madVR - high quality video renderer (GPU assisted)

[chros]

overlay mode is error resilient and even in window mode and can make the best use of GPU gamma ramps.

there is nothing wrong with FSE in general. if you have 2 displays you will notice some issue with FSE in general and FSE only works in fullscreen unlike overlay.
Thanks again.
I wanted to try it out on my setup (see my signature) but madvr says "This GPU/Driver doesn't support overlay...".
Can anybody use overlay mode on Win8.x and with nvdia gpu? If yes, with which gpu and driver version?
Thanks

[kalston]

Yep, overlay is amazing. Ever since I found out about it I haven't looked back.

[XMonarchY]

I wonder if the next version of madVR will include updated algorithms for Chroma Upscaling SuperRes Filter.

[xabregas]

so why i'm lying? why do all people need FSE or are lying?

overlay mode is error resilient and even in window mode and can make the best use of GPU gamma ramps.

there is nothing wrong with FSE in general. if you have 2 displays you will notice some issue with FSE in general and FSE only works in fullscreen unlike overlay.

I know that with 2 displays FSE mode is very problematic but u can correct it by changing some of your gpu default settings. Disabling some stuff makes wonders. Overlay used to be the best way to render videos since windows 7 appeared. With Windows 7 came the crap aeros and the intensive use of vertycal sync in triplicate or more. So FSE mode is the only Vsync i trust, because i dont have my own gpu Vsync setting enabled, that way i know there is only one VSync being used, with overlay if i disable Vsync in my gpu settings i wont have vsync in Full Screen and if i enable it i may be using 2x Vsync ( OS + madvr overlay ). So for me FSE mode is here to stay. ;)

[huhn]

if you really think it works this way...

[alx000]

I know that with 2 displays FSE mode is very problematic but u can correct it by changing some of your gpu default settings. Disabling some stuff makes wonders. Overlay used to be the best way to render videos since windows 7 appeared. With Windows 7 came the crap aeros and the intensive use of vertycal sync in triplicate or more. So FSE mode is the only Vsync i trust, because i dont have my own gpu Vsync setting enabled, that way i know there is only one VSync being used, with overlay if i disable Vsync in my gpu settings i wont have vsync in Full Screen and if i enable it i may be using 2x Vsync ( OS + madvr overlay ). So for me FSE mode is here to stay. ;)

What settings can I change to improve FSE mode?
I have a r9 280

[huhn]

just to make it clear it is highly recommended to leave all GPU settings at default for madVR.

[xabregas]

What settings can I change to improve FSE mode?
I have a r9 280

try put everything in "let the application decide" in your gpu 3d settings and disable all video enhancements. That should do. On Nvidia i found 2 settings that was screwing up madvr video playback. I put it like this:

-Multi-Display/mixed Gpu acceleration = single display performance mode

-Power management mode = prefer maximum performance


This makes the gpu preform without saving power crap auto settings.

As for Vsync u shoud set it as "let application decide".

Now answering @huh,

If u use madvr overlay windowed in full screen and u let your gpu settings by default i dont even want to know wtv crap u are running without even knowing. ;)

[huhn]

nothing madVR avoids them by default

[nussman]

No problems here with default GPU Settings and with/without FSE.
What xabregas says is not true for sure.

[chros]

Thanks again.
I wanted to try it out on my setup (see my signature) but madvr says "This GPU/Driver doesn't support overlay...".
Can anybody use overlay mode on Win8.x and with nvdia gpu? If yes, with which gpu and driver version?
Thanks
I just realized that maybe it's the intel gpu/driver that can be blamed here? (It's so confusing, at least for me :) )
So, let me add this to the previous question:
Can anybody use overlay mode on Win8.x and with intel HD4xxx? If yes, with which gpu and driver version?
:) Thanks

[huhn]

it works on my hd 4400 and it should work on intel in general. i'm running 10.18.15.4256.

and about driver settings. there is nothing work with disabling these terrible post processing effect even through they shouldn't effect madVR but you really shouldn't change the default 3D application settings.

if i'm not mistaken nvidia has disabled all post processing things by default and AMD has only 2 active in newer driver. and I'm pretty sure I have disabled these.

