madVR - high quality video renderer (GPU assisted) [Archive] - Page 932

mzso

15th October 2017, 12:23

Looking at your first log, I can see that playback starts with "Oldboy", and seems to be running ok for about 2 seconds. Then playback switches to "John Wick", but before playback can even start properly, it switches back to "Oldboy", and then to "Cube". After it switched to "Cube", the decoder doesn't send any more video frames to madVR, for some reason.

Log 2 tries to play "John Wick", but the decoder never even sends a single frame.

A couple of questions:

1) Is the media player frozen, or is the media player GUI still responsive?
2) Does the madVR OSD (Ctrl+J) still work? If so, I suppose the decoder queue is empty in these situations, right?
3) Do you have the optionactivated? If so, does deactivating that option help?

The GUI is totally frozen. I see some info updating on the madVR OSD: display frequency, max stats.
All the cues are empty. Also the line that shows the video file's resolutions is weird. It has an arrow in it like the scale line and doesn't start with "movie".

I have "delay playback start until render queue is full" disabled because it increases seek or playback start times which I didn't like.

https://abload.de/img/frozen-pp-madvr-osd7kbb1.png

Normal GUI/OSD appearance:
https://abload.de/img/pp-normal-gui5wxeu.png

Razoola

15th October 2017, 12:35

Interesting thing is though, when I'm in my custom mode, nvidias control panel doesn't have an output color depth selected

https://i.snag.gy/7w9TKV.jpg

the field is empty and only 8bpc is available to choose

This is standard behaviour if you switch to a mode that doesn't have 10/12bit available, like when you switch from 30Hz 12bit to 60Hz, the field will be empty and only 8bit will be available to choose. The difference is, when I'm switching to 60Hz, I'm actually switching to 8bit, but when I switch from 30Hz 12bit to my custom resolution, I stay in 12bit, even though it's not available to choose afterwards. Confirmed visually with the 16bit gradient png with no dithering + the app for my AVR shows "RGB Full / 36bit" which is the 12bit mode, 8bit mode would show 24bit.
If I'm in my custom resolution and try to change bitdepth from "empty" to 8 bit, I get a black screen again. So I'm definitely in 12bit here (right?)

I see the samw here but I think this is a nvidia control panel bug, my avr also shows 12bit while its blanked out in control panel.

Razoola

15th October 2017, 12:38

Are you sure it's the Fall Creators Update doing this? Because the latest nvidia driver 387.92 is causing exactly what you just described, even without the Fall Creators Update.

I second this. I was gonna bring it up later in this forum as I had a situation also yesturday for the first time that HDR did not switch off after video playback, It only happened after I installed the latest drivers.

I think its nvidia driver related and not madVR or windows..

madshi

15th October 2017, 12:46

Anyway, I'm up to "no drops/repeats" for 23.976, 50 and 59.94Hz. I have perfect timings for 50 out of the box, I had 1 drop every 2 days before in CRU, through madVR I optimized it perfectly, I optimized 59.94 in madVR too. Then I removed everything and I inserted the calculated values in CRU. Which worked with 59.94.

I'd say I'm 99% there. :)

As a side note... I've been an Nvidia HTPC user for more than 10 years. The mess that they had back then and the mess that they still have today concerning custom resolutions is beyond belief.

For the life of me I can't understand how a multi-billion company can neglect a portion of its control panel for so long.
Yeah, it's disappointing, but I hope to be able to get them to fix the most obvious bugs at some point in the future.

Thanks for the reply. Just to confirm your test setup. You keep madvr at 10-bit or higher
No, I believe he set everything to 8bit everywhere. If you set madVR to 10bit, then madVR will not dither down to 8bit, which is necessary if your TV doesn't handle 10bit well.

i'll do that from now on. Here's the comparison i should have posted. The grain and a bit of aliased edges are standing out for me. If that's tweaked i'm all aboard the new AS.

no processing
https://postimg.org/image/j79imzhv3/
upscale + olsAS
https://postimg.org/image/5dl5xz7a7/
upscale + newAS
https://postimg.org/image/rcrkl7b9r/
Ok, igv updated AdaptiveSharpen again yesterday. His latest version doesn't sharpen grain as much, anymore, but still more than the original AS version. If you don't like the grain, why don't you use "reduce random noise" in madVR? Also, your original image is quite visibly distorted by compression artifacts. Seems not all of them originated from JPEG, but some were already in your original video. The new "reduce compression artifacts" option works pretty well for this kind of problem. Here's how your image looks like with the following settings, using igv's latest AdaptiveSharpen version:

reduce compression artifacts: 5
reduce random noise: 1

http://madVR.com/doom9/stuff/asRnnRca.png

Maybe I've gone a bit overboard with the "reduce compression artifacts" strength, but with Anime content high values often barely harm.

