models for land cover image classification,” ISPRS J. Photogramm. Remote Sens., vol. 195, no. July 2022, pp. 250–268, 2023, doi: 10.1016/j.isprsjprs.2022.11.012.
S. Costa, F. Neves, P. Machado, P. Moreira, and J. Dias, “Engineering Applications of Artificial Intelligence Benchmarking edge computing devices for grape bunches and trunks detection using accelerated object detection single shot multibox deep learning models,” Eng. Appl. Artif. Intell., vol. 117, no. October 2022, p. 105604, 2023, doi: 10.1016/j.engappai.2022.105604.
Z. Yu, W. Wang, and Y. Chen, “Benchmark on the accuracy and efficiency of several neural network based phase pickers using datasets from China Seismic Network,” Earthq. Sci., vol. 36, no. 2, pp. 113–131, 2023, doi: 10.1016/j.eqs.2022.10.001.
S. Huang, E. Tang, S. Li, X. Ping, and R. Chen, “Hardware-friendly compression and hardware acceleration for transformer : A survey,” EURASIP J. Adv. Signal Process., vol. 30, no. July, pp. 3755–3785, 2022, doi: 10.3934/era.2022192.
P. Kang and J. Jo, “Benchmarking Modern Edge Devices for AI Applications ∗ SUMMARY,” IEICE Trans. Inf. Syst., vol. Vol. E104-, no. 3, pp. 394–403, 2021, doi: 10.1587/transinf.2020EDP7160.
A. S. Aguiar et al., “Visual Trunk Detection Using Transfer Learning and a Deep Learning-Based Coprocessor,” IEEE Access, 2020, vol. 8, no. 1, pp. 78308–78317, doi: 10.1109/ACCESS.2020.2989052.
Y. Liu, M. Yue, H. Yan, and L. Zhu, “Single-image super-resolution using lightweight transformer-convolutional neural network hybrid model,” Inst. Enginering Technol., no. May, pp. 2881–2893, 2023, doi: 10.1049/ipr2.12833.
D. Zhang et al., “A Full-Stack Search Technique for Domain Optimized Deep Learning Accelerators,” Assoc. Comput. Mach., pp. 27–42, 2022, doi: 10.1145/3503222.3507767.
P. Kang and A. Somtham, “An Evaluation of Modern Accelerator-Based Edge Devices for Object Detection Applications,” Mathematics, vol. 10, pp. 1–14, 2022.
A. P. Munanday, N. Sazali, W. Sharuzi, and W. Harun, “Analysis of Convolutional Neural Networks for Facial Expression Recognition on GPU , TPU and CPU,” vol. 3, no. 3, pp. 50–67, 2023.
F. Pacini, T. Pacini, G. Lai, A. M. Zocco, and L. Fanucci, “Deployment of a CNN for Motor Imagery Classification in Brain-Computer Interfaces,” IEEE Trans. Neural Syst. Rehabil. Eng., 2024, vol. 32, no. 1, pp. 1412–1421, doi: 10.1109/TNSRE.2024.3359218.
J. Lu and S. X. Tan, “Thermal Map Dataset for Commercial Multi / Many Core CPU / GPU / TPU”, in Proc. ACM/IEEE Int. Symp. Low Power Electron. Design (ISLPED), 2024, pp. 1–6, doi: 10.1145/3670474.3685963.
C. Rao et al., “ICARUS : A Specialized Architecture for Neural Radiance Fields Rendering,” vol. 41, no. 6, 2022, doi: 10.1145/3550454.3555505.
E. Rapuano et al., “An FPGA-Based Hardware Accelerator for CNNs Inference on Board Satellites : Benchmarking with Myriad 2-Based Solution for the CloudScout Case Study,” Remote Sens., 2021, vol. 13, no. 8, pp. 1519–1536, doi: 10.3390/rs13081519.
M. Vaithianathan, M. Patil, F. S. Ng, and S. Udkar, “Comparative Study of FPGA and GPU for High-Performance Computing and AI,” ESP Int. J. Adv. Comput. Technol., 2024, vol. 1, no. May 2023, pp. 37–46, doi: 10.56472/25838628/IJACT-V1I1P107.
How To Cite This :
Refbacks
- There are currently no refbacks.