[JarrettH]

Maybe madshi can clear up windowed overlay versus full screen exclusive for us.

What are the differences? Switching with overlay is FAST and you can also take screenshots. Is the quality the same? Video untouched?

[huhn]

both have the same quality that's for sure.

FSE takes full control of the presentation so it's unlikely any program in the back ground can effect it making it more solid than normal full screen window mode.

if you don't have any problems than there is no reason to switch to any other presentation mode.

same is for present frames in advance or backbuffer. present frames in advance less chance to be interrupted that doesn't you need it at all.

a user had a problem with FSE mode and this problem is gone without it and he isn't losing anything in term of quality so it's totally fine to disable it.

to my experience overlay is the most stable and reliable mode for nvidia user.

[Manni]

One last quick update.

I've done more tests with 14.12 / 88.21 / FSE 10bits, no dropped frames and no MPC-BE crashes.

So I tried to upgrade 14.12 to 15.7.1, and I immediately got MPC-BE crashes.

I disabled FSE and D3D11, enabled D3D9, and apparently the crashes are gone and so are the dropped frames (although I haven't been able to test a full movie).

I'm going to do more tests with 15.7.1 / 88.21 / D3D9 over the week-end, but if that works I'll settle for that as it means I can use the latest AMD drivers with the latest MadVR in non FSE mode.

Madshi, I know you're busy on other things at the moment but if you'd like me to do more tests or provide more info regarding what causes MPC-BE crashes and dropped frames with D3D11 in my setup, please let me know.

[Akeno]

thx for you info.

What settings in sharpness and superRes would you suggest?

Whatever settings you're comfortable with, really. The differences in SuperRes settings is minimal and the Super-xbr settings depends on your preference for sharpness. I suggest staying away from anything lower than 75 though as the image starts to get really soft compared to other algorithms. I personally use 100. Sharp, but not overly so. For SuperRes, I use a strength of 1 to keep down the temperatures on my laptop and a radius of 0.75. The radius settings denote aliasing vs other sharpening artifacts so play around with that. madshi has stated that he's going to raise the default radius values in the next update so 0.75 might be a good choice over the default 0.66.

[baii]

With my Amd, overlay mode on will show a osd error in d3d9 but nothing happens on d3d11. D3d11 fse switch is pretty quick for me on win8.1 or 10. Dual display fse dont really have problem for me unless I try to have 2 exclusive instance (say a game). Ofc. With fse, it is hard to change media player option on the fly.

Sent from my 306SH

[StinDaWg]

I'm having some issues with D3D11 and FSE mode. With "Use a separate device for presentation" checked, when I exit out of fullscreen into windowed mode the sound keeps playing but the video is black. If I uncheck this then no issues.

However, when going out and then back into fullscreen, audio sync gets off and the render/present queues only fill up half way. Sometimes the video will just play in slow motion until I pause it and hit play again, then it works.

Switching from fullscreen to windowed is also extremely slow compared to D3D9.

This is on Windows 10 with the latest AMD drivers, but the problem was the same on 8.1.

[Sunset1982]

I'm thinking of switching back to win 7 with catalyst 14.12 because I had unreported glitches with D3D9 and FSE with Win10/Cat. 15.7.1 yesterday...

[Manni]

I'm thinking of switching back to win 7 with catalyst 14.12 because I had unreported glitches with D3D9 and FSE with Win10/Cat. 15.7.1 yesterday...

Got an MPC-BE crash first time this morning with 15.7.1 and MadVR 88.21 in non-FSE (D3D9) so going back to 14.12 too.

That seems to be the most stable platform for now in my setup, allowing me to use the latest build.