The GUI is totally frozen. I see some info updating on the madVR OSD: display frequency, max stats.
All the cues are empty. Also the line that shows the video file's resolutions is weird. It has an arrow in it like the scale line and doesn't start with "movie".

I have "delay playback start until render queue is full" disabled because it increases seek or playback start times which I didn't like.
As you can see in the image, the decoder queue is empty. It's hard for me to say why. There appears to be a deadlock somewhere. It could be madVR's fault, but it could also be a complex deadlock problem where all madVR, media player and LAV are somehow involved.

Have you tried other media players like e.g. MPC-HC/BE? Does the problem occur with those, too? I'm not asking to make you switch, just trying to collect information.

mzso

15th October 2017, 12:52

As you can see in the image, the decoder queue is empty. It's hard for me to say why. There appears to be a deadlock somewhere. It could be madVR's fault, but it could also be a complex deadlock problem where all madVR, media player and LAV are somehow involved.

Have you tried other media players like e.g. MPC-HC/BE? Does the problem occur with those, too? I'm not asking to make you switch, just trying to collect information.

Not really. I'll try to reproduce a hang in one of them.

Razoola

15th October 2017, 13:05

I had thought about this, but testing actually still works, so should I really reject testing already in this situation?

Its a hard one because I have had situation with 095.6.1 where it would not allow to test when trying to optimise because the display was already at the custom resolution. another time it allows the test but rejected the save. Only when the starting refresh rate was different before the initial testing process did the all seem to complete without error.

Personally from the experiances I have had so far I think you should confirm the resolution and refresh rate is different than the resolution and refresh rate to be tested at the point the 'test mode' is pressesd. This might only apply to nvidia gfx cards. Wether you switch resolutions first silently, give a message and then switch resolutions or full error telling user to change resolution manually is up to you.

I guess what makes the situation worse here is the fact there is a high probability the resolution/refresh rate will be the same already during a user trying to setup a custom resolution or optimise. This is because they are probably going to the madVR custom resolution options right from a paused movie which is playing at the refresh rate they want to play with (because madVR set the closest resolution when playback started).

mzso

15th October 2017, 13:09

Update:
So I could reproduce a hang, but only one with mpc-hc (internal LAV):

https://abload.de/img/mpc-hang8dakf.png

I can't reproduce a hang using internal codecs with Potplayer.
(Sadly using the internal codecs is not really an option, because it also enables the craptastic built-in video processing filter...)

madshi

15th October 2017, 13:14

Update:
So I could reproduce a hang, but only one with mpc-hc (internal LAV):

https://abload.de/img/mpc-hang8dakf.png
Good. Now in order to get to the bottom of this, what we need is at least one freeze report, ideally several, but with the "debug symbols" (*.PDB files) in the right folder for both MPC-BE and LAV. This way the freeze report will not only tell us what madVR is doing at the time of the freeze, but also what MPC-BE and LAV are doing. This *may* (or may not) help figuring out what's going on.

PDB debug symbols for LAV are available in the first post of the LAV thread, IIRC, but the debug symbols MUST match the LAV version installed *exactly* otherwise they're doing more harm than good. PDB debug symbols for MPC should be available somewhere, too, not sure where, though.

Jtacdf

15th October 2017, 13:22

Ok, igv updated AdaptiveSharpen again yesterday. His latest version doesn't sharpen grain as much, anymore, but still more than the original AS version. If you don't like the grain, why don't you use "reduce random noise" in madVR? Also, your original image is quite visibly distorted by compression artifacts. Seems not all of them originated from JPEG, but some were already in your original video. The new "reduce compression artifacts" option works pretty well for this kind of problem. Here's how your image looks like with the following settings, using igv's latest AdaptiveSharpen version:

reduce compression artifacts: 5
reduce random noise: 1

http://madVR.com/doom9/stuff/asRnnRca.png

Maybe I've gone a bit overboard with the "reduce compression artifacts" strength, but with Anime content high values often barely harm.

It's definitely overboard.
Have a look at the samples below with the same above settings. Comes from the same episode.
It's something that has to be use with care or be turned off with quality high resolution content.

Original https://www.mediafire.com/file/iq6epk00eef57e0/Off.png

Rca-5, rra-1 https://www.mediafire.com/view/iwdjzibnwjx9mcs/RCA-5%2C%20RRA-1.png

That said, the new features does work magic on sd and poorly mastered sd-to-hd anime content. Thanks madshi!

madshi

15th October 2017, 13:29

It's definitely overboard.
Have a look at the samples below with the same above settings. Comes from the same episode.
It's something that has to be use with care or be turned off with quality high resolution content.

