[ / / / / / / / / / / / / / ] [ dir / arepa / ck / cumshop / fascist / hisrol / imouto / loomis / trs ][Options][ watchlist ]

/tech/ - Technology

You can now write text to your AI-generated image at https://aiproto.com It is currently free to use for Proto members.
Email
Comment *
File
Select/drop/paste files here
Password (Randomized for file and post deletion; you may also set your own.)
* = required field[▶ Show post options & limits]
Confused? See the FAQ.
Expand all images

New Volunteer

[–]

 No.940336>>940338 >>940357 >>940406 >>940581 >>940637 [Watch Thread][Show All Posts]

I've never experimented with hardware video encoding until now, and my findings are that it's complete and utter dogshit. Yes, it can encode at speeds higher than 1K fps, but the result is always a larger file. It doesn't matter what preset I use. None of them compare to CPU placebo. The reason I bring this up is because AV1 absolutely must have hardware accelerated encoding to be usable. Is this simply Nvidia's implementation for x264 and x265 which are shit, or is all hardware encoding this poor?

 No.940338>>940343 >>940381

>>940336 (OP)

Hardware video encoding is mostly intended for streamerfags with shit CPUs.


 No.940343>>940359 >>940381

It's Nvidia's implementation.

>>940338

Hardware encoding / decoding goes beyond CPUs and GPUs. Anyone who actually needs to do video encoding / decoding uses hardware acceleration. Do you honestly think that big platforms like YouTube and Netflix do it all in software?


 No.940357

>>940336 (OP)

>The reason I bring this up is because AV1 absolutely must have hardware accelerated encoding to be usable.

Hardware encoding and hardware accelerated encoding aren't the same thing. x264 has support for some GPU acceleration through OpenCL.


 No.940359>>940371

>>940343

They probably do transcodes in software. The current video encoding algorithms don't parallelize well so you get a choice of a limited form that parallelizes well or a quality form that needs a CPU that efficiently handles branching.


 No.940371

>>940359

I don't know how widespread it is, but I do know that Netflix used an FPGA implementation at least at some point in time.


 No.940381>>940397

>>940343

Why not? You only need to encode once for a given resolution and then send the files around.

Hardware is more important for decoding.

As >>940338 pointed out, it's when you need realtime encoding for streaming purposes that there is any point.


 No.940386>>940406

>but the result is always a larger file.

I noticed this using intel GPU hardware acceleration too. Do they use the same algorithm as to make the size so bloated?


 No.940397>>940398

>>940381

Yes, but when you are working at a large scale the savings you get by using dedicated hardware is worth it.


 No.940398

>>940397

*in comparison of dedicating the eqivalent amount of "non-dedicated hardware" to that problem.


 No.940406>>940428 >>940580

>>940386

That's the cost of saving time. You either get a quick encode at a larger file size/less quality or a smaller encode with better quality that takes more time.

>>940336 (OP)

>The reason I bring this up is because AV1 absolutely must have hardware accelerated encoding to be usable.

No it doesn't. Divx/Xvid encoding was done at 1fps and less in the late 90s. x264 was the same when it first came out. Fansub encoders used to use filters that caused encode times to go down below 0.10fps and would let a machine run for days for the finale encode. These are not new problems newfags are just accustomed to real time streaming and the software being mature.


 No.940428

>>940406

Patience is dead


 No.940553

Nvidia always chooses the faster and lower quality route for everything.


 No.940580>>940582 >>940593

>>940406

Anon, there's a reason fansubbers prefer HEVC over VP9 despite the latter performing better at low bitrates.

Encoding 23 minutes of 320x240p24 AV1 for days on end may have been attractive during the early days of digital video but these days it's plain inexcusable when x265 can deliver the same quality with a mild increase in filesize in a matter of hours.


 No.940581>>940582

>>940336 (OP)

>AV1 absolutely must have hardware accelerated encoding to be usable

1) The reference encoder is not optimized AT ALL. It's stupid to compare its speed with others.

2) rav1e will probably end up as the default encoder.


 No.940582>>940586 >>940591

>>940580

>Anon, there's a reason fansubbers prefer HEVC over VP9 despite the latter performing better at low bitrates.

Fansubbers don't know how to encode anymore, boy.

>Encoding 23 minutes of 320x240p24 AV1 for days on end may have been attractive during the early days of digital video but these days it's plain inexcusable

See >>940581

>x265 can deliver the same quality with a mild increase in filesize in a matter of hours

Only if you know how to tune it to avoid landing in blurland. Mainly --no-sao and --no-strong-intra-smoothing; x265 is actually where x264 was before it started disregarding metrics and going full psy optimizations.

Why do I have faith in AV1? Because Xiph is behind rav1e, and the people at Xiph really know their shit.


 No.940586>>940591

>>940582

>Xiph

This. The Xiph guys seem to know their shit.


 No.940591>>940594

>>940586

>>940582

How does rav1e compare to aomenc when using identical bitrates+available options?

I heard it omits many of AV1's advanced features for the sake of speed.


 No.940593

>>940580

>Anon, there's a reason fansubbers prefer HEVC over VP9 despite the latter performing better at low bitrates.

Nice rant that had nothing to do with what I was talking about. You just saw me mention fansubs and went full retard. I don't really care what lazy faggots these days are using to re-encode crunchyroll rips with.


 No.940594

>>940591

The 1.0 spec (bitstream freeze) was JUST reached, m8. You'll have to wait AT LEAST 6 months to get something very good (could be lower seeing all the money behind it, though).


 No.940637>>940667

>>940336 (OP)

Still not much support for hwenc on AMD + linux, is there?


 No.940667>>940669

>>940637

VAAPI can do hardware encoding and decoding through AMD GPUs. which means anything with VAAPI support has it.


 No.940669

>>940667

Anecdote: My Radeon 5670HD with the radeon KMS driver does tear-free decodes with mpv --vo=vaapi.


 No.940670>>940672

Is AV1 really an "open" standard if only Google and Twitch end up with a usable, proprietary encoder that they won't share?


 No.940672

>>940670

Reas the posts above you, retard.




[Return][Go to top][Catalog][Screencap][Nerve Center][Cancer][Update] ( Scroll to new posts) ( Auto) 5
24 replies | 0 images | Page ?
[Post a Reply]
[ / / / / / / / / / / / / / ] [ dir / arepa / ck / cumshop / fascist / hisrol / imouto / loomis / trs ][ watchlist ]