HEVC successor: Versatile Video Coding

[ksec]

Originally Posted by excellentswordfight

TSMC 5nm node has higher transistor cost then 7nm, and 3nm will probably follow that trend, so not sure how 3nm will help if the reason for the lack of av1 decoders was cause of roi on transistor budget.

Edit. initially ofc, but the trend is that transistor cost isnt drastically reduced anymore from die shrinks.

If you are comparing non EUV 7nm aka N7, then barely. Otherwise no. ( EUV 7nm and EUV 5nm. ) Wafer Price are on the rise and per transistor cost isn't yet. It is not hitting the same cost reduction but per transistor cost has never risen per generation during their time of introduction.

Design Cost is the main factor in cost increase, but that bare little relevance to unit cost without specific volume accounted for.

5nm and 3nm are relevant not only because of cost but Die Space and Power Usage. Where in Mobile devices you are expecting a budget of sub 100mW for Decoding. In Laptop you have the luxury of budget increase by 10x / 1W if not more on a GPU.

[hajj_3]

MPEG-LA has released a press release about their VVC patent pool: https://www.mpegla.com/wp-content/up...2021-01-27.pdf

[excellentswordfight]

Originally Posted by ksec

If you are comparing non EUV 7nm aka N7, then barely. Otherwise no. ( EUV 7nm and EUV 5nm. ) Wafer Price are on the rise and per transistor cost isn't yet. It is not hitting the same cost reduction but per transistor cost has never risen per generation during their time of introduction.

Design Cost is the main factor in cost increase, but that bare little relevance to unit cost without specific volume accounted for.

5nm and 3nm are relevant not only because of cost but Die Space and Power Usage. Where in Mobile devices you are expecting a budget of sub 100mW for Decoding. In Laptop you have the luxury of budget increase by 10x / 1W if not more on a GPU.

I was talking about the non euv N7 node, as its rumored that 5nm wafer price was 1.8x higher, while comparing something like apples a12 to a14, density ”only” increased by 1.5x.

But thats intresting, but to me it sounds a bit odd... ”We cannot make an sub 100mW decoder on current silicon so lets use software decoding instead!”

I though that the key factor was that they didnt wanna waste money on silicon for an feature that might not see wide adoption, and that av1 needed quite a bit of silicon budget.

[benwaggoner]

Originally Posted by excellentswordfight

I though that the key factor was that they didnt wanna waste money on silicon for an feature that might not see wide adoption, and that av1 needed quite a bit of silicon budget.

I've heard estimates that an AV1 decoder can take twice the silicon budget of a VVC decoder. That extra cost of AV1 HW support is likely higher than any savings from not having a per-decoder licensing fee.

[GTPVHD]

https://www.streamingmedia.com/Artic...wo-144949.aspx

VVC is dead on arrival as many have predicted.

[Funky080900]

Originally Posted by GTPVHD

https://www.streamingmedia.com/Artic...wo-144949.aspx

VVC is dead on arrival as many have predicted.

“Note that two pools aren't necessarily a bad thing; many successful technology standards involve multiple pools”

The article doesn't say that VVC is dead only that “potential licensees won't know total VVC royalty costs until both pools release their royalty terms”

[ksec]

Originally Posted by Funky080900

“Note that two pools aren't necessarily a bad thing; many successful technology standards involve multiple pools”

The article doesn't say that VVC is dead only that “potential licensees won't know total VVC royalty costs until both pools release their royalty terms”

Access Advance already have some preliminary announcement [1] with regards to pricing a long time ago. Considering they are making it a HEVC + VVC pool, you can bet it won't be cheaper than HEVC. It will be at best, the same as current HEVC pricing but also includes VVC.

And again, ValosMedia, or specifically Qualcomm. Given all other members of Valos are already in either MPEG LA or Access, Qualcomm are not in any the pools ( yet ).

[1] https://accessadvance.com/hevc-advan...-hevc-license/

[benwaggoner]

Originally Posted by Funky080900

“Note that two pools aren't necessarily a bad thing; many successful technology standards involve multiple pools”

The article doesn't say that VVC is dead only that “potential licensees won't know total VVC royalty costs until both pools release their royalty terms”

HEVC has three pools, and certainly isn't dead. The licensing situation is Not Good, certainly, but pretty much every device you've bought in the last 4-5 years has a HW HEVC decoder.

[foxyshadis]

Originally Posted by benwaggoner

I've heard estimates that an AV1 decoder can take twice the silicon budget of a VVC decoder. That extra cost of AV1 HW support is likely higher than any savings from not having a per-decoder licensing fee.

I wonder how much of the spec they implement, since nothing has trickled down for consumers to test yet. AV1 bucks the general trend of having a very stripped-down baseline MPEG profile that hardware can implement; its baseline includes most of the more complex tools and colorspaces that hardware pretends don't exist and therefore the state of the art never really budges as much as you'd expect.

[benwaggoner]

Originally Posted by foxyshadis

I wonder how much of the spec they implement, since nothing has trickled down for consumers to test yet. AV1 bucks the general trend of having a very stripped-down baseline MPEG profile that hardware can implement; its baseline includes most of the more complex tools and colorspaces that hardware pretends don't exist and therefore the state of the art never really budges as much as you'd expect.

Yeah, it's a tradeoff. It's awesome we can do 444 10-bit in the baseline decoder, but that definitely bumps up complexity. That alone requires 2.5x the sample fill rate of 420 8-bit.

MPEG has been great at making very thoughtful choices to optimize for HW decoder complexity, but AV1 didn't get nearly the breadth of input about HW implementations, and efforts were very much focus on SW decoders for a long time.

[hajj_3]

VTM 12.0 is out.

[LigH]

New upload: VTM Encoder Version 12.0 [Windows][GCC 10.2.0][64 bit] 091d3bb5