madVR - high quality video renderer (GPU assisted)

[madshi]

Originally Posted by aufkrawall

there is a problem with defining a custom resolution via Custom Resolution Utility.
I use it to force the display to stay at 75Hz. To do so set up a custom resolution and disable the checkbox "Include extension block" on Nvidia hardware.
In windowed mode now madVR reports 75Hz, but Windows still offers only 60Hz resolution. I don't know why this is that way, maybe oddness of Nvidia driver.
The problem is that madVR fails to enter FSE this way. The Nvidia Control Panels reports 75Hz only.
Could you please extend the refreshrate lookup so that madVR can also detect such custom resolutions, making FSE working?

That won't work. I have to tell D3D9 which display mode to use, and it must be a display mode which really exists. The D3D9 API requests it that way, at least for FSE mode.

Originally Posted by leeperry

I'm currently having troubles getting any kind of DXVA2(in LAV) to work in FS as when going from windowed to FSW(with the latest stable build of PotP) the picture is vertically stretched, then after a second the AR is applied and playback finally starts. OTOH with FSE, the backbuffer queues never fill and it's stuttering like hell. Need to run more tests I guess.

I hope you aren't using the old FSE mode, like you did in XP? In newer OSs you should really be using the new FSE mode. I wouldn't recommend to use DXVA2 decoding, unless your CPU is rather weak. I prefer software decoding. Of course DXVA2 deinterlacing is a different topic, it's currently needed for native video content.

Originally Posted by leeperry

I haven't gone through all previous pages yet but it seems that the backbuffer queues problem is being investigated.

No, it's not.

Originally Posted by cyberbeing

Yes, and when you exit playback madHcCtrl is closed, as would be the normal case of accessing madVR settings outside of playback. My current tray icon setting is "show tray icon on local pc" and "Lan access" disabled.

I switched to a newer NVidia GPU in my dev PC and tried with GraphStudioNext and I can now reproduce the 5 second delay... I'll look into that next.

Originally Posted by cyberbeing

Is it expected for OpenCL dither to increase madVR CPU load significantly when scaling? GPU load also seems to change a lot, depending on output size. Are you dithering after resize rather than prior? It would also seem that GPU load exactly doubles if you enable Smooth Motion on top of OpenCL dither, which I find a bit strange.

Originally Posted by nevcairiel

Dithering should be one of the very last steps in rendering, after scaling, after smoothmotion.

Dithering reduces the bitdepth of the image, so doing it any earlier defeats its purpose.

^

Random dithering is currently done before drawing the OSD. Error diffusion is done even later, after drawing the OSD. Both is done after scaling and after smooth motion FRC. As nevcairiel said, it must be that way. Of course when using smooth motion FRC, the number of rendered frames increases, so that will drive up error diffusion cost.

I've read somewhere that when using NVidia OpenCL, the CPU consumption goes up because for some reason NVidia OpenCL just works that way. I think the thread waiting for OpenCL to finish its work is probably running at 100%. Don't ask me why. I don't know if this happens with AMD or Intel, too, haven't really tested that.

Originally Posted by JarrettH

Maybe I don't understand this correctly, but with NNEDI, what are the chances of upscaling a video by precisely 2x? Does that mean your video would need to be half of the 1920 pixel width (960) to work?

Originally Posted by ThurstonX

If I understand it correctly, nnedi will double a small video, then whatever scaling method you've selected will take that and stretch it to the target rectangle size. That's just from reading along the past few pages.

If you double past the target size, then the downscaling method will kick in.

^

Originally Posted by fairchild

Hopefully at some point someone can explain how to setup a profile+script to use NNEDI to say just do less than 720p content (not sure if it benefits 720p to 1080p) to get max image quality. I guess the purpose of NNEDI is to get even more quality out of your image/chroma scaling when going from say a DVD which is at 720x480 to a 1920x1080 screen.

Edit: kinda figured it out a bit, for lower res media, then can use the following options which bypasses using NNEDI for 720p and up content. (since difference is probably sooooo minute anyway)

use NNEDI3 to double luma if scaling 2.0x 64 neurons (bypasses 720p)
use NNEDI3 to quadruple luma if scaling is 3.0x 16 neurons

You can either do that, or use profiles if you want even more control over which algorithm is used when.

Originally Posted by dansrfe

You've said that you recommend we select a "cheap" upscaler algorithm such as Lanczos3 AR when NNEDI3 is selected. What happens when NNEDI3 is not in use and only Lanczos3 AR is used to upscale Luma? In that case wouldn't you recommend it switches to another algorithm, such as Jinc3 AR, automatically?

You can do that with profiles, if you want.

-------

I've found a bug in v0.87 which could explain some of the new problems. So all of you who have D3D9 failures or performance regressions, please check this test build:

http://madshi.net/madVR871a.rar

[madshi]

Originally Posted by truexfan81

I have to say i love this profile stuff.

