Vidaio posts −23.5% AV1 benchmark

On May 21, X user @bittingthembits posted a chart claiming Vidaio, running on TAO, beat both an FFmpeg AV1 reference and AWS MediaConvert in a video-encoding benchmark. The post said Vidaio achieved a −23.5% BD-Rate versus the FFmpeg baseline, while AWS MediaConvert came in at +97.5%. The thread also said the result could translate into large streaming-infrastructure savings, but it did not include vendor confirmation, a full methodology document, or public comment from Vidaio or Amazon Web Services. ### What exactly was posted on May 21? The May 21 X post compared two encoder outputs against an FFmpeg AV1 reference and expressed the result in BD-Rate, a standard compression metric used to summarize bitrate differences across rate-distortion curves. In that framing, a negative BD-Rate means an encoder needs fewer bits than the reference to reach the same measured quality, while a positive number means it needs more. (forasoft.com) The post’s most cited numbers were −23.5% for Vidaio and +97.5% for AWS MediaConvert. Those figures, if reproduced under the same test conditions, would imply Vidaio used materially less bitrate than the FFmpeg reference at matched quality and that the MediaConvert configuration used in the test required substantially more. The post did not, in the material reviewed, provide a full public test pack with source clips, command lines, presets, or the quality metric used to generate the BD-Rate curves. (forasoft.com) ### What does BD-Rate measure in plain terms? BD-Rate, short for Bjøntegaard Delta rate, is used in codec testing to compare average bitrate differences between two rate-distortion curves over a shared quality range. The metric is widely used in video-compression research and engineering, but published guidance also warns that results depend heavily on the underlying test design, including the clips selected, the number of points on each curve, and the quality metric used. (forasoft.com) A result such as −23.5% does not mean a fixed 23.5% savings on every file. It means that, across the tested range and with the chosen quality metric, the encoder under test averaged lower bitrate than the reference for similar measured quality. A +97.5% result likewise describes the tested setup, not every possible AWS MediaConvert AV1 workflow. (arxiv.org) ### How does AWS MediaConvert fit into this comparison? AWS says MediaConvert is a file-based video transcoding service for preparing on-demand video assets for delivery across devices. AWS documentation also shows that MediaConvert supports AV1 output and uses quality-defined variable bitrate, or QVBR, for AV1 rate control, with tunable quality levels. (forasoft.com) AWS’s own documentation does not support drawing a broad conclusion from a single third-party benchmark without the exact settings. QVBR quality level, tuning choices, and other job parameters can materially change bitrate and quality outcomes, according to the service guides. ### Why was FFmpeg used as the reference? FFmpeg is commonly used as a front end for AV1 encoders, including libaom-av1, and serves as a familiar baseline in engineering tests. (docs.aws.amazon.com) FFmpeg’s AV1 documentation shows support for libaom-based AV1 encoding and a range of rate-control modes and parameters, which means benchmark outputs can vary significantly with configuration. (docs.aws.amazon.com) The choice of FFmpeg as a reference does not, by itself, establish that one commercial service is universally better than another. It establishes only the baseline chosen for that test. That matters because encoder comparisons can shift with presets, content mix, target ladder, and the quality metric used to build the rate-distortion curves. ### What is still missing before the claim can be treated as settled? (trac.ffmpeg.org) Neither Vidaio nor AWS had publicly verified the benchmark in the material reviewed by May 22. The post also did not include, in the reviewed source set, a reproducible package naming the source corpus, resolution ladder, objective metric, encoder versions, and exact commands or job settings. (arxiv.org) The next concrete step is public replication. A reproducible benchmark would need the test clips, FFmpeg commands, MediaConvert job settings, and the BD-Rate calculation method so other engineers can rerun the comparison against the same AV1 outputs. (arxiv.org) (docs.aws.amazon.com)