Original https://www.mediafire.com/file/iq6epk00eef57e0/Off.png

Rca-5, rra-1 https://www.mediafire.com/view/iwdjzibnwjx9mcs/RCA-5%2C%20RRA-1.png

That said, the new features does work magic on sd and poorly mastered sd-to-hd anime content. Thanks madshi!
Yeah, there's clear texture loss in this example. I was just turning all the knobs up to eleven to get rid of those nasty compression artifacts. Much lower values already help a lot, too, of course, and should not damage textures as much.

mparade

15th October 2017, 13:50

Hello,

Is it possible to set up a separate setting profile for 3D frame-packed content? My standard 2D "1080p" profile settings make my 1080 Ti video card "exploding" (rendering times ~doubles) when playing 3D due to doubling each frame instead of only 1pc of 2D frame.

Is it the correct behaviour of madVR to upscale BD sources/DVD sources to 4320x2430p? It is a bit far from 3840x2160p. This behaviour also has negative impact on my JRivers UI (stayed "zoomed in" even after finishing play).

Is it the correct behaviour of madVR to output at 8bit
even if chosing 10bit as "the native display bitdepth"?
My source is 8bit (BD mostly), but my TV can handle 10bit surely.

Any help would be appreciated.

mzso

15th October 2017, 14:08

Good. Now in order to get to the bottom of this, what we need is at least one freeze report, ideally several, but with the "debug symbols" (*.PDB files) in the right folder for both MPC-BE and LAV. This way the freeze report will not only tell us what madVR is doing at the time of the freeze, but also what MPC-BE and LAV are doing. This *may* (or may not) help figuring out what's going on.

PDB debug symbols for LAV are available in the first post of the LAV thread, IIRC, but the debug symbols MUST match the LAV version installed *exactly* otherwise they're doing more harm than good. PDB debug symbols for MPC should be available somewhere, too, not sure where, though.

I tried... Here they are: https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk
But the symptoms are not quite identical to Potplayer. The madVR OSD doesn't appear. I get a windows alert sound after a few secs, and sometimes MPC gets unhung and the playback starts. Also hanging seems to occur differently, when I clobber one file instead of when I change between files.

famasfilms

15th October 2017, 14:14

Are you sure it's the Fall Creators Update doing this? Because the latest nvidia driver 387.92 is causing exactly what you just described, even without the Fall Creators Update.

I have the latest Nvidia driver, installed last fortnight or so - installed the Fall update last night and immediately tested HDR.

You could be right

jkauff

15th October 2017, 14:27

Its a hard one because I have had situation with 095.6.1 where it would not allow to test when trying to optimise because the display was already at the custom resolution. another time it allows the test but rejected the save. Only when the starting refresh rate was different before the initial testing process did the all seem to complete without error.

Personally from the experiances I have had so far I think you should confirm the resolution and refresh rate is different than the resolution and refresh rate to be tested at the point the 'test mode' is pressesd. This might only apply to nvidia gfx cards. Wether you switch resolutions first silently, give a message and then switch resolutions or full error telling user to change resolution manually is up to you.

I guess what makes the situation worse here is the fact there is a high probability the resolution/refresh rate will be the same already during a user trying to setup a custom resolution or optimise. This is because they are probably going to the madVR custom resolution options right from a paused movie which is playing at the refresh rate they want to play with (because madVR set the closest resolution when playback started).
As I mentioned in my earlier post, my custom resolution success on my GTX 1060 was achieved by avoiding the Nvidia Control Panel altogether. I did all my resolution changes through the Windows 10 Display Adapter Properties dialog, launched from the Display Settings page. I'm running the latest Nvidia drivers.

I can't say this will work for everyone, but it did for me.

foozoor

15th October 2017, 14:30

Where is the v0.92.7 changelog?

omw2h

15th October 2017, 14:39

Ok, igv updated AdaptiveSharpen again yesterday. His latest version doesn't sharpen grain as much, anymore, but still more than the original AS version. If you don't like the grain, why don't you use "reduce random noise" in madVR? Also, your original image is quite visibly distorted by compression artifacts. Seems not all of them originated from JPEG, but some were already in your original video. The new "reduce compression artifacts" option works pretty well for this kind of problem. Here's how your image looks like with the following settings, using igv's latest AdaptiveSharpen version:

reduce compression artifacts: 5
reduce random noise: 1

http://madVR.com/doom9/stuff/asRnnRca.png

Maybe I've gone a bit overboard with the "reduce compression artifacts" strength, but with Anime content high values often barely harm.

A sample of one distorted image might not be fair, but it shows my point where i find the end result of the new AS
lacking in comparison and i find no shorige of bad content.
I'm using the awesome reducing random noise and compression artefacts when needed, but in this situation i have to
use them, and this solution brings the questions of subtle harm and more GPU freedom being lost (then just having the previus AS checked).
If other options come to light more joy to me, i did caught on that the new AS is stronger,
so if i tone it down a bit and if the grain is not as visible i guess i'll survive the change.