Profiles for Scaling, profiles for deband/deinterlace.
Now i can watch any file i want to without having to ever manually change settings.

madshi thank you so much :-)

Glad you like it...

[huhn]

Originally Posted by madshi

I've found a bug in v0.87 which could explain some of the new problems. So all of you who have D3D9 failures or performance regressions, please check this test build:

http://madshi.net/madVR871a.rar

freeze with file change in fullscreen exclusive is fixed for me.

the report from Today 09:47 is fixed too.

but video-film mode switch with refresh rate change done through a hot key results in a black screen now.

only happens from video to film mode and only about 50 %

[madshi]

Some progress, at least, thx for the quick feedback. And you can confirm that the remaining issue definitely didn't occur in v0.86.11 *at all*, is that correct?

[huhn]

Originally Posted by madshi

Some progress, at least, thx for the quick feedback. And you can confirm that the remaining issue definitely didn't occur in v0.86.11 *at all*, is that correct?

with 86.11 the player crashes now i'm sure this didn't happen before but is a new player version always after 3 switches. or even after the first one and that defiantly didn't happen before
this never happen yesterday when when 86.11 settings where reseted after downgrading from 87.1
87.1 instant crash
87.1a blackscreen
86.11 player crash

i try it again after clearing the registry and restarting the pc but at the moment is looks like a old bug in a new form.

[romulous]

Originally Posted by romulous

Apologies if this is already a known issue, but video playback fails with all 0.87.x versions so far when I enter fullscreen mode on a second display (not using any of the new features of 0.87.x, and FSE mode disabled). Returning to 0.86.11 makes it work again.

Originally Posted by madshi

I've found a bug in v0.87 which could explain some of the new problems. So all of you who have D3D9 failures or performance regressions, please check this test build:

http://madshi.net/madVR871a.rar

Hi madshi,

This new build finally fixes my problem. However, when you fullscreen with madVR (FSE disabled), there is a delay in playback (a slight pause) before playback continues. This delay is in 0.86.11 as well - but I think it is a slightly longer delay in this new 0.87 build. So the problem is fixed, but I don't think performance is back to 0.86.11 levels as yet.

[DragonQ]

Originally Posted by madshi

I've found a bug in v0.87 which could explain some of the new problems. So all of you who have D3D9 failures or performance regressions, please check this test build:

http://madshi.net/madVR871a.rar

None of these fixes the performance regression.

[leeperry]

Originally Posted by madshi

I hope you aren't using the old FSE mode, like you did in XP? In newer OSs you should really be using the new FSE mode. I wouldn't recommend to use DXVA2 decoding, unless your CPU is rather weak. I prefer software decoding. Of course DXVA2 deinterlacing is a different topic, it's currently needed for native video content.

I'm of course using the new rendering path

Well, I'm confused because Nev told me that DXVA2 works even better than CUVID, I was hoping to get the same kind of performance with DXVA2CP off this HD7850 with mVR as I was getting with CUVID....that'll be a no-go?

Software decoding works perfectly but I would really rather have the GPU taking care of video decoding as I bought a low-end CPU expecting to be able to use GPU video decoding.

If DXVA2 really cannot provide the same kind of performance as CUVID, I guess I might switch teams again....this said, IME Asus/nvidia graphic cards are dead silent and this cheap HD7850 is already getting on my nerves as the rest of the rig is whisper quiet, I'll try to make a custom fan speed curve as the fan is fairly capable.

Originally Posted by madshi

No, it's not.

Oh, well then I'll try to come back with screenshots. Are the backbuffer queues supposed to fill up equally whatever NNEDI is enabled or not?

[romulous]

Originally Posted by romulous

Hi madshi,

This new build finally fixes my problem. However, when you fullscreen with madVR (FSE disabled), there is a delay in playback (a slight pause) before playback continues. This delay is in 0.86.11 as well - but I think it is a slightly longer delay in this new 0.87 build. So the problem is fixed, but I don't think performance is back to 0.86.11 levels as yet.

Scrub this - the delay is dependant on the file being played. I tested with an AVI file on 0.87 and a MOV file on 0.86.11 - and when I check that AVI file with 0.86.11, that delay is longer than with the MOV file, and seems to be about the same as this new 0.87 build.

I do note however that if I enable the nnedi3 box, the problem returns with this fixed build (fullscreen to second display, playback halts and player goes to 100% CPU). The only difference this time being previously, I could still close the player as normal. Now I can't - task manager intervention is required.

So basically if I used the fixed build with no changes to my 0.86.11 configuration, everything is ok. When I start enabling the new 0.87 options, that is when things start to break.

[madshi]

Originally Posted by huhn

i try it again after clearing the registry and restarting the pc but at the moment is looks like a old bug in a new form.

