GPU Performance and Power Consumption Analysis: A DCT based denoising application

• Autores: Pi Puig, Martín; De Giusti, Laura Cristina; Naiouf, Marcelo; De Giusti, Armando Eduardo
• Año de publicación: 2017
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; denoising; power; energy; GPU; NVML; RAPL
• Institución: Universidad Nacional de La Plata

It is known that energy and power consumption are becoming serious metrics in the design of high performance workstations because of heat dissipation problems. In the last years, GPU accelerators have been integrating many of these expensive systems despite they are embedding more and more transistors on their chips producing a quick increase of power consumption requirements. This paper analyzes an image processing application, in particular a Discrete Cosine Transform denoising algorithm, in terms of CPU and GPU performance and energy consumption. Specifically, we want to compare single-threaded and multithreaded CPU versions with a GPU version, and characterize the execution time, true instant power and average energy consumption to deflate the idea that GPUs are non-green computing devices.
XVIII Workshop de Procesamiento Distribuido y Paralelo (WPDP).
Red de Universidades con Carreras en Informática (RedUNCI)