It works fine with FSE and D3D11 (10bits) as I've tested it for a while, I'm going to try it without FSE in D3D9, which I haven't tested yet.

Looks like the issue is more profound than just D3D11.

I'll keep the thread updated.

[Sunset1982]

is fullcreen exclusive mode not working with 2160p resolution?

[aufkrawall]

I can confirm that it doesn't work here (at least with Nvidia DSR 5k).
MPC HC UI elements are hidden, but madVR seems to "forget" to enter FSE.

[nevcairiel]

I can confirm that it doesn't work here (at least with Nvidia DSR 5k).
MPC HC UI elements are hidden, but madVR seems to "forget" to enter FSE.

It only doesn't work properly when there is any hidden scaling going on, possibly like with DSR, or the player you are using is using Windows DPI scaling (which would be bad, as Windows would also upscale the video image then).

With MPC-HC, you may need a newer nightly if you run a multi-monitor setup with only one being 2160p, as it only recently fixed a problem with such a setup.

[Manni]

Last update, watched a full movie with 14.12 / 88.21 / D3D9, no MPC-BE crashes and no dropped frames (at least reported).

So that's the winning combo for me, as it takes me too long to use the player interface if I use FSE, which also works fine here, at least with D3D11 and 10bits which is what I tested formerly.

My problem is solved, available to help Madshi with more tests if he wants to get to the bottom of this at some point.

[Sunset1982]

I'm doing some testings with Windows 10/Catalyst 15.7.1 and 4k Display right now.

I got two problems right now:

- FSE isnt working if I set my Desktop to 4K and use MadVR to upscale 1080p to 4K

- Windowed overlay Mode isn't working, Madvr tells me, driver dont support this mode.

[Sunset1982]

Last update, watched a full movie with 14.12 / 88.21 / D3D9, no MPC-BE crashes and no dropped frames (at least reported).

So that's the winning combo for me, as it takes me too long to use the player interface if I use FSE, which also works fine here, at least with D3D11 and 10bits which is what I tested formerly.

My problem is solved, available to help Madshi with more tests if he wants to get to the bottom of this at some point.


Good to know.


Does anyone know if there is a working amd driver for windows 10 other than 15.7 and 15.7.1?

[Akeno]

What are the requirements for running overlay mode? I have an NVIDIA GTX 860M running the latest drivers on windows 8.1 but madVR reports that these don't support overlay mode.

[madshi]

I'll reply to all your posts later (probably next weekend). For now I just wanted to share a test build for all those of you who are eager to play with new features:

http://madshi.net/madVRblackBarTest1.rar

This test build doesn't really do anything "useful" yet. But it does try to automatically detect black bars encoded into the movie. If detection succeeds, the detected "active video area" is marked with a thin green rectangle.

For those of you who want to play with this, just test some movies with different aspect ratios. Maybe even some weird stuff like IMAX encodings where aspect ratios change in the middle of the movie. This test build is supposed to handle all that just fine.

Detection of black bars might not always be super fast. I've put more weight on avoiding false detections. So if you find a situation where a detected black bar is incorrect, please provide a small video sample for me. But I think it should not occur. If you find samples where you think black bars *should* be properly detected but are not, I'd like a sample for that, as well.

In unusual situations (e.g. non-symmetrical black bars) madVR might decide to wait multiple seconds before accepting/confirming a detected black bar. So have a bit of patience in such situations.

[Xaurus]

I can also confirm that there is some strange frame drops / presentation glitches going on that does not show up in the OSD. Win 10, madvr 88.21, latest lav. I tried using the DX11 presentation path to see whether it would make a difference, but without any success really.

[Manni]

I'll reply to all your posts later (probably next weekend). For now I just wanted to share a test build for all those of you who are eager to play with new features:

http://madshi.net/madVRblackBarTest1.rar

This test build doesn't really do anything "useful" yet. But it does try to automatically detect black bars encoded into the movie. If detection succeeds, the detected "active video area" is marked with a thin green rectangle.

