madVR - high quality video renderer (GPU assisted)

[boday]

I meant 10 % usage with mpv not 10 ms.

I guess you're not using high quality rendering in mpv (vo=opengl-hq)?

If I disable opengl-hq the GPU load is only 4% on my laptop.

[huhn]

if you mean --vo=opengl-hq i used that the problem is it ueses my main GPU too unlike madVR. if i disable the main gpu i get ~30% lowest powerstate. it's hard to measure mpv with more than 1 gpu...

[StinDaWg]

What is the biggest cause for high deinterlace stats? I have a 7850, and deinterlacing is set to vector adaptive. Sometimes max stats (5s) will get into the 20-27 range which causes occasional dropped frames. Movie frame interval is 20.85ms. Is there any settings I can change to lighten the load?
Any suggestions for this?

[huhn]

Sorry...still new to this forum. Here are the links:
https://www.dropbox.com/s/1w42vcywzcdoo8z/Capture.jpg?dl=0
[url]https://www.dropbox.com/s/ihc0h154r4r5kzc/Capture1.jpg?dl=0[/url

your screen shows a huge problem with the decoder queue.

did you run out of ram and who much ram did you give your igpu?

[huhn]

What is the biggest cause for high deinterlace stats? I have a 7850, and deinterlacing is set to vector adaptive. Sometimes max stats (5s) will get into the 20-27 range which causes occasional dropped frames. Movie frame interval is 20.85ms. Is there any settings I can change to lighten the load?

madVR has little to no influence to the way the gpu driver deinterlace the image.

max stats like your are totally normal for me but I don't drop a fame

Movie frame interval is 20.85ms.

this is a very strange number. normally it is 16.67 ms from 30i 60 fileds or 20 ms from 25i 50 fields.

are you sure your file is a normal interlaced file?

[nekrovski]

Hello

I have a problem with madvr. I watch videos/movies with MPC-HC and madvr as the video renderer. When I change the video renderer the problem goes away.
Anyways, this is kind of tough to explain. When I watch fullscreen videos/movies, most of the time there are 2 long bright/pixely lines from end to end. One vertical to the left end of the screen and one horizontal at the top end of the screen. The colors of the brighty/pixely lines usually depend on what's going on on the actual screen. These are only visible in fullscreen.

I'll provide any other info you need.

[huhn]

what hardware do you use and what are the version of your filters.
can you make a screen of the lines?
where these lines always here? since you use madVR?
what settings are you using in madVR?

[nekrovski]

what hardware do you use and what are the version of your filters.
can you make a screen of the lines?
where these lines always here? since you use madVR?
what settings are you using in madVR?

Hi and thanks for your reply.
Amd Phenom II x4 3.4ghz
radeon hd 5770

madvr 0.87.10
lav 0.61.2

I cannot make a screen of the lines.
They weren't always here and I can't really remember when they started showing up but probably after I started using madvr.

I remember following this guide for video watching (but I think I didn't make all the changes mentioned there)
https://imouto.my/watching-h264-videos-using-compute-unified-device-architecture-cuda/

Now I went back to that site and read this
https://imouto.my/madvr/

After configuring almost everything as said in the link above, the lines disappeared. It's one of the settings in madvr and if I have enough time I'll "trial and error" it.

[edit] after messing with the settings, i messed up a lot, there's not picture showing in fullscreen mode now.
lol
can I revert madvr to defaults?
[edit2] ok i found "C:\Program Files (x86)\LAV Filters\x86\madVR\restore default settings.bat"
picture is fine now, but the problem with the bright/pixely lines is back
now im gonna go ahead and follow that guide to see which option gets me rid off the brighty/pixely lines (i love naming them like this).

[romulous]

@madshi: Question from Blight. Are mouse-move events no longer filtered in recent versions? They are supposed to be routed through a call-back function.

Thanks,

romulous

[boday]

if you mean --vo=opengl-hq i used that the problem is it ueses my main GPU too unlike madVR. if i disable the main gpu i get ~30% lowest powerstate. it's hard to measure mpv with more than 1 gpu...
Ah I see.
your screen shows a huge problem with the decoder queue.

did you run out of ram and who much ram did you give your igpu?

I just received a reply from @cyberbeing on madshi's bug tracker (I reported the problem there before I can post here), and it makes a lot of sense to me:

http://bugs.madshi.net/view.php?id=232

The problem you're describing sound similar to what occurs when there is not enough GPU memory bandwidth for baseline functionality in madVR. For discrete GPUs, this usually occurs around 15-17GB/s. For iGPUs and APUs, you'll need to aim higher than that, since bandwidth is shared with system memory.

