Intel "Arctic Sound M" Enterprise Accelerator Shown Encoding AV1 Video

[btarunr]

Intel showed off the video encode acceleration capabilities of its upcoming Data Center GPU codenamed "Arctic Sound M," which features AV1 video encode hardware-acceleration. Intel has been pushing for AV1 to be adopted as the streaming video standard for some years now, as it offers comparable bitrate savings and quality to H.265, but is royalty free, resulting in tens of millions of Dollars of royalty savings for streaming content providers such as Netflix, as well as consumer electronics manufacturers, particularly smart TV makers.

Intel's accelerator is a single-slot, full-height add-on card with a PCI-Express x16 interface. It relies on rack airflow for cooling, and features a metal-channel heatsink. Power is drawn from a single 8-pin PCIe power connector. There are no encoding performance numbers put out, except the 30% bitrate savings AV1 offers compared to the current industry standard, H.264. This 30% saving adds up in a big way for a streaming content provider.



Intel's video presentation follows.

View at TechPowerUp Main Site

 

[ModEl4]

Sure 30% reduction in file size is fine and dandy but it's too heavy without dedicated acceleration, i hope YouTube to delay the AV1 adoption as far as it can be

 

[Steevo]

I notice it says 2 of the GPU's on the test setup.

Its a fair trade off in quality, the H264 looks washed out in some areas, the AV1 looks more blurry in some.

 

[windwhirl]

Sure 30% reduction in file size is fine and dandy but it's too heavy without dedicated acceleration, i hope YouTube to delay the AV1 adoption as far as it can be

It's already done. Youtube just chooses to send you a VP9 stream or any other type according to whatever your browser is.

 

[ModEl4]

It's already done. Youtube just chooses to send you a VP9 stream or any other type according to whatever your browser is.

Isn't AV1 internal YouTube encoding be dependent on the resolution, the views and the age of the video (when uploaded, I'm not talking for new content only)?
Sometime in the future for new content it may not stream VP9 for 4K resolution depending also on the view count making AV1 accelerated capable hardware a necessity in essence.
I don't know just asking, I haven't seen a roadmap regarding future YouTube requirements regarding AV1 acceleration.

 

[noel_fs]

wasnt av1 supposed to be more efficient than 265? why are the even comparing it to 264? 30% is already done by 265?

 

[windwhirl]

Isn't AV1 internal YouTube encoding be dependent on the resolution, the views and the age of the video (when uploaded, I'm not talking for new content only)?
Sometime in the future for new content it may not stream VP9 for 4K resolution depending also on the view count making AV1 accelerated capable hardware a necessity in essence.
I don't know just asking, I haven't seen a roadmap regarding future YouTube requirements regarding AV1 acceleration.

They've been rolling it out slowly. Originally only available for very select UHD content and later for low-res videos (up to 480p). However, it's now more common (though not all content is available in AV1)

wasnt av1 supposed to be more efficient than 265? why are the even comparing it to 264? 30% is already done by 265?

I don't think that was the purpose, beating H265 I mean. The whole point of AV1 is being royalty free and doing away with closed standards. Besides, HEVC never really took off as H264 did (licensing being one of the reasons why).

Though, that aside, AV1 has been shown to beat H265 by a small margin (10-20%) in terms of compression efficiency in some tests.

Compression efficiency analysis of AV1, VVC, and HEVC for random access applications | APSIPA Transactions on Signal and Information Processing | Cambridge Core

Compression efficiency analysis of AV1, VVC, and HEVC for random access applications - Volume 10

www.cambridge.org

 

[Vya Domus]

These comparisons are always hilarious because you can easily pick and choose videos that make it seem more impressive than it actually is at large, not to mention that once you turn everything into a JPEG it all looks the same anyway.