This work evaluates how far x264 can be pushed and what per-title encoding ladders deliver, using FFmpeg/x264 on 13 diverse 1080p clips and VMAF as the quality metric. Incremental tuning - veryslow preset, a 10s GOP, two-pass, extra reference frames - plus switching from a fixed Netflix-style ladder to per-title ladders increased average VMAF by 0.66 points while cutting average bitrate from 4.90 Mbps to 3.27 Mbps (≈33%). Encoding time grew 3.2× on an Intel Core i9-14900 workstation; baseline ladder cost ≈$1.25/hr of content encoded, optimized ≈$3.99/hr, so the extra encoding expense is recovered after about 186 hours of viewing at $0.02/GB bandwidth. Results use a Top-Heavy viewer distribution (≈71.6% on top rungs) and tune top rungs to VMAF≈93 because quality above 93 is visually indistinguishable.
The practical conclusion is clear: implement per-title encoding before assuming gains from HEVC/AV1, since untuned H.264 leaves significant bandwidth on the table. Benefits scale with how inefficient the existing ladder is (big wins if current 1080p top rungs are 5.5-6.5 Mbps) and with clip complexity (news/animation gain more than sports). For mobile-heavy audiences the VMAF gain grows but bandwidth savings shrink, extending the breakeven point, so per-title tuning should be judged against audience profile and encoding throughput.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Yantra is an LALR(1) parser generator designed for C++ that helps developers create parsers efficiently. It is available as an open-source project on GitHub with ongoing development and community support. (github.com)
Jevotron is a command-line tool for integrating multiple Jev workflows, enabling anomaly detection and data assessment across various file formats. It offers features like preview, scoring, review, and caching to streamline data analysis and review processes. (cmungall.github.io)
EDG C++ Compiler has been released as open source on GitHub, allowing developers to access and contribute to its codebase. The project includes extensive commits, files, and development activity. (github.com)
PLDB catalogs over 33,000 programming languages papers from 1973 to 2026, covering top venues like POPL, PLDI, and ICFP. It allows searching by topics such as concurrency, effect handlers, and ownership types. (pldb.kirancodes.me)
Perspica is a tool that provides a semantic diff for reviewing code changes. It aims to improve code review accuracy by analyzing the meaning of code differences. (github.com)
Halfspace is an experimental IDE for solid modeling using distance fields, allowing real-time rasterization and mesh export. It aims to simplify working with implicit surfaces by providing both a shape library and tools for incremental model building. (mattkeeter.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.