What is the system memory bandwidth like on your NUC? I believe you need a bare minimum of 20GB/s to not severely bottleneck an Intel iGPU with madVR.

A simple test is to see if render times improve significantly with 8bit 1080p video compared to 10bit 1080p video. There should be only a negligible difference between the two, unless you are bandwidth contained.

I only bought a single 4 GB DDR3L 1600 ram for the NUC as I thought that is more than enough for just a HTPC, but the problem is in this way I can only get 12.8 GB/s memory bandwidth. It seems that I need to bought another ram to form a dual-channel and get memory bandwidth doubled.

So I tested several 8-bit and 10-bit files and it looked like the case: rendering time for 8-bit files are mostly within 10 ms, but for 10-bit files they are about 25 ms. (Again, no imaging scaling, just chroma upsampling here)

[nekrovski]

Found the option that makes those brighty/pixely lines.
rendering -> general settings -> enable automatic fullscreen exclusive mode
when I uncheck this, the problem is gone.

[vivan]

So I tested several 8-bit and 10-bit files and it looked like the case: rendering time for 8-bit files are mostly within 10 ms, but for 10-bit files they are about 25 ms. (Again, no imaging scaling, just chroma upsampling here)What about debanding? (processing -> artifact removal)

[whitestar999]

i am using madvr for years now but with a very weak graphics card(mobility radeon 4570) so never tried more taxing processing algorithms.now i am upgrading my system & decided to get a graphics card.i know that even high end cards struggle with 1080p/i24fps to 1080p60Hz screen but my sources are mostly 480p/720p 23.976fps to be played on 1080p60Hz 32" led tv through hdmi & most of them are animation(japanese anime to be precise).would a card in the league of R7 250 be able to cut it or do i need a more powerful card to ultra-enhance my 480p/720p animation videos using 128 neuron nnedi3 to make it look good?

[huhn]

Found the option that makes those brighty/pixely lines.
rendering -> general settings -> enable automatic fullscreen exclusive mode
when I uncheck this, the problem is gone.

think about a bug report this shouldn't happen and this is still the best presentation mode madVR got.

http://bugs.madshi.net/my_view_page.php

[nekrovski]

think about a bug report this shouldn't happen and this is still the best presentation mode madVR got.

http://bugs.madshi.net/my_view_page.php
Hmm now even when I enable it, the bug doesn't appear.
BUT I had a different problem with a game (dota 2) terrain was missing, I was googling and found out that I had to restore the amd driver settings back to default. I've made some changes once in the drivers that apparently messed up my game and probably was the culprit for this mpchc/madvr problem as well.
Now all is fixed

[nekrovski]

OK this is a final confirmation. The culprit: enabling anti-alising samples 8x in

catalyst control center -> gaming -> 3d application settings -> system settings

this messes my game and shows the bright/pixely lines in mpchc when using madvr and exclusive fullscreen mode is enabled.

[Seedmanc]

How do I configure madvr with the goal of reducing lag it seems to introduce and at the same time maintain stability? I'm using it to output and upscale video I'm capturing in realtime from PS2 emulator window to watch it on a bigger screen, so minimizing lag is important as I'm playing the game. Another problem is that sometimes with too high settings for madvr or too much processing being applied to my realtime setup (I'm also processing the video in avisynth before output) madvr's framerate severely drops after a while and almost grinds to a halt, only relaunching the player helps, but only for a while. Why does this happen? It's like it clogs instead of just having lower but steady framerate. I tried looking at OSD, there doesn't seem to be any queue overflows.

[huhn]

the min cpu queue is 4 frames so lag free playback is impossible.

[StinDaWg]

madVR has little to no influence to the way the gpu driver deinterlace the image.

max stats like your are totally normal for me but I don't drop a fame



this is a very strange number. normally it is 16.67 ms from 30i 60 fileds or 20 ms from 25i 50 fields.

are you sure your file is a normal interlaced file?
Ya, it's 1080i25 H.264 file played back at 50hz. It's 20ms, that first number was a typo. I use fullscreen exclusive mode but I had to change to windowed to take a screenshot. I end up dropping frames every couple of minutes. This is with software decoding and my 7850 overclocked to 1050 core. If I use dxva2 copyback decoding instead i get ~45ms (5s) deinterlace.

http://t1.someimage.com/3J0vKSz.jpg (http://someimage.com/3J0vKSz)

[huhn]

try lower gpu and cpu queues. I did some test about this topic over a year ago.

and this looks like a TV recoding are you sure the recording is not broken things like this happen all the time with TV records.

are you using the newest driver?

