Public media guide
AV1 vs HEVC vs H.264: In-Depth Codec Efficiency & Compression Benchmark
An engineering comparison of AV1, HEVC (H.265), and AVC (H.264). We analyze Bjontegaard-Delta rate (BD-rate) gains, VMAF visual fidelity scores, licensing royalties, and hardware decoding across consumer silicon.
Digital video delivery represents over 82% of all downstream internet bandwidth. At the center of this infrastructure are video compression standards—mathematical transforms designed to eliminate spatial, temporal, and psychoacoustic redundancies from raw pixel rasters.
For over two decades, Advanced Video Coding (AVC / H.264) served as the universal lingua franca of web video, delivering ubiquitous hardware playback across virtually every connected screen. However, as 4K, 8K, and high-frame-rate (HFR) streaming became mainstream, the bandwidth overhead of H.264 became unsustainable.
Today, digital video engineers primarily evaluate two successor generations: High Efficiency Video Coding (HEVC / H.265), standardized by the ITU-T and ISO/IEC, and AOMedia Video 1 (AV1), the open-source, royalty-free standard developed by the Alliance for Open Media (AOMedia). Understanding the architectural distinctions, rate-distortion curves, and silicon decoder roadmaps is essential for any modern media delivery pipeline.
In this engineering benchmark, we evaluate the algorithmic foundations of each standard, examine empirical BD-rate gains across standard 1080p and 4K sequences, analyze licensing implications, and provide production-ready FFmpeg encoding recipes.
1. Transform Block Partitioning & Prediction Architectures
The compression advantage of modern codecs originates in how they subdivide macroblocks and predict motion between adjacent video frames. While H.264 relies on fixed 16x16 macroblocks with sub-partitions down to 4x4, HEVC introduces the Coding Tree Unit (CTU) architecture supporting flexible quadtree decomposition up to 64x64 blocks. AV1 extends this further with a multi-type recursive block partitioning tree reaching 128x128 superblocks with 10 different partition shapes, including non-square 1:4 and 4:1 rectangular splits.
- H.264: Fixed 16x16 macroblock grids with limited directional intra-prediction modes (9 luma modes).
- HEVC: Dynamic Coding Tree Units (CTUs) up to 64x64 with 35 intra-prediction directions and asymmetric motion partitioning.
- AV1: 128x128 superblocks featuring 56 directional intra angles, recursive split-merge syntax, and wedge-based compound prediction.
- Result: AV1 isolates fine textures, sharp edges, and subtle gradients with significantly fewer residual transform bits.
2. In-Loop Filtering & Artifact Suppression
Transform coding inevitably produces blocking boundaries and ringing artifacts, particularly at aggressive quantizer levels. While H.264 employs a rudimentary deblocking filter, HEVC introduces Sample Adaptive Offset (SAO) to correct banding in flat regions. AV1 incorporates a sophisticated three-stage in-loop filtering pipeline: a standard deblocking filter, the Constrained Directional Enhancement Filter (CDEF) to eliminate edge ringing, and Loop Restoration based on Wiener filtering and self-guided restoration filters.
- CDEF operates along directional lines in 8x8 blocks to suppress ringing without smoothing legitimate edge details.
- Wiener restoration calculates mathematical Wiener coefficients to invert blur introduced during quantization.
- Film Grain Synthesis in AV1 allows grain to be stripped prior to encoding and regenerated procedurally at decoder runtime, saving up to 40% bitrate in filmic content.
3. Licensing, Royalties, and Commercial Viability
A critical factor differentiating codec adoption is the intellectual property structure. H.264 has a stable, well-understood royalty structure governed by MPEG LA. In contrast, HEVC is fractured across multiple competing patent pools—MPEG LA, HEVC Advance (Access Advance), Velos Media, and independent patent holders—leading to unpredictable royalty liabilities for software distributors and web streaming services.
- H.264: Capped annual royalties with royalty-free internet streaming distribution.
- HEVC: Highly fragmented patent pools with ongoing disputes regarding content distribution fees and hardware per-device royalties.
