Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Video Encoding > VP9 and AV1

Reply
 
Thread Tools Search this Thread Display Modes
Old 31st October 2019, 14:09   #1901  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
There was some sort of AOM event a couple of weeks ago apparently.

A load of presentation slides can be found at this link.
soresu is offline   Reply With Quote
Old 31st October 2019, 19:44   #1902  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Lots of Deep, Neural and ML related stuff discussed at the AOM event it seems - I think we can guess the main direction of AV(x) codecs in the future.
soresu is offline   Reply With Quote
Old 31st October 2019, 19:46   #1903  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by soresu View Post
Unfortunately not fast enough, I bought an Amazon Fire Stick for my dad in 2017 assuming that the advertised HEVC support meant up to 10 bit, only to find out to my horror that most of the HEVC encoded videos I have did not work on it.

At least the newer FS 4K I bought this year has 10 bit capability, still the experience was somewhat disheartening considering the sheer amount of 10 bit content available at the time I bought the first Fire Stick, 5 years after HEVC was standardised.
Any HDR-capable Fire Stick/TV supports 10-bit decode. The feature did come to FireTV first, though, a couple of generations back.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 31st October 2019, 19:53   #1904  |  Link
benwaggoner
Moderator
 
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,750
Quote:
Originally Posted by soresu View Post
Lots of Deep, Neural and ML related stuff discussed at the AOM event it seems - I think we can guess the main direction of AV(x) codecs in the future.
Do you mean the encoders or the bitstream definition itself?

I get nervous about actually tuning bitstream features based on ML, because we still lack in well subjectively-correlated metrics. VMAF is the least bad ever, but is SDR only. and the whole question of how individual frame metrics get aggregated into good metrics for interframe encoding remains barely examined. A mean of individual frame values isn't that useful for a clip that is more than a few second of a single shot.

We've seen that AV1 shows better VMAF to MOS ratios than other codecs, which could be a result of this sort of curve fitting to one metric. It's generally true that the more a metric gets used, the lower its subjective correlation becomes, as encoders get increasingly tuned to the metric instead of to subjective ratings.
__________________
Ben Waggoner
Principal Video Specialist, Amazon Prime Video

My Compression Book
benwaggoner is offline   Reply With Quote
Old 31st October 2019, 21:21   #1905  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Quote:
Originally Posted by benwaggoner View Post
Do you mean the encoders or the bitstream definition itself?

I get nervous about actually tuning bitstream features based on ML, because we still lack in well subjectively-correlated metrics. VMAF is the least bad ever, but is SDR only. and the whole question of how individual frame metrics get aggregated into good metrics for interframe encoding remains barely examined. A mean of individual frame values isn't that useful for a clip that is more than a few second of a single shot.

We've seen that AV1 shows better VMAF to MOS ratios than other codecs, which could be a result of this sort of curve fitting to one metric. It's generally true that the more a metric gets used, the lower its subjective correlation becomes, as encoders get increasingly tuned to the metric instead of to subjective ratings.
I've only skimmed 2-3 of the presentation slide decks, but there is certainly an appreciation for bitstream compatibility where the ML models are concerned.

Though Google at least have been knocking on this particular door for at least a few years now, I figure they would not still be at it if they thought it could not be harnessed in a standardised way for a codec bitstream.

Just to be clear, my own level of understanding of all of this is fairly amateur compared to experts on here - I mostly have an avid interest in codecs and more recently ML too (due to various ML optimisations in CG rendering and production fields).

Last edited by soresu; 31st October 2019 at 21:23.
soresu is offline   Reply With Quote
Old 1st November 2019, 19:59   #1906  |  Link
soresu
Registered User
 
Join Date: May 2005
Location: Swansea, Wales, UK
Posts: 196
Found an interesting slide deck called"Adaptive Optimal Linear Estimators for Enhanced Motion Compensated Prediction".

It goes over several things, though I'm not sure if they are alternative solutions or potentially additive improvements.

Assuming they are additive, it discusses at least an average 11.5% BD rate improvement over baseline (presumably AV1 is the baseline).

Link here.

Last edited by soresu; 1st November 2019 at 20:01.
soresu is offline   Reply With Quote
Old 2nd November 2019, 11:00   #1907  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
TLDR:
Win7 64bits, i7-4770k, 3.40GHz (stock), improvement between 0.2.1 and 0.5.1 using only SSSE3 accelerated routines, single thread:
Chimera: 33.2%
Dua Lipa: 34.9%

Included are some AVX2 tests too, because yes.
Ok, so you replied to different issues than those I questioned with my results, but still your results are interesting.
It seems that single threaded performance has increased for SSSE3 but AVX2 over SSSE3 is very tiny.
Still, both SSSE3 and AVX2 in single threaded mode are ~30 something % faster for 0.5.1 vs 0.2.1

BUT as I have already stated in my results, the CPU utilization during real-world multi-threaded decoding, eats ALL of the single threaded performance in case of Dua Lipa clip and most of the single threaded gain for the other clips for both SSSE3 and AVX2 versions.

In other words, for the real-world multi-threaded decoding the absolute sum of gain and loss between 0.2.1 vs 0.5.1 is dead zero for Dua Lipa and so small for the other clips.

Sorry, but I can't call this situation as progress after seven months, if overall multi-threaded decoding performance gain is zero.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 11:49   #1908  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
FFMpeg says it's going to use 4 frame threads and 3 tile threads to decode the files, so I'll be using those numbers.



Chimera: 34%

Dua Lipa: 29.7%
Ok, now your results are even more interesting.
Where can I can I find those executables of 0.2.1 and 0.5.1 versions to run them on my systems ?
I have used the LAV filters versions posted above.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 12:02   #1909  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
FFMpeg says it's going to use 4 frame threads and 3 tile threads to decode the files, so I'll be using those numbers.



