NVIDIA RTX 30 Series Supports AV1 Accelerated Video Decoding

[Mario Junior]

AMD doesn't even have full VP9 HW decoding

[aphysically]

Well, this doesn't have a lot of consequence on Linux anyway except for stuff like mpv. Even the new firefox x11 vaapi support doesn't work on nvidia's proprietary drivers (even though there's a vdpau-to-vaapi library) since it requires DMA-BUF that nvidia can't use since it's exported GPL in this Linux kernel.

[wswartzendruber]

Some day, between all of the useless AV1 encoders currently out there, maybe just one of them will rise up to be usable.

[SyXbiT]

though I always hope to game on my GPU, I spend a lot more time watching h265 video, and Nvidia has been excellent at that with NVDEC. Is AMD just as good as Nvidia for 4k hardware decode of h264/h265?
AV1 is a great thing for futureproofing, as it will be used by Amazon, Netflix and Youtube

[rene]

You have zero computers, right? Because no fucking way you could be running without UEFI BIOS (closed source blob), ROMs for your HDD/SSD, NIC and your classy AMD card which doesn't even work without firmware. Open Source fans never fail to impress with their idolatry and total lack of knowledge.

What a total lack of knowledge on your side. There is a vast difference in having some (unfortunate) binary only ROM for startup, and having a huge, Windows codebase based Nvidia blob running within your single dress space monolithic kernel. https://www.youtube.com/watch?v=PLvNEguPPw8

The boot ROM one can also replace with core boot et al. However, with a driver BLOB and no open register level specs you can't even fix bugs or run the card on other open systems, such as Haiku or other state of the art research.

[mdedetrich]

It takes up chip space, drives up manufacturing cost and is superseeded.

This is misleading. Putting in hardware level acceleration for codecs in general does take up more die space but once you have hardware level acceleration for codecs, supporting additional codes is trivial when it comes to "die space". A lot of the die is reused between the different codecs for the fixed function algorithms that are used.

[rene]

This is misleading. Putting in hardware level acceleration for codecs in general does take up more die space but once you have hardware level acceleration for codecs, supporting additional codes is trivial when it comes to "die space". A lot of the die is reused between the different codecs for the fixed function algorithms that are used.

lol "trivial" and "not much die space", ever tried implementing a modern video encoder in RTL?

[mdedetrich]

lol "trivial" and "not much die space", ever tried implementing a modern video encoder in RTL?

You misunderstood what I was saying. Adding hardware level accelerated codecs in the first place is definitely not trivial, but once its added supporting extra codecs is not that much more die space. This is especially true if there is already support for a more advanced codec, i.e. h.265

You would have a point if the graphics card in question didn't have codec acceleration at all in the first place, but this is not the case.

We are also not talking about AVX 512 here

[Mike Frett]

Some day, between all of the useless AV1 encoders currently out there, maybe just one of them will rise up to be usable.

Some day I hope well get some of that nice AI neural network upscaling that we're missing out on...even though the software is open source. It's a shame really.

[OneTimeShot]

lol "trivial" and "not much die space", ever tried implementing a modern video encoder in RTL?

Well.. you clearly haven't. Most of the raw pixel graphics processing parts (transforms/filters/motion compensation) are the same between codecs. The big differences are in the protocol encoding/decoding algorithms which is done with a programmable coprocessor.

As long as a codec doesn't add new graphics transforms you can use the same hardware. So (e.g.) MPEG-2 is just a firmware program using hardware that supports h264. AV1 and VVC will actually share a lot of hardware - the pixel stuff has been out of patent for a while now.