For those of you who want to play with this, just test some movies with different aspect ratios. Maybe even some weird stuff like IMAX encodings where aspect ratios change in the middle of the movie. This test build is supposed to handle all that just fine.

Detection of black bars might not always be super fast. I've put more weight on avoiding false detections. So if you find a situation where a detected black bar is incorrect, please provide a small video sample for me. But I think it should not occur. If you find samples where you think black bars *should* be properly detected but are not, I'd like a sample for that, as well.

In unusual situations (e.g. non-symmetrical black bars) madVR might decide to wait multiple seconds before accepting/confirming a detected black bar. So have a bit of patience in such situations.

This sounds great, but if I understand correctly does this mean that at the moment, you're only showing us what you're detecting, so that in a later stage we can tell MadVR what we want to do for each A/R?

For example, I have a 16/9 screen and I use a single mask in the lower part of the screen to mask an area equal to both black bars once I've shifted the picture up so that the top of the picture will always be at the top of the screen, leaving one area to be masked at the bottom. Will you be able to handle that?

I'm using the Zoom Memory of the X500 at the moment, and it works great, but it's limited to 5 memories and also if the database in MyMovies is wrong, it selects the wrong A/R, while your method would be more exhaustive and error-proof if accurate.

And what will you do with mixed A/R like The Dark Knight? I'm not sure this can be handled unless you start implementing non linear stretch like on the radiance to stretch the 1.78 parts horizontally for those who have a 2.35 screen, but if it's not instant in won't be usable really.

[madshi]

I have a CIH setup myself, so don't worry, I'll put the detection to good use in the near future. For now I'd just like some feedback about whether the detection works reliably.

[chros]

it works on my hd 4400 and it should work on intel in general. i'm running 10.18.15.4256.

What are the requirements for running overlay mode? I have an NVIDIA GTX 860M running the latest drivers on windows 8.1 but madVR reports that these don't support overlay mode.
I have just upgraded both intel and nvidia drivers:
- intel : 10.18.10.4176 -> 10.18.10.4226 (there's no newer driver for HD4000 not like for yours :) )
- nvidia: 347.88 -> 353.62

Results by selecting Overlay mode:
- intel: black picture (no report of "doesn't support"), no osd, but there's sound (with the previous driver there wasn't any sound) -> so I consider it as a bogus driver version
- nvidia: still no overlay support


it does try to automatically detect black bars encoded into the movie.
Woow!!!

[naomatrix]

Currently MadVR does not work with Bluesky Frame Rate Converter (http://bluesky23.yukishigure.com/BlueskyFRC.html), or not ?
I could not run their joint work in Windows 8.1, green screen always got, but running with EVR. As I understand that, MadVR does not accept RGB32 input, but can not switch to another stream output format in BFRC.
Bluesky Frame Rate Converter lets you watch 60 fps smooth video using the AMD Fluid Motion on AMD graphics cards in the player MPC-HC,BE,Pot.

[Anima123]

I didn't get the black-bar detection, will it include videos actually 16:9 encoded as 4:3, by adding black bars up and down the video content area?

[DarkSpace]

This test build doesn't really do anything "useful" yet. But it does try to automatically detect black bars encoded into the movie. If detection succeeds, the detected "active video area" is marked with a thin green rectangle.

Nice idea, but one thing to consider: How will you handle partially overscanned video (some scenes have black bars, some scenes use the full image)? I should guess that you can't reliably detect such a situation, and even then, the treatment of such things may vary depending on the content.
For example, the content could be a compilation of unrelated videos, some of which have an aspect ratio different from 4:3 / 16:9. In that case, you'd probably want to crop each "segment" individually.
Then again, there could be a case (I've only seen this once, and I don't remember where right now) where the black bars appear to be completely useless, until a color inversion happens in the video - and they turn white! I think this would be one of those misdetections you mentioned, because this example's black bars actually hold information.

Just food for thought, rather now than later when you need to rewrite half of your logic...