Chimera: 34%

Dua Lipa: 29.7%
Also, I have to say that you are using half of your threading power meaning only 50% CPU logical utilization as you have an 8 threaded CPU and you are using only 4 threads.
I think you have to test it again with at least 8 threads in order to use hyperthreading and all of your CPU's processing power.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 14:45   #1910  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
Is there any particular reason DXVA Checker grays out the CPU usage line when I try to do the benchmarks? Is there some option I need to set?
SmilingWolf is offline   Reply With Quote
Old 2nd November 2019, 15:06   #1911  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
Is there any particular reason DXVA Checker grays out the CPU usage line when I try to do the benchmarks? Is there some option I need to set?
You only have to run the benchmark (I run it 3 times and take the average) and for every run you will see at the end the CPU usage.
It has also min/avg/max value, even for CPU usage.
Just leave it to finish.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 15:50   #1912  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
Yeah I tried doing that and it didn't work, see screenshots


Last edited by SmilingWolf; 2nd November 2019 at 16:02.
SmilingWolf is offline   Reply With Quote
Old 2nd November 2019, 16:06   #1913  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
Yeah I tried doing that and it didn't work, see screenshots

Ok, you have to go to LAV filters settings and choose SW decoding.
It seems to me that you are using HW decoding.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 16:24   #1914  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
I did wonder if that was the case, but when I open LAVFilters' config panel this is what I see:


Fresh installation, nothing touched
SmilingWolf is offline   Reply With Quote
Old 2nd November 2019, 16:59   #1915  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
Quote:
Originally Posted by SmilingWolf View Post
Fresh installation, nothing touched
Latest DXVA Checker v4.2.1 and Connect to Renderer selected ?
Also, when you select the AV1 file, does LAV say unsupported inside DXVA Checker ?
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all
NikosD is offline   Reply With Quote
Old 2nd November 2019, 17:19   #1916  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
Latest DXVA Checker v4.2.1, system equipped with a GTX1080 (440.97).

It says Unsupported, yes.

This version does not seem to have a "Connect to Renderer" option anywhere.
SmilingWolf is offline   Reply With Quote
Old 3rd November 2019, 07:26   #1917  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
Quote:
Originally Posted by NikosD View Post
Ok, but SmilingWolf and you, have tested different things than me.
Firstly, he posted single threaded performance difference and I posted multi-threaded performance difference
Conveniently forgetting about my two posts dedicated to multi threaded performance aren't we?
http://forum.doom9.org/showthread.ph...74#post1889274
http://forum.doom9.org/showthread.ph...89#post1889289

There isn't a DXVA Checker report yet, afternoon spent trying to make it work notwithstanding, but as I said, CPU utilization goes between 70% and 90% with the two sequences used.
SmilingWolf is offline   Reply With Quote
Old 3rd November 2019, 08:21   #1918  |  Link
littleD
Registered User
 
littleD's Avatar
 
Join Date: Aug 2008
Posts: 343
Some older versions of dxva checker shows CPU usage. But i did short test and the results of dxch was around 88% utilization while system monitor was showing 100%. Maybe thats why authors of the program turned off the feature temporarily, because of inconsistent results?
littleD is offline   Reply With Quote
Old 3rd November 2019, 08:50   #1919  |  Link
SmilingWolf
I am maddo saientisto!
 
SmilingWolf's Avatar
 
Join Date: Aug 2018
Posts: 95
I'd love to keep arguing, but I have to agree that won't make the board any favor, so let's let bygones be bygones.

I have already removed all sorts of config files, fresh installations, even reboots etc.
Have you run your own benches on this (4.2.1) version, or on an older one? If the latter happens to be the case, what exact version, so I can download it from the VideoHelp archive?

I don't get what you mean by "no internal commands". They are cmdline applications, just use the same cmdlines I used. I was inside an MSYS shell just so that I could use the "time" command, but I suppose PowerShell on Win10 has got something similar.
Word of advice, they only digest pure IVF files, so at least the Dua Lipa video will have to be freed of its container using "ffmpeg -i Dua_Lipa.mp4 -c:v copy Dua_Lipa.ivf"
SmilingWolf is offline   Reply With Quote
Old 3rd November 2019, 09:45   #1920  |  Link
NikosD
Registered User
 
Join Date: Aug 2010
Location: Athens, Greece
Posts: 2,901
I'm not a developer, so environments like MSYS , Visual Studio etc are not frequently installed on my system.
I have compiled a few apps from time to time, even my own code decades ago (!) but I'm not going to do it now setting up MSYS.
I'll give PowerShell a try of course, as I use it from time to time for my job (although I still do a lot using cmd)
But that IVF thing is another obstacle.
Regarding DXVA Checker I used v4.2.1 which of course has everything, as I told you before.
Min/avg/max for FPS and CPU utilization.
Your main problem is that you see things like Video Engine and GPU utilization and you shouldn't.
You need a cleaner OS.
Tomorrow I'll try setting LAV to single-thread mode and run the same tests with Skylake at work.
If nobody here in this forum can confirm or reject my multi-thread results using so familiar tools like LAV filters and DXVA Checker, I'll try to reproduce yours single-thread results.

P.S
DXVA Checker is a sophisticated and accurate tool and the CPU utilization refers to itself only, as a process, not general CPU utilization during its running.
__________________
Win 10 x64 (19042.572) - Core i5-2400 - Radeon RX 470 (20.10.1)
HEVC decoding benchmarks
H.264 DXVA Benchmarks for all

Last edited by NikosD; 3rd November 2019 at 09:49.
NikosD is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 11:55.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.