How Tencent Cloud’s V265/TXAV1 Revolutionizes 8K Live Streaming

Introduction

As live video pushes toward ultra‑high‑definition, low‑latency, and high‑bitrate streams, the emergence of Apple Vision has further expanded 3D and 8K 120 fps encoding demands, making video encoding optimization increasingly challenging. Tencent Cloud’s Jiang Aojie presents the V265/TXAV1 live‑streaming capabilities and optimization techniques.

Part 1: V265/TXAV1 Live Capability Overview

Live streaming has become a popular medium connecting providers and consumers across education, sports, gaming, and e‑commerce. With rising user expectations for high‑quality video, efficient codecs like AV1 and H.265 are essential for delivering clearer images at lower bitrates.

Tencent Cloud’s Shannon Lab has spent over a year optimizing AV1/H.265 encoders for ultra‑high‑resolution, high‑bitrate, and low‑latency scenarios, achieving significant quality and stability improvements.

In the MSU2022 competition, TXAV1 and V265 outperformed X265, saving over 30 % and 40 % bitrate respectively in live‑stream benchmarks, and ranking first in most metrics.

Performance highlights:

V265 vs. X265 medium: >36 % bitrate reduction with 20 % speed increase.

TXAV1 vs. X265 medium: >40 % bitrate reduction at comparable speed.

TXAV1 vs. V265: ~10 % additional compression at similar speed.

Part 2: Typical Live‑Streaming Use Cases

2.1 8K Live

Key goals: full functionality, low latency, high performance.

Full functionality: supports 8K 60 fps, 10‑bit, 150 Mbps, 4:2:2, HDR, ABR.

Low latency: single‑device encoding with minimal delay.

High performance: up to 9× speedup over X265 medium at 8K 60 fps.

2.2 Fast Live

Targeted at e‑commerce, showcase, and online education where interaction latency must stay within 500‑1000 ms. Optimizations reduced bitrate by 5‑7 % while maintaining speed.

2.3 MV‑HEVC

Apple Vision Pro supports MV‑HEVC hardware encoding, improving 3D video quality. Tencent Cloud’s MV‑HEVC implementation compresses multi‑view video without side‑by‑side stitching, achieving over 20 % average compression gain, especially on 3D movies with small inter‑eye disparity.

Part 3: Encoding Optimization Techniques

3.1 Engineering Optimizations

Data‑structure improvements:

TreeNode – caches node attributes to avoid recomputation.

CoreUnit – stores core encoding information for fast access.

IdenticalCu – reuses identical coding units across nodes.

SwapBuffer – alternates memory buffers to reduce copies, yielding >5 % speedup generally and >20 % on 8K.

Process optimizations: Parallel analysis, frame‑level, slice/tile, macroblock, and post‑processing stages are re‑engineered for higher concurrency, with adaptive parallel control to balance speed and quality.

Multithreading: Tile‑based parallelism and adaptive WPP strategies increase parallelism, especially for 8K where traditional WPP is limited.

3.2 Algorithm Optimizations

Non‑standard DCT applied to 64×64 blocks reduces transform time by 50‑60 %, with CTU‑level scene detection limiting impact on complex textures, achieving 6 % speedup with only 0.5 % compression loss.

Low‑latency mini‑GOP structures and enhanced reference handling improve QP distribution across layers, boosting performance in fast‑live scenarios.

3.3 Subjective Quality Optimizations

ROI‑based encoding prioritizes viewer‑focused regions, reducing bitrate fluctuation from 32 % to 15 % and further to 5 % after adaptive bitrate‑pool adjustments.

Fast‑live encoding benefits from ROI‑aware block partitioning, yielding only a 5 % increase in processing time while improving subjective quality by 32.3 %.

3.4 Additional Optimizations

Further algorithmic enhancements include adaptive intra‑skip, MVP‑skip, pixel‑ME skip, intra‑mode RD search improvements, and reference‑frame selection refinements. Engineering tweaks cover CTU coding, reference mode copying, intra‑pixel copying, tile syntax updates, and cost‑table calculations.

Tencent Cloud also offers the R265 terminal encoder for zero‑latency compression on both x86 and ARM platforms, delivering 30 % bitrate savings at speeds comparable to x264 veryfast.

Conclusion

Through a combination of data‑structure redesign, process re‑engineering, multithreading, algorithmic shortcuts, and ROI‑driven subjective enhancements, V265/TXAV1 achieves over 100 % speedup with negligible compression loss, while significantly improving visual quality for ultra‑high‑definition live streaming.