[StinDaWg]

try lower gpu and cpu queues. I did some test about this topic over a year ago.

and this looks like a TV recoding are you sure the recording is not broken things like this happen all the time with TV records.

are you using the newest driver?
Lower queues make it worse, more dropped frames. Ya, it happens all the time with 1080i videos, both 25 fps and 30 fps. I'm using 13.12 driver. The 14x versions have worse performance with NNEDI3.

[huhn]

here some screens.

http://abload.de/img/highqueueshuj5c.png
http://abload.de/img/queues7oki9.png

as you can see the queues have a huge effect on the deint time but don't use 4...

have you checked the VRAM usage I could easily reach 1 GB limited with my old 6770. most 7850 have 2 gb anyway so unlikely.

are you using something like de noise or stuff like this in the amd driver this is used for deinerlacing and dxva scaling and I guess this is not totally free or did you forced anti aliasing and stuff like this for 3D games?

I use a r9 270 this is a 7870 so I should have the same problem.

[digitech]

Im planning to build an htpc, Im thinking to use windows 7 64 bits vs windows 8, anyone knows which OS ver. suits the best for MadVR movie playback?

[Anime Viewer]

here some screens.

http://abload.de/img/highqueueshuj5c.png
http://abload.de/img/queues7oki9.png

as you can see the queues have a huge effect on the deint time but don't use 4...



Did you forcibly add interlacing into Aikatsu?!? I'm guessing you forcibly added it to that video just to use it as an example of the de-interlacing/queue effect.

Back to the topic at hand. Why wouldn't you recommend the queues being set to 4 for CPU and GPU?

[jkauff]

Im planning to build an htpc, Im thinking to use windows 7 64 bits vs windows 8, anyone knows which OS ver. suits the best for MadVR movie playback?
Since madshi has stated he plans to stick with DirectX 9 and 32-bit for the foreseeable future, it shouldn't matter which one you use.

[StinDaWg]

here some screens.

http://abload.de/img/highqueueshuj5c.png
http://abload.de/img/queues7oki9.png

as you can see the queues have a huge effect on the deint time but don't use 4...
I tried the default settings of 12/8/8 and there is no difference in max deint time, and the presented frames is dangerously low.

http://t1.someimage.com/JuK3Eyq.jpg (http://someimage.com/JuK3Eyq)

have you checked the VRAM usage I could easily reach 1 GB limited with my old 6770. most 7850 have 2 gb anyway so unlikely.
600 MB

are you using something like de noise or stuff like this in the amd driver this is used for deinerlacing and dxva scaling and I guess this is not totally free or did you forced anti aliasing and stuff like this for 3D games?
No, all post processing in CCC is turned off.
Answers in bold.

Edit: I unchecked "use a separate device for dxva processing" and max deint times dropped from 27 to 3-6. Seems that was the issue. Not sure why it's on by default but it's not working well with my card.

Double edit: I spoke too soon. Render times are low but I'm still getting random spikes. A cluster of 8 dropped frames in 5 minutes.

[QBhd]

Im planning to build an htpc, Im thinking to use windows 7 64 bits vs windows 8, anyone knows which OS ver. suits the best for MadVR movie playback?

I find Win 8.1 marginally better for madVR... Issues with desktop composition in Win 7 (HDTV & Monitor setup) are completely gone in 8.1

QB

[boday]

I just received a reply from @cyberbeing on madshi's bug tracker (I reported the problem there before I can post here), and it makes a lot of sense to me:

http://bugs.madshi.net/view.php?id=232



I only bought a single 4 GB DDR3L 1600 ram for the NUC as I thought that is more than enough for just a HTPC, but the problem is in this way I can only get 12.8 GB/s memory bandwidth. It seems that I need to bought another ram to form a dual-channel and get memory bandwidth doubled.

So I tested several 8-bit and 10-bit files and it looked like the case: rendering time for 8-bit files are mostly within 10 ms, but for 10-bit files they are about 25 ms. (Again, no imaging scaling, just chroma upsampling here)
Now I think this is not a madVR bug, but just out of curiosity, I tested some files again on another laptop with i7-3520m and HD 4000, with- or without the second memory chip installed. The laptop is using DDR3 1600 MHz memory, so with single-channel I get 12.8 GB/s bandwidth, and with dual-channel I get 25.6 GB/s.