[Manni]

I have a CIH setup myself, so don't worry, I'll put the detection to good use in the near future. For now I'd just like some feedback about whether the detection works reliably.

Great, I'll take a look and I'll report if it works with my usual test movies for various A/R.

By the way, I have a CIW, not a CIH setup (my screen is 16/9, so it's the height that varies, not the width). Hopefully you'll take this into account too, as it's a fairly common situation.

[iSunrise]

@madshi:
Black bar detection fails for me with these two encodes:

http://www.recordedamigagames.org/recordedamigagames/longplays_1/Amiga%20Longplay%20%5B400%5D%20Turrican%20II%20%28a%29.mp4
http://www.recordedamigagames.org/recordedamigagames/longplays_1/Amiga%20Longplay%20%5B401%5D%20Turrican%203%20%28a2%29.mp4

Not sure what's special about them, but most of the other files I did a quick test on worked fine.

PS: Seems deep linking is not allowed on that site, just paste the URL manually.

[Manni]

I've tested most of my usual movies for various A/R, here are the results:
1.33: pass
1.66: pass
1.85: pass
2.35: mixed. Some detected, others like The Notebook fail.
2:20: fail (Lawrence of Arabia).
2.40: pass

I tried The Dark Knight and it can take up to 6 seconds to detect the A/R change between the iMax scenes in 1.78 and the 2.35 scenes, so it's unlikely to be usable in a CIH setup.

Let me know if you need a short clip from the failed movies if you don't have them, I'll try to make a short clip from my blurays.

[huhn]

I have just upgraded both intel and nvidia drivers:
- intel : 10.18.10.4176 -> 10.18.10.4226 (there's no newer driver for HD4000 not like for yours :) )
- nvidia: 347.88 -> 353.62

Results by selecting Overlay mode:
- intel: black picture (no report of "doesn't support"), no osd, but there's sound (with the previous driver there wasn't any sound) -> so I consider it as a bogus driver version
- nvidia: still no overlay support



Woow!!!

sounds like yet another optimus issue.

[heiseikiseki]

sounds like yet another optimus issue.

Madvr Works Perfectly On OPTIMUS system here.(GTX860M+HD4400,both newest driver on windows 10)
On my computer the output port is on Intel Graphic but seems the high present time issue fixed when I　upgrade to windows 10.(Need to set the
presented video frames to "2"(Present Time :0.89ms on FHD).Higher would cause the high present time(10ms on FHD), Lower would cause drop frames.)

[huhn]

doesn't sound perfect at all.

[madshi]

Nice idea, but one thing to consider: [...]
At this point I'm not looking for a discussion. I'd just like feedback on whether the detection as implemented in the test build works or not.

Black bar detection fails for me with these two encodes: [...]

Not sure what's special about them, but most of the other files I did a quick test on worked fine.
What's special about them is that they have static parts (score bar) which have a couple of non-bright pixels in them at the borders. Furthermore the black bars are not symmetrical (bottom bar is bigger than top bar). The combination of these attributes makes madVR unsure about the detection. The detection really works best with typical video content.

2.35: mixed. Some detected, others like The Notebook fail.
2:20: fail (Lawrence of Arabia).
Samples for those fails would be helpful. Did they fail completely? Or did it just take a long time to detect? 2.20:1 is currently not in my list of "known" aspect ratios. Which means that madVR may wait an extra 10 seconds (or even an extra 30 seconds) on top of the usual time before detecting the black bars properly. But at some point detection should work. How long did you wait?

I tried The Dark Knight and it can take up to 6 seconds to detect the A/R change between the iMax scenes in 1.78 and the 2.35 scenes
The change from 2.35 to 1.78 should be detected immediately. The change from 1.78 to 2.35 may take some time to detect, but I don't think that's a big problem, because you don't lose image content, you just see black bars for a few seconds. Anyway, maybe I can improve the 1.78 to 2.35 change detection speed.

