Original title:
Performance comparison of a signal processing pipeline execution using CPU and GPU
Authors:
Tomašov, A. ; Horváth, T. Document type: Papers
Language:
eng Publisher:
Vysoké učení technické v Brně, Fakulta elektrotechniky a komunikačních technologií Abstract:
The paper compares the execution performance of NumPy and PyTorch mathematical libraries in embedded systems with graphics processing unit (GPU) acceleration. Both frameworks execute a signal processing pipeline from a fiber manipulation detection system, which inspects a signal from a state of polarization analyzer to enhance the security of optical fiber. The performance comparison is evaluated in the NVIDIA Jetson Nano system with 128-core Maxwell GPU. Based on the measured results, the PyTorch library executed on the GPU has performance improvement from 59 % to 84 % on different batch sizes. The results prove the real-time analysis capabilities of such a system with GPU acceleration.
Keywords:
Fiber Optics Security; GPU; NVIDIA Jetson Nano; PyTorch; Signal Processing Acceleration Host item entry: Proceedings I of the 28st Conference STUDENT EEICT 2022: General papers, ISBN 978-80-214-6029-4
Institution: Brno University of Technology
(web)
Document availability information: Fulltext is available in the Brno University of Technology Digital Library. Original record: http://hdl.handle.net/11012/209386