I can see the performance difference between the two conditions, but I can still see the rendering time are quite different (like 4 ms vs 10 ms), not like "only a negligible difference" described above, though I can see the GPU load is reduced to half in some cases thanks to the higher bandwidth.

And what really bothers me is that even working on a single-channel with only 12.8 GB/s bandwidth, madVR still works better on this laptop than the NUC: it can play most of my files without problem if I lower the settings a bit, but the NUC can't even smoothly play with the lowest setting of madVR. I think HD 5000 on the NUC should be significantly better than HD 4000, so I don't know how this could be explained, and it really made me a bit hesitated to buy another memory chip for the NUC.
What about debanding? (processing -> artifact removal)

With debanding enabled I got ~30 ms in average rendering time, playing the same clip without debanding I got ~25 ms. But why do you ask?

[vivan]

With debanding enabled I got ~30 ms in average rendering time, playing the same clip without debanding I got ~25 ms. But why do you ask?Debanding is quite heavy, for me it triples rendering time (when there's no scaling).

[huhn]

Did you forcibly add interlacing into Aikatsu?!? I'm guessing you forcibly added it to that video just to use it as an example of the de-interlacing/queue effect.

Back to the topic at hand. Why wouldn't you recommend the queues being set to 4 for CPU and GPU?

it's a transport stream so the original TV recoding and this one is 30i 60 fields with a 3:2 cadence. if you just play this back with madVR or any other player it is deinterlaced because it most be done most TV would do the same if not all.

of cause this is telecine (still interlaced) so detelecine would be better the cm are not telecine not sure if they are PsF or normal 60 field interlacing.

but EVR would use deinterlacing too so I don't force anything on this.
TV broadcast works like this I can't change this ^^

why shouldn't you use 4 gpu and cpu queues?
get a telecine interlaced file with 3:2 set it to IVTC mode.
you will see the backbuffer/present queue is a 0-1/" when your screen is at 23p and frames get dropped from time to time.
next problem is that the CPU queue is the back buffer for xy subfilter. 4 frames is not the biggest backup buffer 14+ is a lot more save.

or in short the default values are a very good choice between gpu Vram/ram usage and back buffer for "problem"

[huhn]

@StinDaWg
your decoder queue is not 100% full is this always like this it is usually 13-14/14 or like that. I'm pretty sure you know what I mean.

is something running in the back that uses the gpu or the cpu like windows gadgets or stuff like this?

[StinDaWg]

@StinDaWg
your decoder queue is not 100% full is this always like this it is usually 13-14/14 or like that. I'm pretty sure you know what I mean.

is something running in the back that uses the gpu or the cpu like windows gadgets or stuff like this?
No. It's never full when using deinterlacing. It's always full when playing any other progressive file.

[boday]

Debanding is quite heavy, for me it triples rendering time (when there's no scaling).

OK, but since in my case the setup is not quite adequate to use madVR, it's probably not good to serve as a reference. :o

Anyway, I just bought another memory module to form a dual-channel memory, and the performance got much better, but it's still worse than my laptop. I ran some benchmarks and confirmed the GPU is working well enough as it is, so I guess I was just set my expectation too high for a HD 5000 - it's not that much better than HD 4000, at least for video playback with madVR.

Just to be clear: it good enough to play lanczos3+ar for stuff on a 1080p screen, but not good enough to use the same settings on a 1440p screen, which involves a lot of 1080p -> 1440p upscaling.

[ryrynz]

Would anyone be interested in setting up a madvr benchmark thread?

[Seedmanc]

Might be interesting, I'd like to know what effect on speed and lag every option has.

[dansrfe]

Would anyone be interested in setting up a madvr benchmark thread?

Sounds useful. Just need a write-up to guide madVR users on how to bench and what stats to post.

[BetA13]

Question....

What am i doing here? does this even work? Is this giving me a worse experience?

to be honest i cant see a difference, but the blacks seem more fluid, i mean less banding.. could that be?
Total noob here :)

Pic:
http://s27.postimg.org/5a5y5lp33/sdgfsdfs.jpg (http://postimg.org/image/5a5y5lp33/)

edit:
damn, i saved it as gif..argh.. next time ill do better..

[nevcairiel]

I'm sorry, but you're imagining things, or some low quality conversion algorithm is adding noise that blurs out some of your "banding".
You should output video as-is untouched, anything else is just useless conversion and distorts the image.

