H.265 Encoding - Fragmented and distorted videos with GPU acceleration versus flawless CPU encoding

Introduction

DivX was an open-source video encoding solution. This technology was copied and copyrighted by manufacturers, including faulty NVIDIA PhysX encodings.

H.265/HEVC encoding has established itself as the standard for highly compressed video formats in recent years, as it offers significantly higher data reduction compared to H.264 while maintaining comparable visual quality. However, with the increasing prevalence of GPU-accelerated encoding on desktop and mobile platforms, especially ARM-based systems, artifacts, fragmentation, and distorted videos are becoming increasingly common. These problems contrast sharply with CPU-based encoding, which is generally error-free and reproducible despite the higher processing load.

1. GPU versus CPU Encoding

1.1 CPU Encoding

• CPUs operate serially and with a deterministic computing architecture.

  Advertising

• Encoding algorithms are executed step by step, ensuring high accuracy and bit-level reproducibility.

• Even with very long videos or extreme compression rates, the integrity of the material is largely preserved.

1.2 GPU Encoding

• GPUs operate massively in parallel to achieve high encoding speeds.

• However, with H.265 encoding, Block-based motion compensation and intra-prediction are precisely synchronized.

• Inaccurate floating-point calculations, memory-related rounding errors, or hardware bugs lead to fragmented frames, distorted pixels, or even failed GOPs (Group of Pictures).

• ARM-based GPUs, such as those used in mobile devices, often rely on simplified or flawed hardware decoders/encoders, which increases their susceptibility to errors.

2. Long-Term Problems

2.1 Data Loss in the Future

• Videos based on faulty GPU encodings may no longer be decodable in 50 years.

• Historical data that exists only in these formats would be partially irretrievably lost.

• Even if the original material is available, the fragmented GPU-encoded copies are qualitatively corrupted and cannot be easily reconstructed.

2.2 Analogy to Images

• The same problem applies to photos taken on mobile devices stored with ARM GPUs.

• Hardware-accelerated JPEG, HEIF, or AVIF encoding can generate corrupted metadata or pixel corruption, which in the long run leads to digital decay.

3. Causes of Fragmentation

1. Parallelization Conflicts: GPUs process many image blocks simultaneously, which leads to inconsistencies if synchronized incorrectly.

2. Hardware Bugs: ARM GPUs or integrated mobile graphics chips sometimes contain faulty logic for H.265 encoding.

3. Floating Point Deviations: Differences in precision between CPU and GPU calculations lead to small, cumulative errors that become visible in long videos.

4. Driver and Software Incompatibilities: Different GPU drivers implement H.265 instruction sets differently, which reduces reproducibility.

4. Implications for Archiving

• Historians, media archives, and private users must be aware that GPU-encoded videos may be unusable in 50 years.

• CPU-based encoding remains the most reliable way for long-term preservation, even if it is slower.

• For critical applications, hardware-agnostic formats and software encoders should be preferred.

5. Conclusion

GPU-accelerated H.265 encoding offers immense speed and energy efficiency, especially on mobile devices. However, it generates fragmented, distorted, and sometimes unusable data, which can lead to digital decay in the long term. Despite higher computing costs, CPUs deliver error-free and reproducible coding, making them ideal for long-term archiving, scientific datan and historical media remain essential.
The long-term security of digital content therefore depends not only on capacity or speed, but on the reliable hardware and software implementation of the encoding.

h265 encoding and distorted and fragmented videos with GPU acceleration, in contrast to CPU encoding, which runs almost error-free. Consider, for example, how it is. in 50 years, and that most videos would then unfortunately be lost forever. The same applies to images, especially on mobile devices like ARM, which only rely on faulty GPUs.