Cisco Announces "Thor" Royalty-Free Video Codec

[caligula]

There so many next gen video codecs that it starting to get hard to follow.
HEVC, Daala, VP9, Thor.
The question is which will have supported hardware acceleration, patent situation and licensing fees.

Don't forget all the audio codecs. People want at least 10.2 channels of 1536 kHz audio and 64 bit dynamic range. So there definitely needs to be dozens of proprietary lossless codecs for different formats.

[chithanh]

Thor and Daala are not even finalized yet, so asking for hardware acceleration is preposterous.

Why? It seems completely reasonable to me.

Because no sane manufacturer will implement an unfinished codec in their hardware. It's not like software which you can change by offering a simple download.

We know it sometimes happens that manufacturers make hardware before the relevant standard is final. 802.11 draft-n was one of the more notorious examples. But it is quite a risk and thus more the exception than the rule.

Not liking isn't a good description.
And what is claimed isn't always right which often shows over time.

Here is what Cisco wrote:

Unfortunately, the patent licensing situation for H.265 has recently taken a turn for the worse. Two distinct patent licensing pools have formed so far, and many license holders are not represented in either. There is just one license pool for H.264. The total costs to license H.265 from these two pools is up to sixteen times more expensive than H.264, per unit. H.264 had an upper bound on yearly licensing costs, whereas H.265 has no such upper limit.

The existence of two patent pools, each with their own licensing rules, is an undisputed fact.
The existence of entities who claim patents relevant to H.265 and who are not members of either patent pool is a fact too.
I don't know what else Cisco must do or say to substantiate that the patent situation makes H.265 unacceptable for use as universal video codec.

Google didn't pay the MPEG LA extortion money, Google was *successfully* fending off all attacks. But even successful defense costs money, so Google paid a bit to the MPEG LA so that they would tell companies to back off.
That VP8, and VP9 for that matter, don't take off beyond Youtube isn't because the MPEG LA supposedly sunk it, it's because they're effectively proprietary Google formats.

The goal of the MPEG LA in this case was clearly to obstruct and slow down the competing standard through FUD. This is a common tactic. And they mostly reached that goal. Google had to pay up to save what was left in terms of 3rd party support.

I wrote more about that here.

Thanks, I did not know about the bug. A pretty ugly situation if you ask me, and Google did certainly not do The Right Thing? here.

[Gusar]

Regarding your second point, well, there's no "instead of working with VP9". You cannot. Development of the codec (not the code, mind you, the codec itself) is proprietary and owned by google, so there's no "working with". That's why they want a codec under the supervision of a standardisation group, not a single company.

Thank you! At least someone gets it. There's a lot of comments out there objecting that the Thor blog post calls VP9 proprietary. So it's nice to see that a few people get it. "Proprietary" can and does mean something other than "closed source".

[erendorn]

Why? It seems completely reasonable to me. Unless Thor is more than 50% better than VP9, then I'd go for the codec with hardware support every time. Given that I have a hard time telling Theora apart from HEVC, and I have plenty of bandwidth, I don't see the value (for me) in releasing a new codec each year.

Still, if Cisco want to waste their time instead of working with VP9, it's their prerogative.

Due to the difficulty of updating hardware, especially a little integrated part of an entire SoC, it should be obvious to everyone why beta hardware support is not a thing. If the codec has no final definition, there will be of course zero hardware support.

Regarding your second point, well, there's no "instead of working with VP9". You cannot. Development of the codec (not the code, mind you, the codec itself) is proprietary and owned by google, so there's no "working with". That's why they want a codec under the supervision of a standardisation group, not a single company.

[Gusar]

Everyone with a slight clue on that matter (i.e. not you) knows since years that VP8 is based on AVC

Sorry, but no. They share some generic concepts (then again, *all* codecs do), but the actual implementation is quite different. The entropy coder is different, there's no bi-prediction in VP8, or adaptive quantization (while aq is actually an encoder feature, the bitstream must make it possible; VP8 doesn't, it only has the far more limited "segments" - VP9 is more advanced in this regard, but still not as flexible as h264/h265), the VP8 alt-ref and golden frame concept doesn't exist in h264, and there's several other differences that escape my mind right now.

Google didn't pay the MPEG LA extortion money, Google was *successfully* fending off all attacks. But even successful defense costs money, so Google paid a bit to the MPEG LA so that they would tell companies to back off. That VP8, and VP9 for that matter, don't take off beyond Youtube isn't because the MPEG LA supposedly sunk it, it's because they're effectively proprietary Google formats. I wrote more about that here.

Cisco had H.264 patents, so skin in the game then. H.265 and they are crying that they don't have any leverage with their fellow patent holders. Boo hoo.

Err, Cisco has h265 patents too. This isn't them throwing a tantrum because they aren't getting any h265 money. This is them acknowledging that the system is flawed and as such actively preventing adoption of codecs.


And to all those saying "why not hardware acceleration". It should be totally obvious why not, but because apparently it isn't obvious to some, let me explain: It's way too soon. Daala and Thor's bitstream are far from finalized yet. You don't start designing a hardware circuit for a format that will change next week. And then change again the week after that. Once the bitstream is set in stone, then and only then do you start thinking about hardware decoding.

[Marc Driftmeyer]

Cisco had H.264 patents, so skin in the game then. H.265 and they are crying that they don't have any leverage with their fellow patent holders. Boo hoo.

[bug77]

WebRTC

And http://www.webex.com/

Would have been a perfect opportunity for a "Thor is here" article title

[My8th]

VP8 is still being used lots

[spinkham]

That it's cheaper to develop your own codec then licence one says just about everything there is to say about how broken the software patent system is...