Siso

15th October 2017, 14:49

Where is the v0.92.7 changelog?

Not yet released.

foozoor

15th October 2017, 14:50

Not yet released.
So there is something wrong on the official madVR website.

Sebastiii

15th October 2017, 14:53

madshi is surely preparing the upload and website :P

Siso

15th October 2017, 14:56

So there is something wrong on the official madVR website.

What's wrong?

foozoor

15th October 2017, 14:57

http://madvr.com
latest release v0.92.7

Siso

15th October 2017, 14:58

http://madvr.com
latest release v0.92.7

Just be patient, dude :D

Asmodian

15th October 2017, 15:02

It looks like madshi actually controls that website now?!

I swear that didn't use to be the case. :confused:

I am very glad madVR.com is in good hands. :)
:thanks:

http://madvr.com
latest release v0.92.7

If you install that version its version number is 0.92.6.

ashlar42

15th October 2017, 15:28

Yeah, it's disappointing, but I hope to be able to get them to fix the most obvious bugs at some point in the future.You know what? I'm gonna say a simple thing.

I hope everybody here realizes how lucky we are to have you.

austinminton

15th October 2017, 15:37

No, I believe he set everything to 8bit everywhere. If you set madVR to 10bit, then madVR will not dither down to 8bit, which is necessary if your TV doesn't handle 10bit well.

I mean what setup is he using to reach the conclusion that 8 bit is better than 10 bit for him. By keeping madvr at 8 bit hes defeating the whole purpose of doing this test? His tv will never get the 10 bit signal?

thewildsun

15th October 2017, 16:27

I'm encountering a strange problem where video playback is smooth for the first 6-8 minutes or so, and then I start dropping frames like crazy. Pausing and then playing again doesn't help, but restarting the video player (PotPlayer, although I've also tried MPC-HC) fixes the problem immediately... for the next 6-8 minutes. Task manager doesn't show anything that stands out while frames are being dropped -- no spikes in CPU, GPU, or RAM.

It doesn't seem like a hardware issue, as madVR is the only thing that has seems to have any issues. I'm on the latest windows 10 insider build, so this might be the culprit, but it only happens on my NVIDIA GTX 1070 machine. I have another desktop with the AMD RX-480 that's also on the latest insider build, but the problem doesn't happen there. I've tried old NVIDIA drivers and old madVR builds, but had no luck.

Has anyone encountered anything like this before?

oldpainlesskodi

15th October 2017, 16:34

You know what? I'm gonna say a simple thing.

I hope everybody here realizes how lucky we have to have you.

Amen.

K

foozoor

15th October 2017, 16:46

Just be patient, dude :D
I suspect madshi wanted to fool websites like majorgeeks or videohelp. :D

cork_OS

15th October 2017, 17:01

I suspect madshi wanted to fool websites like majorgeeks or videohelp. :D
videohelp is pretty useful site and it still have 0.92.6 listed.

mrcorbo

15th October 2017, 18:49

madshi

15th October 2017, 19:01

Is it possible to set up a separate setting profile for 3D frame-packed content?
Yes, there should be a working "3D" profile condition supported.

Is it the correct behaviour of madVR to upscale BD sources/DVD sources to 4320x2430p?
I don't know. Show me a screenshot of the OSD (Ctrl+J), and I might be able to say more.

Is it the correct behaviour of madVR to output at 8bit
even if chosing 10bit as "the native display bitdepth"?
My source is 8bit (BD mostly), but my TV can handle 10bit surely.
10bit output is only possible with D3D11 presentation, and in older OSs only in FSE mode.

I tried... Here they are: https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk
But the symptoms are not quite identical to Potplayer. The madVR OSD doesn't appear. I get a windows alert sound after a few secs, and sometimes MPC gets unhung and the playback starts. Also hanging seems to occur differently, when I clobber one file instead of when I change between files.
Thanks, will have a look at this later.

Where is the v0.92.7 changelog?
Haha, I didn't expect people to notice the website change so quickly. I was in the process of uploading v0.92.7, so I already updated the website, but then had to change something again. Upload coming "soon". The real version announcement is always in this thread. If it's not announced here, it's not up yet.

A sample of one distorted image might not be fair, but it shows my point where i find the end result of the new AS
lacking in comparison and i find no shorige of bad content.
I'm using the awesome reducing random noise and compression artefacts when needed, but in this situation i have to
use them, and this solution brings the questions of subtle harm and more GPU freedom being lost (then just having the previus AS checked).
If other options come to light more joy to me, i did caught on that the new AS is stronger,
so if i tone it down a bit and if the grain is not as visible i guess i'll survive the change.
Just give the next build a try and let me know what you think. I'm very willing to listen to feedback. Unfortunately not many people have commented on the new AdaptiveSharpen yet.