- AV1: Developed by AOMedia (Google, Apple, Microsoft, Amazon, Netflix, Meta, Intel, NVIDIA) with an explicit, royalty-free defensive patent grant.
4. Hardware Decoding Ecosystem in 2026
Codec efficiency on paper is useless if end-user devices experience thermal throttling or battery drain during software decode. Over the past five years, AV1 hardware silicon has achieved critical mass across mobile, desktop, and smart display processors, rendering software-only fallback necessary only for legacy hardware.
- Mobile Silicon: Apple A17 Pro, A18, Qualcomm Snapdragon 8 Gen 2/3/4, MediaTek Dimensity 9000/9200/9300 series feature native 4K/8K 60fps AV1 hardware decoders.
- Desktop & Laptop: Intel Core Ultra (Meteor/Arrow Lake), AMD Ryzen 7000/8000/9000, and Apple M3/M4 chipsets provide full hardware acceleration.
- Browser Support: Google Chrome, Mozilla Firefox, Microsoft Edge, and Apple Safari (on supported hardware) enable AV1 playback by default.
Format & Use Table
| Specification | AVC (H.264) | HEVC (H.265) | AV1 (AOMedia) |
|---|---|---|---|
| Standardized Year | 2003 | 2013 | 2018 |
| Max Superblock Size | 16x16 | 64x64 | 128x128 |
| Intra Prediction Modes | 9 modes | 35 modes | 56 directional + chroma |
| Relative Bitrate (VMAF 93) | 100% (Baseline) | 65% (-35% bandwidth) | 48% (-52% bandwidth) |
| Royalty Structure | Capped (MPEG LA) | Fragmented (3+ pools) | 100% Royalty-Free |
| In-Loop Filters | Deblocking | Deblocking + SAO | Deblock + CDEF + Loop Rest. |
| Film Grain Synthesis | No (In-stream bits) | No (In-stream bits) | Yes (Parametric runtime) |
Step-by-Step Workflow
Install modern FFmpeg with libsvtav1 and libx265: Ensure your build includes SVT-AV1 v2.0+ for multi-threaded CPU acceleration.
Encode baseline H.264 reference: ffmpeg -i input.mov -c:v libx264 -crf 22 -preset slow -c:a aac -b:a 128k h264_ref.mp4
Encode HEVC reference: ffmpeg -i input.mov -c:v libx265 -crf 24 -preset medium -tag:v hvc1 -c:a aac hevc_test.mp4
Encode SVT-AV1 target: ffmpeg -i input.mov -c:v libsvtav1 -crf 28 -preset 6 -svtav1-params tune=0:film-grain=8 -c:a libopus -b:a 96k av1_target.mp4
Calculate VMAF score: ffmpeg -i av1_target.mp4 -i input.mov -filter_complex "[0:v][1:v]libvmaf=model=version=vmaf_v0.6.1" -f null -
Frequently Asked Questions
Does AV1 encode slower than H.264 and HEVC?
Historically, AV1 encoders required significant CPU time. However, modern implementations like SVT-AV1 (Scalable Video Technology) have introduced optimized presets (preset 6 and 7) that encode at speeds competitive with libx265 medium while preserving superior compression.
Why does Apple support both HEVC and AV1?
Apple was a founding contributor to HEVC and uses it natively across ProRes captures and AirPlay pipelines. However, Apple joined AOMedia and added AV1 hardware decoding silicon starting with the A17 Pro and M3 processors to optimize web streaming battery life on YouTube and Netflix.
Can I replace all my MP4 files with AV1 WebM files today?
For general web delivery, we recommend serving AV1 WebM or MP4 via HTML5 video source tags with an H.264 MP4 fallback to ensure 100% compatibility across legacy Smart TVs and older devices.
What is VMAF and why is it preferred over PSNR/SSIM?
Video Multi-Method Assessment Fusion (VMAF) was developed by Netflix. Unlike simple mathematical pixel comparisons (PSNR) or structural assumptions (SSIM), VMAF uses machine learning trained on human viewer perception to provide an accurate reflection of subjective visual quality.