Ok, good. I'm mostly interested in *new* bugs at the moment.

Originally Posted by DragonQ

None of these fixes the performance regression.

Ok. Will have to look at that next.

Originally Posted by leeperry

Well, I'm confused because Nev told me that DXVA2 works even better than CUVID, I was hoping to get the same kind of performance with DXVA2CP off this HD7850 with mVR as I was getting with CUVID....that'll be a no-go?

I don't know, I don't use DXVA decoding, never did. Copyback has been a problem with older AMD generations, but it's supposed to be improved in 7xxx. Have you tried native DXVA2 decoding instead? Of course you use forced film mode that way. FWIW, I've *always* said that I prefer and recommend software decoding...

Originally Posted by leeperry

Software decoding works perfectly but I would really rather have the GPU taking care of video decoding as I bought a low-end CPU expecting to be able to use GPU video decoding.

As long as the CPU can handle the decoding load, what's wrong with using software decoding? Or do you have AviSynth scripts you want to run on the CPU?

Originally Posted by leeperry

If DXVA2 really cannot provide the same kind of performance as CUVID, I guess I might switch teams again....

However, OpenCL performance is better with AMD.

Originally Posted by leeperry

Oh, well then I'll try to come back with screenshots. Are the backbuffer queues supposed to fill up equally whatever NNEDI is enabled or not?

If your GPU is fast enough then I would expect all queues to fill up nicely. That said, the whole OpenCL stuff in madVR is just a couple of days old, so there's very little experience how it affects the various queues.

Originally Posted by romulous

I do note however that if I enable the nnedi3 box, the problem returns with this fixed build (fullscreen to second display, playback halts and player goes to 100% CPU). The only difference this time being previously, I could still close the player as normal. Now I can't - task manager intervention is required.

So basically if I used the fixed build with no changes to my 0.86.11 configuration, everything is ok. When I start enabling the new 0.87 options, that is when things start to break.

Maybe I could say something if I knew which GPU/OS you were using? Anyway, my priority right now is on making the "old" features work alright again. Only after that I'm ready to polish the OpenCL stuff...

[huhn]

Originally Posted by leeperry

I'm of course using the new rendering path

Well, I'm confused because Nev told me that DXVA2 works even better than CUVID, I was hoping to get the same kind of performance with DXVA2CP off this HD7850 with mVR as I was getting with CUVID....that'll be a no-go?

Software decoding works perfectly but I would really rather have the GPU taking care of video decoding as I bought a low-end CPU expecting to be able to use GPU video decoding.

If DXVA2 really cannot provide the same kind of performance as CUVID, I guess I might switch teams again....this said, IME Asus/nvidia graphic cards are dead silent and this cheap HD7850 is already getting on my nerves as the rest of the rig is whisper quiet, I'll try to make a custom fan speed curve as the fan is fairly capable.


Oh, well then I'll try to come back with screenshots. Are the backbuffer queues supposed to fill up equally whatever NNEDI is enabled or not?

if you got a intel cpu try quicksync it's faster than nvidia dxva and worlds faster than amd dxva.

and nev is totally right he meant dxva2 on a nvidia card.
cuvid is very useful on old nvidia cards but in newer cards both use the same "chip" so cuvid is kinda like dxva copy back just puts your card in highest power state wasting power like a pro.
and dxva is a dead end you can't use this for newer codecs like h265 or vp9.

@ the new backbuffer settings they don't make sense to me.
the max backbuffer valve you can get is always gpu queue -1 at default it is 8 gpu and 8 backbuffer so the backbuffer stays at 6-7 at best.

but these are only my observations with intel, nvidia and amd.

[madshi]

Originally Posted by huhn

@ the new backbuffer settings they don't make sense to me.
the max backbuffer valve you can get is always gpu queue -1 at default it is 8 gpu and 8 backbuffer so the backbuffer stays at 6-7 at best.

but these are only my observations with intel, nvidia and amd.

Which new backbuffer settings do you mean? Are you talking about windowed or FSE mode? In FSE mode you don't specify the number of backbuffers, but you specify the number of prepresented frames. And that is the only default value I changed in v0.87, IIRC. So I'm confused with what you mean...

-------

@DragonQ, and everyone else who's having performance problems. Does this build help?

http://madshi.net/madVR871b.rar

Is it correct to say that all performance issues have to do with DXVA deinterlacing or DXVA decoding somehow?

@cyber, this should also fix the settings delay.

[huhn]

Originally Posted by madshi

Ok, good. I'm mostly interested in *new* bugs at the moment.

at the moment i can't get 86.11 running properly.
i get 40 ms render time with default settings with a 1080p23 source on a 1080p screen. so i broke it and my observations are "useless" atm.

i didn't bother you until i now if it's new bug or a old one.