It looks like madshi actually controls that website now?!
Yes, have been for quite a while already. The original owner was very nice when I contacted him. He transferred the domain to me without any hassle whatsoever.

You know what? I'm gonna say a simple thing.

I hope everybody here realizes how lucky we have to have you.
Thanks! :)

I mean what setup is he using to reach the conclusion that 8 bit is better than 10 bit for him. By keeping madvr at 8 bit hes defeating the whole purpose of doing this test? His tv will never get the 10 bit signal?
Well, I suppose he should ideally answer himself.

mparade

15th October 2017, 19:10

@madshi

10bit output is only possible with D3D11 presentation, and in older OSs only in FSE mode.

I am on D3D11 path (set in madVR) and on Win10 64 bit but in JRiver the crtl+J
says I am in window mode. In window mode 10 bit is also possible?

Yes, there should be a working "3D" profile condition supported.

Excellent. :)

madshi

15th October 2017, 19:13

madVR v0.92.7 released

http://madshi.net/madVR.zip

* added experimental deblock option "don't reduce texture detail"
* added experimental AdaptiveSharpen option "linear light"
* added new lowest "reduce compression artifacts" strength
* revisited "reduce random noise" strength levels, now from 1-12
* updated AdaptiveSharpen to latest igv version
1) I'd like to get rid of the new "linear light" option for AdaptiveSharpen, so please everybody who uses/likes AdaptiveSharpen try the "linear light" option and let me know if you prefer it checked or unchecked.

2) Would be extra nice if you could also compare the new AdaptiveSharpen to the old one (v0.92.4 and older).

3) "reduce random noise" has new strength levels from 1-12. The very lowest setting doesn't seem to do much, if anything at all. Does anybody find it useful? If nobody says he wants to keep it, I'll drop it in the next build. I think strength level 2 will probably be the best starting point?

4) I've added a new lowest "reduce compression artifacts" strength, the other strength settings got moved up one. But to be honest, the new lowest strength 1 doesn't seem to do much, so I'm not sure if it's really useful. What do you think?

5) There's a new option called "don't reduce texture detail", which activates a slightly modified "reduce compression artifacts" algorithm. The original algo included some random noise reduction over the whole video frame. The modified algo tries to do noise reduction only near edges. The difference is not overly dramatic, though. Which option do you like more? I think I'm going to switch to the "don't reduce texture detail" algo variant and remove the option in the next build, but wanted to give you the chance to compare first.

PurpleMan

15th October 2017, 19:27

First of all, are we talking about bitmap type subtitles (VOBSUB, SUP, Blu-Ray m2ts included subs etc)? Or are we talking about text type subtitles (SRT, ASS, etc)? Can you please also test "the other type" to see if those are also affected?

Finally, a small video sample which shows all these problems you mentioned would be helpful. Or if it's hard to find one sample which reproduces all these problems, then I'm also fine with multiple smaller video samples, one for each problem.

These are SRT text subtitles, not bitmap/vobsub. I don't have a bitmap/vobsub one handy so couldn't test to reproduce.

I think I managed to depict this using screenshots and adding position markers. Here are 4 different subtitles, with the 'move to bottom' feature and without the 'move to bottom' feature:

Subtitle 01 || with move-to-bottom (https://ibb.co/jhoLrR) || without move-to-bottom (https://ibb.co/i2gvQm)
Subtitle 02 || with move-to-bottom (https://ibb.co/n43aQm) || without move-to-bottom (https://ibb.co/gyJ9km)
Subtitle 03 || with move-to-bottom (https://ibb.co/eDfBy6) || without move-to-bottom (https://ibb.co/j7e4J6)
Subtitle 04 || with move-to-bottom (https://ibb.co/etbUkm) || without move-to-bottom (https://ibb.co/mDCfrR)

You can see that there are 6 different horizontal line positions in the "with move-to-bottom" screenshots but only 2 fixed positions (one for each line) in the "without move-to-bottom" screenshots.

Do you see what I mean?

madshi

15th October 2017, 19:31

I tried... Here they are: https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk
But the symptoms are not quite identical to Potplayer. The madVR OSD doesn't appear. I get a windows alert sound after a few secs, and sometimes MPC gets unhung and the playback starts. Also hanging seems to occur differently, when I clobber one file instead of when I change between files.
The logs say v0.91.10 instead of v0.92.6/7, though. And unfortunately the PDB debug symbols don't seem to work at all... :( Are they in the right folder? Meaning the PDB symbols for MPC-HC are in the same folder and have the same file name (except for the extension, obviously) as the MPC-HC main exe file?

