Hacker News Reader

Comment by LeoPanthera | Hacker News Reader

Last time I checked (very unscientific test using Handbrake) the AV1 encoder was something like 10x slower than the hevc encoder on the same system. Is this likely to improve?

View on HN

Yes, currently there is no hardware encoder/decoder for AV1, but there is for h264 on many processors. This may change in the future if companies and users adopt it.

I might be mistaken, but I believe most high quality encodes in both h264 / hevc are not done using dedicated hardware encoders built into processors or graphics cards. While it's much faster there's usually some quality tradeoff. Instead, I believe that newer processors have additional instruction sets that enable much faster software encoding on the same codecs. Someone correct me if I'm wrong about this.

You're right, there are fewer h264 encoders than decoders in consumer software, but still dozens. https://en.wikipedia.org/wiki/H.264/MPEG-4_AVC_products_and_... There are precisely 0 for AV1 so far (except some may have a subset of the standard in FPGA by now).

> Yes, currently there is no hardware encoder/decoder for AV1, but there is for h264 on many processors. This may change in the future if companies and users adopt it.

Does x86/x64 have any? What's the name of one of the instructions (for example)? I feel like mostly all I see is just bigger vector instructions...

It's not a collection of x86_64 instructions, it's an separate specialized core on Intel CPUs. https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardwar... and AMD https://en.wikipedia.org/wiki/Video_Coding_Engine You need OS-dependent driver code to access it. Otherwise, if a decoder just uses x86_64 instructions, it's a software decoder.

Whoa!!! So there's a CPU, a GPU, and a dedicated video decoder?? I had no idea. Thanks for sharing this!!

Yeah, its called a VPU aka Video Processing Unit. If your hardware implements video decoding, they likely have licensed or designed their own VPU IP block.

This is dangerously close to becoming a standard talking point nthis topic. It’s dangerous because it’ll be wrong as soon as anyone implements an encoder with speed as an objective, and especially when dedicated hardware starts to appear in devices (about 12 to 18 months from now)

It did with H265, the prototype implementations were horribly slow.

Except that "The AV1 reference encoder is [as of today] a hundred times slower than an HEVC one"

https://www.ibc.org/delivery/codec-wars-the-battle-between-h...

Does it matter? It's still a reference encoder. AFAIK, the AV1 bitstream is barely frozen. Is he talking about the working model HEVC encoder right after they froze it, or the final reference encoder before people started working on hardware implementations? The Current AV1 reference being some amount slower than some HEVC reference says nothing in itself about room for improvement.

Reply on news.ycombinator.com