Logo image
100x Speedup of HEVC Motion Estimation Using Multicore and GPGPU
Thesis

100x Speedup of HEVC Motion Estimation Using Multicore and GPGPU

Chen, Wei-Chih
Masters, 國立清華大學, 資訊工程學系所
2016

Abstract

高效率視頻編碼 圖形處理器通用計算 多核心處理器 HEVC GPGPU Multicore
The High-Efficiency Video Coding (HEVC) standard provides higher compression ratio than the older H.264/MPEG-4 AVC at the expense of immense computation complexity. Motion estimation (ME) is the most intensive computation in an HEVC encoder. To speed up HEVC encoding, we propose a data parallel computing method to perform the motion estimation part using a GPU and a multithreaded encoder using a multicore CPU. Running on a Nvidia Titan X GPGPU, our successively-refined methods can speedup the motion estimation part by 102 times and the total encoding time by 91 times while suffering very low bit-rate increase (0.177%) and PSNR loss (0.005dB).

Metrics

1 Record Views

Details

Logo image