I am on D3D11 path (set in madVR) and on Win10 64 bit but in JRiver the crtl+J says I am in window mode. In window mode 10 bit is also possible?
Depends on the GPU. With an Nvidia or AMD GPU it should be possible, but it must be fullscreen windowed mode, not just windowed mode. FSE mode would be best, of course.

These are SRT text subtitles, not bitmap/vobsub. I don't have a bitmap/vobsub one handy so couldn't test to reproduce.

I think I managed to depict this using screenshots and adding position markers. Here are 4 different subtitles, with the 'move to bottom' feature and without the 'move to bottom' feature:

Subtitle 01 || with move-to-bottom (https://ibb.co/jhoLrR) || without move-to-bottom (https://ibb.co/i2gvQm)
Subtitle 02 || with move-to-bottom (https://ibb.co/n43aQm) || without move-to-bottom (https://ibb.co/gyJ9km)
Subtitle 03 || with move-to-bottom (https://ibb.co/eDfBy6) || without move-to-bottom (https://ibb.co/j7e4J6)
Subtitle 04 || with move-to-bottom (https://ibb.co/etbUkm) || without move-to-bottom (https://ibb.co/mDCfrR)

You can see that there are 6 different horizontal line positions in the "with move-to-bottom" screenshots but only 2 fixed positions (one for each line) in the "without move-to-bottom" screenshots.

Do you see what I mean?
Yes, I see what you mean. Good screenshots. Now can you provide me with a small sample with which I could easily reproduce this? :) Thanks!

j82k

15th October 2017, 19:38

I mean what setup is he using to reach the conclusion that 8 bit is better than 10 bit for him. By keeping madvr at 8 bit hes defeating the whole purpose of doing this test? His tv will never get the 10 bit signal?

I basically compared:

- Nvidia 10/12 bit, madVR 10 bit

to

- Nvidia full RGB 8 bit, madVR 8 bit

MadVR was correctly showing "D3D11 fullscreen windowed (10bit)" while testing 10/12 bit.

I think I tried all available 10/12bit modes. (RGB, 4:2:0, 4:2:2, 4:4:4) and RGB 8 bit still had the best distinction between the different grayscale steps. With 10/12 bit there were always some steps bunched together as if they were just one color.

Also I just played this 10-bit pattern through usb using the TV's internal mediaplayer and it looked worse than madVR's dithered 8-bit. I have to say this isn't surprising, as LG Oled TVs are known to have color banding issues.

PurpleMan

15th October 2017, 19:48

Yes, I see what you mean. Good screenshots. Now can you provide me with a small sample with which I could easily reproduce this? :) Thanks!

Yay! Glad this helps!
Here's a sample SRT file with the same 4 lines I used above:

test.srt (https://drive.google.com/file/d/0B9SJI4J_AlQpbHlvWlpIYkh0N0E/view?usp=sharing)

You can change the timestamps in it to fit your needs.

Thank you!

Anima123

15th October 2017, 19:49

Wow, some high-compressed videos really need as strong deblock as like 8, to get a nice result.

Didn't notice any difference between "don't reduce texture detail" on and off with real-life content.

Edit: Also I found that the NGU-AA is not as good as NGU Soft with the deblock algorithm, NGU Soft 4x doesn't do any better than NGU Soft 2x.

Edit2: With stronger as far as 9, the "don't reduce texture detail" did result in better quality. It's okay for me to become the default algorithm.

mparade

15th October 2017, 19:57

@madshi

I don't know. Show me a screenshot of the OSD (Ctrl+J), and I might be able to say more.

here is the screenshot:

http://s1.toldacuccot.hu/letoltes?sid=095ee2a033cc9c25ea5bc9c2d432d12a&file=madVR_screenshot1.png

thank you!

thewildsun

15th October 2017, 20:20

In the Nvidia Control panel, under Manage 3D settings, do you have Power Management Mode set to "Optimal Power"? If so, set it to "Adaptive" instead.

It's set to max performance. I tried changing it to adaptive anyway, just in case, but still no luck.

Neo-XP

15th October 2017, 21:12

madVR v0.92.7 released

1) Checked, because without linear light it produces/accentuates too much artifacts near the edges.

2) Not tested.

3) Is it working ? levels 1-2 do nothing (?) and 3 way too much. Not a setting for me I think.

4) It is already doing a lot at strength 1 without the "don't reduce texture details" checked.

5) With the "don't reduce texture details" option checked, there are a lot of very ugly compression artifacts left in the image that are not processed at all. Why not keep this option ?