[BetA13]

I'm sorry, but you're imagining things, or some low quality conversion algorithm is adding noise that blurs out some of your "banding".
You should output video as-is untouched, anything else is just useless conversion and distorts the image.

hehe, i knew i was wrong :)
guess its this placebo effect or what you said..

anyhow, Thanks a lot for the "quick" answer.

cheers..

[vivan]

I'm sorry, but you're imagining things, or some low quality conversion algorithm is adding noise that blurs out some of your "banding".Isn't debanding is on by default?

[ryrynz]

Isn't debanding is on by default?
Yup. This would likely be what he's seeing. Low quality videos are going to have blocking in the blacks that's improved how he described. No placebo there.

[nevcairiel]

I think you misunderstood his question. He disabled 8-bit output modes from LAV Video and forced LAV to convert to 10-bit before sending the image to madVR, and this "seems" to change the picture for him.
I don't think its related to madVRs debanding at all, unless debanding works differently on 10-bit input.

[sneaker_ger]

And madVR debanding is off by default.

[ryrynz]

He disabled 8-bit output modes from LAV Video and forced LAV to convert to 10-bit before sending the image to madVR, and this "seems" to change the picture for him.
I don't think its related to madVRs debanding at all, unless debanding works differently on 10-bit input.
There was someone earlier who said the exact same thing.. I compared the images and they were identical bar dithering for me.. Is it possible the display has some effect here? If that's two people saying that they see differences I'd like to see some actual proof of it.

[Arm3nian]

Im planning to build an htpc, Im thinking to use windows 7 64 bits vs windows 8, anyone knows which OS ver. suits the best for MadVR movie playback?

I had better luck with 7. Sometimes on 8.1 skipping through a video would cause madvr to crash. No problems whatsoever on 7 64bit.
Now I think this is not a madVR bug, but just out of curiosity, I tested some files again on another laptop with i7-3520m and HD 4000, with- or without the second memory chip installed. The laptop is using DDR3 1600 MHz memory, so with single-channel I get 12.8 GB/s bandwidth, and with dual-channel I get 25.6 GB/s.

I can see the performance difference between the two conditions, but I can still see the rendering time are quite different (like 4 ms vs 10 ms), not like "only a negligible difference" described above, though I can see the GPU load is reduced to half in some cases thanks to the higher bandwidth.



Dual channel is important for an igpu config so the cpu can use 1 channel and the igpu use the other.

[flashmozzg]

I had better luck with 7. Sometimes on 8.1 skipping through a video would cause madvr to crash. No problems whatsoever on 7 64bit.


Don't think thats windows problem and not your setup.
Have no such probs on 8.1 x64 but it's much better in terms of WPM and other. + it'd be muh easier to upgrade to 9 which supposedly comes spring/summer 2015.

[Seedmanc]

I thought there's no madvr x64.

[michkrol]

I personally had no problems with playback on madVR on both Windows 7 (x64) and Windows 8.1 (x64), so it might be anything from configuration differences to driver problems.
I thought there's no madvr x64.
There isn't, but you can use x86 (32 bit) applications on Windows x64 Vista/7/8/8.1. Most of us are on Windows 7/8.1 x64 and just download x86 (32 bit) media players to use madVR.

[burfadel]

Yes, you can run 32-bit Windows applications on x64 Windows, and there isn't a penalty performance wise than doing so. Actually technically, a 32-bit application on 64-bit Windows should actually perform fractionally faster due to other factors in the system. Having a 64-bit version, especially one that is optimised, will have advantages in speed though and should perform better or have a bit lower CPU utilisation. Since LAV Filters and MPC-HC are available in 64-bit and kept congruent with their 32-bit counterparts, it makes sense to go to a 64-bit version of madVR now. However, madshi hasn't expressed any inclination in doing so.

[huhn]

Yes, you can run 32-bit Windows applications on x64 Windows, and there isn't a penalty performance wise than doing so. Actually technically, a 32-bit application on 64-bit Windows should actually perform fractionally faster due to other factors in the system. Having a 64-bit version, especially one that is optimised, will have advantages in speed though and should perform better or have a bit lower CPU utilisation. Since LAV Filters and MPC-HC are available in 64-bit and kept congruent with their 32-bit counterparts, it makes sense to go to a 64-bit version of madVR now. However, madshi hasn't expressed any inclination in doing so.

he multiple time said there will be a 64 bit version at least for 1.0.

he said it would take about 2 days to create a 64 bit version and he like to spend the time on something more meaning full to him like adding feature and things like that.