Benchmarks

Benchmarks for comparing the two codecs are still quite complex at present. Because VVC is relatively new on Linux, you need to build the libvvenc library from source and then compile it into FFmpeg. However, FFmpeg has steadfastly refused to integrate the library in Debian. Ubuntu 25.10 already provides an updated FFmpeg version, but it cannot use libvvenc either. Consequently, I developed a script using ChatGPT that allows comparison of both codecs and displays the results graphically.

I used an excerpt from the freely available Blender movie Big Buck Bunny [7] as my source material. I processed the same short clip consisting of 300 frames with both AV1 and VVC. It was impossible to overlook the fact that VVC took about 4.7 times longer than AV1 to process the same clip under identical conditions.

For AV1, the script uses the libaom reference library and Opus for the audio track (Figure 1). The approximately 10-second sequence resulted in a WebM format with a file size of 4.8MB and a bitrate of 2400kbit/s. For VVC, the script used the libvvenc library (Figure 2), which produced a video with a size of just 640KB at the same bitrate. The file generated by AV1 is 7.5 times larger than the one generated by VVC. However, VVC also takes almost five times as long.

Figure 1: The benchmark script uses a YUV file as a starting point and outputs the file compressed in WebM format.

Figure 2: The script's basic settings for VVC.

Figure 3 shows the various presets used. In AV1, the AV1-4 and AV1-8 presets are a tradeoff between encoding speed and video quality. AV1-4 is relatively slow and aims to achieve very high quality and strong compression with a small file size. AV1-8 works significantly faster, but with less efficient compression and slightly lower quality. The file size is sometimes larger than with the slower presets, but the computing time is greatly reduced.

Figure 3: In practical tests, VVC proved to be significantly more efficient. Depending on the preset, the files are two to eight times smaller than those encoded with AV1, with comparable image quality.

For VVC, I used the slower and faster presets. The slower variant focuses on quality and compression efficiency. The encoder uses more complex algorithms and searches through larger areas of the image, which significantly slows down processing, but ensures the best possible quality and smallest file size. The faster preset is optimized for maximum speed. The encoder uses less complex methods and significantly reduces the number of compute steps. This reduces the time required, but compression efficiency suffers a little, and the video file is larger and of slightly lower quality.

Figure 3 shows that VVC produces significantly smaller files than AV1 in all variants. While the VVC-encoded clips in the test were mostly around 600KB in size, the corresponding AV1 versions were 1300KB and 5000KB in size. VVC is therefore two to eight times more efficient than AV1 at comparable image quality. In practical terms, this means that VVC allows videos of the same quality to be stored in a far more space-saving manner and streamed at a lower data rate. The price for this, in addition to the license fees, is more computational load during encoding, which is often only worthwhile for professional video productions and streaming services.

Conclusions

AV1 is the natural ally for Linux and open source in particular: license-free, implemented as open source, and without legal uncertainties. AV1-encoded media consumes more bandwidth when streaming due to larger file sizes, but it saves you from licensing fees, which is why organizations such as Google, Netflix, and Amazon prefer to use AV1. You'll find VCC used more often in applications such as broadcasting, professional media production, or ultra-high-end applications, where license fees are already factored in. To help you decide which codec meets your needs, Figure 4 provides a cheat sheet for comparing AV1 and VVC.

Figure 4: AV1 and VVC at a glance.