Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Wednesday February 06 2019, @01:55AM   Printer-friendly
from the commoditize-your-complement dept.

Intel Releases Open Source Encoder for Next-Gen AV1 Codec

Intel published its own open source CPU-based encoder for the next-generation and royalty-free AV1 codec (a codec is a program for encoding / decoding a digital data stream or signal). Intel is one of the main founding members of the Alliance for Open Media (AOM), the non-profit group behind the development of the AV1 codec.

Intel's new encoder, called Scalable Video Technology AOMedia Video 1 (SVT-AV1), aims to fill the role of a good CPU-based encoding software tool until dedicated AV1 encoders are ready for prime time. The encoder supports the Linux, macOS and Windows operating systems.

A CPU-based encoder requires a beefy system, so it's no surprise the real-time encoding specifications for SVT-AV1 are no joke. SVT-AV1 requires Skylake-generation or newer Xeon processors with at least 112 threads and at least 48GB of RAM for 10-bit 4K video encoding. Outside of video streaming companies, these type of systems are out of reach for most. Consumers that want to encode AV1 videos may want to wait for dedicated AV1 encoding hardware to appear, which make take another year or so.

Here's a recent 42-minute talk (no transcript) about AOMedia Video 1 (AV1). Hardware support for AV1 should begin appearing around 2020.

Related: Alliance for Open Media Announces Release of AOMedia Video Codec 1.0 (AV1) Specification
YouTube and Netflix Upload AV1-Encoded Videos for Testing


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by sjames on Thursday February 07 2019, @05:39AM (1 child)

    by sjames (2882) on Thursday February 07 2019, @05:39AM (#797625) Journal

    It still says something about the encoding complexity. I can encode h.265 in better than real time with an 8 core Opteron system.

    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 2) by takyon on Thursday February 07 2019, @06:06AM

    by takyon (881) <reversethis-{gro ... s} {ta} {noykat}> on Thursday February 07 2019, @06:06AM (#797631) Journal

    Further optimizations could be made to the encoder. Intel likely slapped its encoder together using code it used for its HEVC encoder (this is info I got from the linked video talk in TFS). But in a world where you can have 16-32 cores for relatively cheap, we've finally found something to do with those cores. 16-core Ryzen is probably coming this year. I think we could see another doubling of cores on the TSMC "5nm" [anandtech.com] node.

    https://rethinkresearch.biz/articles/ao-media-looks-ahead-to-av2-as-av1-picks-up-momentum/ [rethinkresearch.biz]

    AV1 is also slightly hampered by the encoding overhead, although that will be a diminishing burden given progress in hardware. After all AV1 was designed to trade hardware for efficiency.

    I'm excited to see what AV2 will bring to the table. Some potential features were dropped during AV1 development [wikipedia.org] because they would have made performance even worse than it is now:

    Daala Transforms were the major innovation behind the daala codec. They implement "lapped" discrete cosine and sine transforms that its authors describe as "better in every way" than the txmg set of transforms that prevailed in AV1. Both the txmg and daala_tx experiments have merged high and low bitdepth code paths (unlike VP9), but daala_tx achieved full embedding of smaller transforms within larger, as well as using fewer multiplies, which could have further reduced the cost of hardware implementations. The Daala transforms were kept as optional in the experimental codebase until late January 2018, but changing hardware blocks at a late stage was a general concern for delaying hardware availability.

    The encoding complexity of Daala's Perceptual Vector Quantization (PVQ) was too much within the already complex framework of AV1. The Rate Distortion dist_8x8 heuristic aims to speed up the encoder by a sizable factor, PVQ or not, but PVQ was ultimately dropped.

    ANS was the other non-binary arithmetic coder, developed in parallel with Daala's entropy coder. Of the two, Daala EC was the more hardware friendly, but ANS was the fastest to decode in software.

    --
    [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]