SVT-AV1: Proven Performance, Promising Horizons
SVT-AV1 technology has amassed quite a history. While it’s now a product of the Alliance for Open Media (AOMedia) Software and Implementation Working Group (SIWG), the origins of the technology pre-date the working group’s formation, beginning as an Intel and Netflix-led project in 2019. Momentum grew over the years, and additional companies, including Meta and Tencent, joined the SVT-AV1 effort. In 2020, AOMedia created the SIWG, which took up SVT-AV1 as its flagship project. We connected with Ioannis Katsavounidis and Roman Chernyak, co-chairs of the SIWG, to learn more about the benefits of SVT-AV1, its progress, and what the future may hold.
Strong Foundation, Ongoing Improvements
Scalable Video Technology (SVT) refers to the ability of an encoder to scale its performance according to the compute capabilities of the platform on which it is running, essentially allowing it to use the maximum amount of allocated resources to achieve the highest possible video quality.
The SVT-AV1 encoder has been the SIWG’s primary focus since the group’s launch, and for good reason. This encoder offers three orders of compute scalability (that's more than 1,000 times speed difference from the slowest to the fastest presets) with the most beneficial tradeoff between compute and bitrate efficiency among all open software implementations of video encoders of various coding standards. The project is also targeted to cater to multiple Image and Video compression use cases ranging from VOD, to Broadcast, to Live streaming, to Real-time Communications (RTC) and last but not least, Image compression implementing the AVIF standard. The SIWG team keeps working tirelessly to offer a publicly available, state-of-the art production-grade implementation to fast-track the productization of AV1.
Since its adoption by AOMedia, SVT-AV1 has seen twenty releases and more than 2,300 commits, currently sitting at version 3.1.0. “We have delivered constant improvements in compute-bitrate efficiency tradeoffs, reaching levels of x264-veryfast speeds, while still delivering very significant (>50%) bitrate savings,” explained Katsavounidis. One of the biggest contributions is the development of a so-called "fast-decode" mode of operation. This allows for the generation of AV1 fully compliant bitstreams that require much less (up to 50% less) compute when decoded on mobile phones (using the dav1d open-source software encoder library). “This advancement allows for the delivery of AV1 streams to older devices that would otherwise be unable to decode AV1, and thus miss out on the opportunity that AV1 offers over older video coding standards,” he added.
SVT-AV1 in the “Reel” World
As SVT-AV1 continues to improve, adoption continues to grow. Meta has publicly stated that all its AV1 streams (which cover more than 50% of the watch time of videos on Facebook and Instagram) are produced using SVT-AV1. Additionally, SVT-AV1 allowed Meta to stream high-quality short-form videos (Reels) for the first time. At the same time, the research, industry and open-source communities have also embraced SVT-AV1 and are beginning to employ it, so the SIWG expects to see adoption grow even further in the coming years. “SVT-AV1 has allowed billions of people to enjoy high-quality videos at lower bitrates, even on older mobile phones,” said Katsavounidis. “In this way, SVT-AV1 has greatly improved the video experience for countless consumers.”
And it’s only going to get better. Katsavounidis expects SVT-AV1 to continue evolving and improving over the next 5-10 years, with the goal of offering the best compute-bitrate efficiency tradeoff available across multiple video and image streaming use cases. With AOMedia celebrating its tenth anniversary this year, the SVT-AV1 success story serves as a strong proof-point for the impact of the group’s work over the past decade, as well as the potential of what’s to come.