x265 HEVC Encoder

[benwaggoner]

Originally Posted by jlpsvk

yeah... avx512 produces much more heat... so probably safety issue...

This was also true of the initial AVX2 implementation, but the more recent ones get much better. I hope we'll see much more perfomant AVX-512 in the next-gen Xeon.

Still, MCW claims that performance gains can be found with it at slower presets at 4K, and I've got some test 8K samples to check.

Sent from my SM-T837V using Tapatalk

[Atak_Snajpera]

Originally Posted by benwaggoner

This was also true of the initial AVX2 implementation, but the more recent ones get much better. I hope we'll see much more perfomant AVX-512 in the next-gen Xeon.

Still, MCW claims that performance gains can be found with it at slower presets at 4K, and I've got some test 8K samples to check.

Sent from my SM-T837V using Tapatalk

Avx512 won't help xeons when AMD just Has epyc2/threadriper 3990x with 2x more cores
Personally I prefer 2x more cores supporting avx-256 than 28 cores with avx-512.
Besides Intel Has created a huge mess with that extension.

https://pbs.twimg.com/media/DXSRqJXX...jpg&name=small

[nevcairiel]

Originally Posted by Atak_Snajpera

Besides Intel Has created a huge mess with that extension.

https://pbs.twimg.com/media/DXSRqJXX...jpg&name=small

That "mess" looks far better if you remove the Knights Landing/Knights Mill stuff, which is only on Xeon Phi and of no relevance to something like x265 anyway - and Cannon Lake is dead, so could also remove that circle.

So what software currently does is develop against the Skylake-X subset (like x265 did), or the Icelake subset if they want to use more advanced features in the future. Two levels to care about, really. Before AVX512, we also had a number of small additional sets to augment SSE and AVX, but those get rarely talked about, since many of them arrived with AVX(2) at the same time. There were FMA3 and FMA4 of course (one for Intel, one for AMD, although AMD switched over in later CPUs), BMI1 and BMI2, XOP, and probably more that are of no direct video relevance that I forgot.

[benwaggoner]

Originally Posted by Atak_Snajpera

Avx512 won't help xeons when AMD just Has epyc2/threadriper 3990x with 2x more cores
Personally I prefer 2x more cores supporting avx-256 than 28 cores with avx-512.

I'm testing on the hardware I have, which is basically a c5.18xlarge.

[benwaggoner]

Originally Posted by benwaggoner

I'm testing on the hardware I have, which is basically a c5.18xlarge.

Anyhoo, I encoded 2400 frames of some 8K source with preset slower with and without AVX512.

Just one test run, but I got 1.41 fps with AVX512 and 1.38 without. Also, my office got warm .

[Blue_MiSfit]

Originally Posted by Magik Mark

Oh I see. Can you refer me to the table detailing max buffer for every profile? Thanks


Sent from my iPhone using Tapatalk

https://en.wikipedia.org/wiki/High_E...ers_and_levels

[estir]

Originally Posted by froggy1

While SSIM is (much) better than PSNR, it isn't perfect.

It's not much better. it's not even always better than PSNR.
Anyway PSNRHVS is better than both of them (better even than MS-SSIM), but for some reason no modern codec uses it for RDO.

[filler56789]

x265.exe 3.2+29-c5ff729 has arrived.

http://msystem.waw.pl/x265/

[Barough]

x265 v3.2+31-6b348d5b56d8 (32 & 64-bit 8/10/12bit Multilib Windows Binaries) (GCC 9.2.0)

https://bitbucket.org/multicoreware/x265/commits/branch/default

[l00t]

Originally Posted by jlpsvk

yeah... avx512 produces much more heat... so probably safety issue...

But why? The CPU is at 60oC at 100%, it should allow higher clocks...

[redbtn]

Umm, hdr-opt replaced to hdr10-opt. Is it mean that hdr-opt does nothing in previous releases?

[microchip8]

Originally Posted by redbtn

Umm, hdr-opt replaced to hdr10-opt. Is it mean that hdr-opt does nothing in previous releases?

No, it got replaced to make it more clear that it's meant for HDR10 content

[redbtn]

Originally Posted by froggy1

No, it got replaced to make it more clear that it's meant for HDR10 content

Yeah, I mean it does nothing for hrd content. I used it for a year, thinking it doing optimization for HDR.

[microchip8]

Originally Posted by redbtn

Yeah, I mean it does nothing for hrd content. I used it for a year, thinking it doing optimization for HDR.

How do you know?

[iwod]

I am wondering if x265 has any plans for VVC / H.266? We are may be 6 months away from it being finalised.

[redbtn]

Originally Posted by froggy1

How do you know?

This patch deprecates the existing option '--hdr-opt' as the name sounds generic
even though the feature does QP optimizations only for HDR10 content

So i guess for HDR it doesn't nothing.

[nevcairiel]

Originally Posted by redbtn

So i guess for HDR it doesn't nothing.

"HDR" by itself means nothing. Its a marketing name, and does not describe any sort of specific video format for which one could optimize for.

What kind of HDR? HDR10/PQ? HLG? DV?

Thats why the name of the option was changed. HDR by itself is a category of formats, not any specific format one could apply any optimizations for.

[Boulder]

As far as I know, --hdr-opt simply did/does some adjustments to the encoding parameters. You could apply it to SDR content if you like. Not that it would be useful though..

[benwaggoner]

Originally Posted by Boulder

As far as I know, --hdr-opt simply did/does some adjustments to the encoding parameters. You could apply it to SDR content if you like. Not that it would be useful though..

It would be counterproductive in fact. IIRC, hdr10-opt adjusts delta qp for chroma based on luma levels to better match the characteristics of the PQ EOTF. By default x265 is more optimized for gamma. hdr10-opt would thus be bad for 709 or HLG. Dolby Vision, except in profiles that have a backwards compatible PQ base layer, doesn't even use Y'Cb'Cr, and in Profile 5 does some crazy dynamic range adjustments, so not helpful there.

There's probably fruitful research to be done for how to adapt chroma qp relative to luma based on each macroblock's luma levels in a psychovisual model.

[utack]

I am surprised how much difference "slower" makes compared to "slow"
It is probably expected due to all the extra bframes giving a quality gain and much more lookahead making different choices what to prioritize, but was not fully aware of it before testing.