[huhn]

found a clear misdetection:

http://abload.de/img/misdetectionktbdp.png

[Manni]

Samples for those fails would be helpful. Did they fail completely? Or did it just take a long time to detect? 2.20:1 is currently not in my list of "known" aspect ratios. Which means that madVR may wait an extra 10 seconds (or even an extra 30 seconds) on top of the usual time before detecting the black bars properly. But at some point detection should work. How long did you wait?


The change from 2.35 to 1.78 should be detected immediately. The change from 1.78 to 2.35 may take some time to detect, but I don't think that's a big problem, because you don't lose image content, you just see black bars for a few seconds. Anyway, maybe I can improve the 1.78 to 2.35 change detection speed.

I left The Notebook run longer and it detected the A/R after about 50 seconds. It might be due to the fact that the opening shot is a sunset, so might be harder to detect the edge of the black bars with certitude. We should probably leave that one.

I left Lawrence run for 1mn before giving up (past the audio only intro of course!). I can try to make a clip, or maybe you could add 2.20 in your list to see if it helps detection, if that's the reason, and I'll test again.

Re The Dark Knight, it's not really a problem for me because on a CIW setup you just have to live with the black bars, but in a CIH setup if the detection in both directions is not instant and if you can't do whatever you want to do to get rid of the bar instantly, it's going to be weird because you'll have 1.78 then 2.35 with black bars then 2.35 with NLS, should be very distracting. Anyway not my problem as I don't have CIH, so I'll let users comment on that when it's implemented.

Very promising overall, can't wait to see more on that front.

[madshi]

found a clear misdetection:

http://abload.de/img/misdetectionktbdp.png
Ouch. Can I have a small video sample with which I can reproduce this problem, please?

I left The Notebook run longer and it detected the A/R after about 50 seconds. It might be due to the fact that the opening shot is a sunset, so might be harder to detect the edge of the black bars with certitude. We should probably leave that one.
What happens if you seek into the middle of the movie right at the start? Does detection then work as fast as with the other movies you tested?

I left Lawrence run for 1mn before giving up (past the audio only intro of course!). I can try to make a clip, or maybe you could add 2.20 in your list to see if it helps detection, if that's the reason, and I'll test again.
Same here: If you seek into a bright scene, does detection work then, after maybe max 40 seconds or so? In any case, a 1 minute sample from Lawrence would be great.

[maxstep]

Absolute heartbreaking shame on nvidia with dropping nnedi3 support. Can't believe it. Just jumped ship from AMD and everything is much worse with Nvidia. Now I find out that I can't even watch a video. Lovely.

Thank you to madshi for this incredible software!

[Warner306]

Absolute heartbreaking shame on nvidia with dropping nnedi3 support. Can't believe it. Just jumped ship from AMD and everything is much worse with Nvidia. Now I find out that I can't even watch a video. Lovely.

Thank you to madshi for this incredible software!

It sounds like it isn't anything that can't be fixed with some patience. madVR still has a long way to go. I am sure the problem will be fixed on one or both ends of Nvidia or madVR before too long.

[kopija]

Overlay mode working fine here on HD3000 and W10!

[mindz]

It only doesn't work properly when there is any hidden scaling going on, possibly like with DSR, or the player you are using is using Windows DPI scaling (which would be bad, as Windows would also upscale the video image then).

This got my attention. Im using MPC-HC and Im scaling fonts to 125% via the 'Control Panel\All Control Panel Items\Fonts' Change font size setting in Windows 7. Is Windows now first scaling my video, instead of madVR?

[chros]

Can I send standard window messages (like WM_COMMAND) to MadVR through MPC_HC (through it's webinterface)?
I'd like to change couple of things remotely, e.g. the input level.
I tried to capture the hotkeys (CTRL+J, CRTL+SHIFT+ALT+I , etc) in WindowDetective , but mpc-hc doesn't have a clue about them :(
Any idea how to do this?
Thanks