x265 4.2 Lands With Threaded Motion Estimation and 8K-Ready Levels

The x265 project shipped version 4.2 on April 19, and for once the headline feature is not a new rate-control mode or an HDR format - it is raw speed.
 

The release introduces Threaded Motion Estimation, an experimental mode that can make encoding up to 1.5x faster on 1080p and lower-resolution video when the machine has cores to spare.

For a project that has traditionally squeezed speed out of tighter loops and smarter assembly, this is a noticeably different approach: a dedicated thread pool that precomputes motion estimation in parallel with the rest of the encode.

The catch, openly acknowledged by the developers, is that the feature shines on multi-core machines running at lower clock speeds.

On high-frequency systems and at 4K resolution, the overhead can actually make things slower - which is why the feature ships as experimental and must be enabled explicitly with a new --threaded-me flag.

Ready for the Next Generation of Video

The quieter but longer-lasting change is spec-level: x265 4.2 adds support for HEVC Levels 6.3 through 7.2, the new tiers introduced in the ITU-T H.265 version 9 specification from September 2023.

These levels cover very high-resolution and very high-bitrate content - the kind of 8K and high-framerate material that consumer hardware is just starting to reach.

Most users won't touch these tiers today, but encoders need to support them now so that archival-grade pipelines can plan around them.

The Rest of the Highlight Reel

• Frame parallelism for Motion-Compensated Spatio-Temporal Filtering (MCSTF) - the denoising-style feature that landed in 4.x now runs in parallel across frames instead of serially.
• Quality improvements in Segment-Based Rate Control (SBRC) - cleaner bitrate distribution across long videos.
• Updated Dolby Vision Profile 5 VUI defaults - better HDR compatibility out of the box.
• Roughly 8% faster encoding on ARM processors thanks to expanded NEON and SVE optimizations.
• Around 2x faster encoding on RISC-V with a broad set of newly written SIMD kernels for SAD, SATD, DCT, SAO, loop filter, and more - a notable nod toward the architecture's growing relevance.
• Updated AOM Film Grain parameter signaling for film-grain synthesis workflows.

There is also a generous list of bug fixes: memory leaks in SEI buffers and analysis save/load, a crash with hist-scenecut on high bit-depth builds, a lookahead concurrency bug, a chroma QP offset fix for non-4:4:4 inputs, and several ARM SIMD regressions from earlier builds.

Known Issues

The release notes call out three known limitations worth being aware of: output mismatches between analysis save and load with cutree at reuse levels under 10, inconsistent hash mismatches with the abr-ladder feature, and the performance regression with threaded-me on high-frequency systems and 4K content mentioned above.

Should You Update?

For most people encoding HEVC through a GUI like X HEVC Encoder or HandBrake, the update will arrive automatically in a future bundled release - there is nothing to do manually.

If you are running x265 directly or building your own pipeline with FFmpeg or StaxRip, 4.2 is worth the upgrade for the stability fixes alone; the new threaded encoding mode is a bonus for anyone with a many-core CPU who encodes mostly 1080p content.

The full release notes and binaries are available now, and the x265 HEVC Encoder 4.2 download on free-codecs has been updated to match.