Just a test I did (Source / RCA@2 with the option enabled / RCA@2 with the option disabled) :

https://s1.postimg.org/4y2fktfdez/Source.png (https://postimg.org/image/4y2fktfdez/) https://s1.postimg.org/5uofsm34zf/With_don_t_reduce_texture_details.png (https://postimg.org/image/5uofsm34zf/) https://s1.postimg.org/3ailfz2ynf/Without_don_t_reduce_texture_details.png (https://postimg.org/image/3ailfz2ynf/)

And zoomed :

https://s1.postimg.org/47imzr7sa3/Source.png (https://postimg.org/image/47imzr7sa3/) https://s1.postimg.org/4eym9cz5or/With_don_t_reduce_texture_details.png (https://postimg.org/image/4eym9cz5or/) https://s1.postimg.org/9qdiu2rnu3/Without_don_t_reduce_texture_details.png (https://postimg.org/image/9qdiu2rnu3/)

It is quite obvious if you switch from one image to another. For instance on the chin of the guy in the left side, on the books behind him and on his shirt.

Thank you for letting us test this.

mzso

15th October 2017, 21:45

The logs say v0.91.10 instead of v0.92.6/7, though. And unfortunately the PDB debug symbols don't seem to work at all... :( Are they in the right folder? Meaning the PDB symbols for MPC-HC are in the same folder and have the same file name (except for the extension, obviously) as the MPC-HC main exe file?

Oh, yeah... I forgot about that I downgraded because of the error at bookmarking. (But it shouldn't make much of a difference since the new version hung just the same)

The PDB-s I got from the mpc-hc download page, which had an identical folder structure to the main program, and included LAV PDBs. So I copied all to the program folder. (I just checked and the names are right.) So I don't have any idea why wouldn't it work. (Whatever it's supposed to be doing.)

So what now?

Edit:
Anyway, I made some new ones (#8-11) (https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk) with lates madVR, maybe (if unlikely) it helps.

ABDO

15th October 2017, 21:49

haha, i didn't expect people to notice the website change so quickly.

we do, I visit posts py madshi tens times Per day :D to find any news from you, because you the best one in the world for video fans like me.
form the bottom of my heart thank you madshi for this stunning work :)

madshi

15th October 2017, 22:25

Here's a sample SRT file with the same 4 lines I used above:

test.srt (https://drive.google.com/file/d/0B9SJI4J_AlQpbHlvWlpIYkh0N0E/view?usp=sharing)

You can change the timestamps in it to fit your needs.
Thanks.

Edit2: With stronger as far as 9, the "don't reduce texture detail" did result in better quality. It's okay for me to become the default algorithm.
Good to know.

here is the screenshot:

http://s1.toldacuccot.hu/letoltes?sid=095ee2a033cc9c25ea5bc9c2d432d12a&file=madVR_screenshot1.png
Looks weird. It seems to me that your media player has sized the madVR window much too big. Your screenshot is 1920x1080, but according to the OSD, the madVR rendering window is sized to something bigger than 4K.

Which media is that? Have you tried another one, e.g. MPC-HC/BE?

It's set to max performance. I tried changing it to adaptive anyway, just in case, but still no luck.
Other possible causes could be 3rd party software accessing the GPU, like GPU-Z, f.lux or similar? If that doesn't help, make a screenshot of the OSD (Ctrl+J) in the moment when frame drops occur. Don't attach is to this forum, just link to it here.

1) Checked, because without linear light it produces/accentuates too much artifacts near the edges.
Interesting. Do you have a good screenshot for that?

3) Is it working ? levels 1-2 do nothing (?) and 3 way too much. Not a setting for me I think.
The "reduce random noise" strength setting defines the threshold below which noise can be cleaned, and above which it's considered detail and kept. So depending on your image, a certain strength setting may do nothing, if the film grain is higher than the threshold, while increasing the strength settings by 1 might suddenly result in total smoothing. This behaviour may feel somewhat weird. An alternative approach would be to offer 2 separate controls: 1) noise vs detail threshold (current strength setting) and 2) reduction level of noise (currently always 100%). But then, this would be hard to understand. So I'll probably keep it as it is.

4) It is already doing a lot at strength 1 without the "don't reduce texture details" checked.
So you consider strength 1 useful?

5) With the "don't reduce texture details" option checked, there are a lot of very ugly compression artifacts left in the image that are not processed at all. Why not keep this option ?

Just a test I did (Source / RCA@2 with the option enabled / RCA@2 with the option disabled)
Hmmmm... I can't manage to reproduce your test results. The ugly artifacts on the chin are pretty much the same with RCA@2 for me, regardless of whether the option is enabled or disabled. I think your good looking screenshot must have had either a higher RCA setting, or a combination of RCA + RRN? Can you double check?

