By Dieison Silveira (Ph.D. Student, PGMICRO/UFRGS)
& Vignesh V Menon
A detailed energy consumption analysis, accounting for the consumption components of CPU, cache memories and main memory, for all x265 presets executing in a multi-core system was presented in the paper titled ‘Performance and Energy Consumption Analysis of the x265 Video Encoder’ published in 2017 25th European Signal Processing Conference (EUSIPCO). The Holy Grail of preset options in x265 is the optimized trade-off between encoding speed and compression efficiency. This work, on the other hand, presents a detailed performance and energy consumption analysis of the x265 video encoder, considering the energy consumption related to CPU, main memory, and caches. The figure below depicts the energy efficiency considering the energy consumption per encoded byte.
According to the paper, ultrafast preset has a bitrate 45% higher than the placebo preset. The results also showed that the system energy consumption increases 45x, from ultrafast preset to the placebo preset. This happens because slower presets tests more encoding options and are more computationally intensive, spending more energy. The paper recommends the user to choose the presets depending on the video resolution, memory, and target platform. Some of the key noteworthy findings from the paper:
- Fast, medium and slow presets present the best bitrate/energy trade-offs. In this way, if the device has bitrate and energy constraints these three presets are good options.
- For energy-constrained devices, such as battery-powered embedded systems, a better solution is the use of fastest presets, which have the lowest energy consumption.
- For high-resolution videos such as UHD 4K and 8K, slower presets are recommended, since these presets achieve best compression ratio.
The figure above presents the behavior of bitrate and energy consumption for all presets. Please find the full paper here.
We encourage everyone working on x265 or using x265 as a tool to evaluate any video quality metric, pre and post processing tools to write to us about their observations and recommendations to improve x265. We welcome to share their findings on some visual quality or performance improvement algorithms which could be beneficial if implemented in x265.
Disclaimer: The graphs presented in the blog are from the research paper.