[James Freeman]

but video-film mode switch with refresh rate change done through a hot key results in a black screen now.

only happens from video to film mode and only about 50 %

87.1b

Happens here too.

When changing to Film mode & automatic refresh rate the screen/window turns black.
When force disabling Deinterlacing (in Film mode, after the refresh changed), the video appears again.
BUT, when I manually change the refresh to 72Hz in windows, then open MPC and change to Film mode, everything is OK.


Render times (1080p24) & 87.1b:

Average Stats,
rendering 2.24ms
present 0.08ms

Max Stats:
rendering 2.50ms
present 0.15ms

Full buffers on everything.

I have never had speed issues with any of the builds from 86.11 to 87.1b.

[nevcairiel]

Originally Posted by leeperry

Well, I'm confused because Nev told me that DXVA2 works even better than CUVID, I was hoping to get the same kind of performance with DXVA2CP off this HD7850 with mVR as I was getting with CUVID....that'll be a no-go?

AMDs hardware decoder is rather bad compared to NVIDIA. Its extremely slow.

So while DXVA2CB will work on AMD 7000 series GPUs, it'll not reach the levels of NVIDIA hardware because AMDs decoder is just really slow. It can't even be guaranteed that it can do 60fps properly. LAV can only work with the hardware you have, so while i do favor DXVA2CB over CUVID, that doesn't translate that it will magically be fast on other hardware.

[huhn]

Originally Posted by madshi

Which new backbuffer settings do you mean? Are you talking about windowed or FSE mode? In FSE mode you don't specify the number of backbuffers, but you specify the number of prepresented frames. And that is the only default value I changed in v0.87, IIRC. So I'm confused with what you mean...

you are right the present in advance changed to 8.

but the "rule" max gpu queue -1 = max backbuffer/present is still "true". same rule goes for cpu and ivtc(haven't check the ivtc part for a while).
and that's why i don't get why gpu and present are set to 8.
but this has nothing to do with backbuffer mixed them up sorry. and this normally didn't harm only found problems with settings like these with my intel hd4000 and i think this was a long time ago.

i updated back to 87.1a and this version is working, how can i break a old version...

[huhn]

Originally Posted by James Freeman

87.1b

Happens here too.

When changing to Film mode & automatic refresh rate the screen/windows turns black.
When force disabling Deinterlacing (in Film mode, after the refresh changed), the video appears again.
BUT, when I manually change the refresh to 72Hz in windows, then open MPC and change to Film mode, everything is OK.


Render times (1080p24) & 87.1b:

Average Stats,
rendering 2.24ms
present 0.08ms

Max Stats:
rendering 2.50ms
present 0.15ms

Full buffers on everything.

can you plz try this with 86.11 my version is crashing now and render times are about 10 ten higher than normal

[omarank]

Hi Madshi

I just tried the latest test build of madVR (madVR871b). As others have reported I too see a regression in performace while playing interlaced videos as compared to v0.86.11. At the moment I am not using any of the new features, not even debanding though debanding works on my system and I find it very impressive.

I have encountered one minor bug in the new release which I guess hasn't been reported. When I press the shortcut keys for <chroma upscaling algorithm - set to "Jinc">, it uses NNEDI3 algorithm and I see a green screen.

I have a GTX260 and I am using v332.21 drivers.

[ryrynz]

HD3000 on 8.1 with 3347 drivers using 0.87.1, when fullscreen exclusive is enabled and not ticking 'use separate device for presentation' and after transitioning to next file after playback I get a black screen.

The black screen is locked and I can't go into windowed mode without bring up Task Manager. Once doing so I can play the video fine in windowed mode, entering fullscreen again will show a frozen picture, audio plays fine.
Entering windowed mode and fullscreen again and FSE mode activates with no issues and video plays fine.

Ticking 'use separate device for presentation' fixes this issue, no issue with deband build 14.

[madshi]

Originally Posted by James Freeman

87.1b

Happens here too.

When changing to Film mode & automatic refresh rate the screen/window turns black.
When force disabling Deinterlacing (in Film mode, after the refresh changed), the video appears again.
BUT, when I manually change the refresh to 72Hz in windows, then open MPC and change to Film mode, everything is OK.

As I said, I'm mostly only interested in *new* bugs atm. So please double check whether this problem does not occur in v0.86.11. If it also occurs in v0.86.11 then it's an old bug, which you can enter into the madVR bug tracker, but I'll not look at it right now. Just want to make sure that v0.87.x doesn't contain new bugs compared to v0.86.11 right now.

Originally Posted by huhn

but the "rule" max gpu queue -1 = max backbuffer/present is still "true". same rule goes for cpu and ivtc(haven't check the ivtc part for a while).

No, it's not true. E.g. when playing 24fps content @60Hz in FSE mode, you can easily have a render queue of 7-8/8 and a present queue of 11-12/12.