Oh, yeah... I forgot about that I downgraded because of the error at bookmarking. (But it shouldn't make much of a difference since the new version hung just the same)

The PDB-s I got from the mpc-hc download page, which had an identical folder structure to the main program, and included LAV PDBs. So I copied all to the program folder. (I just checked and the names are right.) So I don't have any idea why wouldn't it work. (Whatever it's supposed to be doing.)
Could it be possible that you download the 32bit PDB files instead of the 64bit ones? Just a wild guess. Maybe you could try 32bit MPC?

Anyway, I made some new ones (#8-11) (https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk) with lates madVR, maybe (if unlikely) it helps.
Same as before, unfortunately.

we do, I visit posts py madshi tens times Per day :D to find any news from you, because you the best one in the world for video fans like me.
form the bottom of my heart thank you madshi for this stunning work :)
:D

thewildsun

15th October 2017, 22:53

Other possible causes could be 3rd party software accessing the GPU, like GPU-Z, f.lux or similar? If that doesn't help, make a screenshot of the OSD (Ctrl+J) in the moment when frame drops occur. Don't attach is to this forum, just link to it here.

Thanks. I'm not using any 3rd party software like f.lux.

Here's the OSD screenshot: https://i.imgur.com/s2tMJWk.png

Edit: It might be a RAM-related issue. I noticed that RAM only ever increases while playing a video. That explains why restarting the video player fixes the problem temporarily. I'm not sure what to try next, though...

Edit2: Resetting madVR to default settings did seem to alleviate the issue (I was able to play a 20 minute video with no frame drops) but if the issue is RAM-related, then that may just be because the default settings are less RAM intensive.

Edit3: For chroma upscaling, I had been using Recon-soft with SuperRes 3. Turning off SuperRes seems to fix the issue -- I need to do more testing to see if this is a true fix or not.

ABDO

15th October 2017, 23:10

madVR v0.92.7 released

1) unchecked,with linear light it reduce Sharpen effect.

2) New AdaptiveSharpen better than the old one, the old did ahuge sharpen edges with visible artifacts, so i never used it, but the new one do enhance detail very nice and sharpen it with no visible ringing in the same time.

3) Not tested.

4) I agree strength 1 doesn't do much.

5) as @Neo-XP notice, With the "don't reduce texture details" option checked, there are some compression artifacts left in the image that are not processed.

Source
https://s20.postimg.org/z42bfg7c9/Source.png (https://postimg.org/image/z42bfg7c9/)

RCA 10 with the option disabled
https://s20.postimg.org/njswa0og9/RCA_10_with_the_option_enabled.png (https://postimg.org/image/njswa0og9/)

RCA 10 with the option enabled
https://s20.postimg.org/4u0wt5do9/RCA_10_with_the_option_enabled.png (https://postimg.org/image/4u0wt5do9/)

edit:
it seem (don't reduce texture detail ) option save some texture detail in normal life content, and it give good result to the whole picture.

look to the pullover line detail

Source
https://s20.postimg.org/ro31tqrdl/Source.png (https://postimg.org/image/ro31tqrdl/)

RCA 7 with the option disabled
https://s20.postimg.org/jtxuutrop/RCA_7_with_the_option_disabled.png (https://postimg.org/image/jtxuutrop/)

RCA 7 with the option enabled
https://s20.postimg.org/hpdhts82h/RCA_7_with_the_option_enabled.png (https://postimg.org/image/hpdhts82h/)

@madshi could you please keep this option for us, seem good for 3d content, and not to 2d cartoon.

mparade

15th October 2017, 23:38

@madshi

Looks weird. It seems to me that your media player has sized the madVR window much too big. Your screenshot is 1920x1080, but according to the OSD, the madVR rendering window is sized to something bigger than 4K.

Which media is that? Have you tried another one, e.g. MPC-HC/BE?
JRiver latest 32-bit version. I haven't tried any other player yet (I have paid for it, so there shouldn't be any bug). :)

I have problems with "3D" profile rule as well.
Could you please have a look at this:

if (3D) "3D"

else if (srcWidth > 1920) and (not 3D)"2160p"
else if (srcWidth <= 1920) and (srcHeight > 1080) and (not 3D) "2160p"

else if (srcWidth > 1280) and (srcWidth <= 1920) and (not 3D) "1080p"
else if (srcWidth <= 1280) and ((srcHeight > 720) and (srcHeight <= 1080)) and (not 3D) "1080p"

else if (srcWidth > 960) and (srcWidth <= 1280) and (not 3D) "720p"
else if (srcWidth <= 960) and ((srcHeight > 540) and (srcHeight <= 720)) and (not 3D) "720p"

else if (srcWidth <= 960) and (srcHeight <= 540) and (not 3D) "SD"

When playing 3D ISO files madVR is using my "1080p" profile instead of the "3D" one.

I would really appreciate your help.