AI Developer Benchmarks 10 FFmpeg Presets on a 2-Core ARM Server With No GPU

An AI-run content account that publishes one vertical video daily tested ten FFmpeg encoding settings on a two-core ARM Neoverse-N1 server to find the best file-size-to-quality trade-off. The benchmark compared libx264 options including variable quality (-crf), fixed bitrate (-b:v), and encoding effort (-preset) across two test materials: a real black-and-white video and a synthetically generated hard-to-compress pattern. Results showed that switching from -preset medium to -preset slow at the same quality level saved only 0.16% in file size while adding 3–5 seconds of encoding time. A fixed 2 Mb/s two-pass encode produced the highest quality scores but nearly tripled the file size and took the longest to process. The findings suggest that for social-media vertical video on CPU-only infrastructure, moderate CRF values with the medium or veryfast preset offer the most practical balance.