Intel To Work On AV1 Decoding Support, FFmpeg / GStreamer Plugins

[phoronix]

Phoronix: Intel To Work On AV1 Decoding Support, FFmpeg / GStreamer Plugins

Yesterday Intel finally announced their SVT-AV1 video encoder as a promising high-performance AV1 encoder but it turns out they also have open-source plans this year for developing a performant AV1 decoder, among other interesting items on their road-map...

http://www.phoronix.com/scan.php?page=news_item&px=Intel-TODO-AV1-Decode-FFmpeg

[microcode]

I wonder what benefit it could have over dav1d, I guess we'll see.

[tildearrow]

There is already dav- never mind

[edwaleni]

Hmmm. This makes me wonder if the new Intel GPU in the works has some secret sauce that will amplify this effort for a "performant" AV1 encoder.

Something that can setup a very large number of parallel workstreams or makes it look like a large vector processing engine to the host.

Clear Linux with 4 Intel GPU's would make a fascinating render farm POC.

[dwagner]

I understand the efforts for a quicker encoder, but the decoder side seems pretty much covered, already. That said, competition may get us even more improvements.

[Go_Vulkan]

@Michael: I think it is time to look at the quality, not the speed only. We must compare the same quality before any performance measurement should be done. I think there will be a drawback with this encoder.

Hmmm. This makes me wonder if the new Intel GPU in the works has some secret sauce

Very unlikely. The reason is more AMDs ZEN2 standing in front of the door. With 16 cores, 7 nm manufacturing process, low power consumption and low prices. Intel trys to find any reason to stay with them instead of just buying AMD. If they tailor this software for the specialties of their CPUs, these benchmarks can hide the real performance difference with average software.

[edwaleni]

@Michael: I think it it time to look at the quality, not the speed only. We must compare the same quality before any performance measurement should be done. I think there will be a drawback with this encoder.


Very unlikely. The reason is more AMDs ZEN2 standing in front of the door. With 16 cores, 7 nm manufacturing process, low power consumption and low prices. Intel trys to find any reason to stay with them instead of just buying AMD. If they tailor this software for the specialties of their CPUs, these benchmarks can hide the real performance difference with average software.

I understand what you are saying, but no one has found anything in Clear Linux that is Intel specific. The tests show a unique compiler flag but it runs on AMD products just as easily and just as well.

As for the quality, I think that is outside the scope of Phoronix and more in scope with other sites. Yes, I have seen renders using prior gen NVidia GeForce cards and some of them make your head shake. Phoronix is performance oriented, not a video A+B comparison deal.

[GreenReaper]

Perhaps so, but video quality is a part of performance, just as noise or power usage is. If you can push out 60FPS of janky, high-bandwidth video, that may look great on a bar chart, but everyone not streaming will use software instead.

[smitty3268]

Perhaps so, but video quality is a part of performance, just as noise or power usage is. If you can push out 60FPS of janky, high-bandwidth video, that may look great on a bar chart, but everyone not streaming will use software instead.

Uhh, this is software.

That said, it's been well documented already that their encoder is trading off quality for higher performance.

[LinAGKar]

As for the quality, I think that is outside the scope of Phoronix and more in scope with other sites. Yes, I have seen renders using prior gen NVidia GeForce cards and some of them make your head shake. Phoronix is performance oriented, not a video A+B comparison deal.

Except comparing the performance without looking at the quality is extremely misleading. You could probably make the claim that x264 is slower than SVT-AV1 by setting it to placebo.