AMD's Advanced Media Framework Promotes RADV Support

[phoronix]

Phoronix: AMD's Advanced Media Framework Promotes RADV Support

AMD today published version 1.4.34 of its Advanced Media Framework (AMF) SDK. This accelerated multimedia framework is used on Linux and Windows for integrating with games/applications leveraging DirectX, OpenGL, and OpenCL with interoperability support for Radeon GPU customers...

https://www.phoronix.com/news/AMD-AMF-1.4.34

[pharmasolin]

This is huge.

[xpris]

This is huge.

Yes but for decoding FFMPEG Vulkan Video maintainders dont want it to be merged into FFMPEG and that really bad. I hope at some point they allow it. In my own testing AMF in encoding works much better than vaapi even with zero-copy. So I hope also for better decoding... If only some people weren't pigs (trying to promote their own solutions and devaluing the competition) we could have this.

[sophisticles]

The biggest problem with AMF is the price of AMD's discrete GPUs, a look through Microcenter's website show ridiculous prices,

Before anyone says "what about NVIDIA's prices", yes they are high as well but not as high as AMD's and I was able double my money by riding the NVIDIA stock wave, so I can stomach the NVIDIA's prices a bit easier.

Having said that, at the moment no one can beat Intel's offerings, in terms of price, what you get for the price and in all honesty they have superior open source support than AMD.

AMF is interesting if someone has or was to build a 8xxxG based system, or something like the Ryzen 9 7900X3D and use the integrated graphics for video editing.

[Daktyl198]

I can't remember, is AMF accessible via the FOSS drivers yet? Or are we still limited to VA-API?

[entropicdrifter]

I can't remember, is AMF accessible via the FOSS drivers yet? Or are we still limited to VA-API?

This article literally says that AMF for encoding, scaling, and tranformations is now fully accessible via RADV, the open source Mesa AMD Vulkan driver. Decode is available too, but it's considered experimental so I think it's hidden behind a flag or something.

[entropicdrifter]

Yes but for decoding FFMPEG Vulkan Video maintainders dont want it to be merged into FFMPEG and that really bad. I hope at some point they allow it. In my own testing AMF in encoding works much better than vaapi even with zero-copy. So I hope also for better decoding... If only some people weren't pigs (trying to promote their own solutions and devaluing the competition) we could have this.

Yeah, hopefully now that this is officially accessible via the open-source drivers they'll soften up on that stance a bit.

[edxposed]

Yes but for decoding FFMPEG Vulkan Video maintainders dont want it to be merged into FFMPEG and that really bad. I hope at some point they allow it. In my own testing AMF in encoding works much better than vaapi even with zero-copy. So I hope also for better decoding... If only some people weren't pigs (trying to promote their own solutions and devaluing the competition) we could have this.

FFmpeg's Vulkan Video implementation is so bad that RADV is the only driver that works perfectly, so no one would want to use FFmpeg's Vulkan Video if they merged the AMF decoder.The same goes for VVC, which, in order to prevent their shit native VVC decoder from being swept to the trash heap, politically blocked thevvdec patches.

[jrch2k8]

FFmpeg's Vulkan Video implementation is so bad that RADV is the only driver that works perfectly, so no one would want to use FFmpeg's Vulkan Video if they merged the AMF decoder.The same goes for VVC, which, in order to prevent their shit native VVC decoder from being swept to the trash heap, politically blocked thevvdec patches.

To be honest here is not entirely FFMPEG's fault.

1. AMDVLK(whichever version) is broken AF(some releases more than others) and they fail in a lot of places include vkd3d, geekbench, dxvk, etc. so not entirely FFMPEG's fault.
2. nVidia basically slapped together enough code to make Vulkan video show something on screen and called it a day, since their only priority is NVENC/DEC, same situation with Vulkan compute where it overlaps with business important CUDA functionality.
3. Intel Vulkan implementation is still under heavy development and misses lot of functionality but is getting better everyday, still not totally FFMPEG's fault.

So, it only works perfectly on RADV because is the only Vulkan driver for Linux that offer full and stable functionality not because of some dumb conspiracy theory.

I assume in the near future RADV and ANV will be the go to drivers for Vulkan video and if it gets commonly used enough in the future nVidia may (big if here) change their implementation to some nvdev/enc shim good enough to be usable most of the time.

[Artim]

If only some people weren't pigs (trying to promote their own solutions and devaluing the competition) we could have this.

This is literally what AMD is doing. Intel shows with their oneAPI how to do things, yet AMD insists on their questionable stuff that most users can't even use. That's not how you compete with Nvidia, that's how you make sure Nvidia will always win.