madVR - high quality video renderer (GPU assisted)

[jmonier]

Hi guys, i am having a problem that i think is related with madvr.
Sometimes when i am playing a tv show and i skip to the next episode, it just shows a black screen, and even returning from fullscreen i can't play the episode. i have to close mpc and open the episode again. this do not happen when i am not in fullscreen.
Maybe is related to the change in refresh rate, i use the option to change refresh rate in madvr.
madvr version 0.89.19.
When i am not in fullscreen i can skip all episodes in a season without any problem, but in fullscreen it sometimes give me this blackscreen.
Using windows 10.

Try 90.10. There was a memory leak which was fixed in 90.10. I had similar problems and 90.10 seems to have fixed it.

[leeperry]

Are you saying that all the Image Downscaling options (and perhaps others?) are just renamed version of Bicubic?
They're all Bicubic and all subjective impressions are pure placebo, shouldn't have taken the red pill. Catmull-Rom = Bicubic50, that's all AFAIK ))

[nevcairiel]

They're all Bicubic and all subjective impressions are pure placebo, shouldn't have taken the red pill. Catmull-Rom = Bicubic50, that's all AFAIK ))

Not "all" are Bicubic, but Catmull Rom and Mitchell Netravali are (and Bicubic and SoftCubic of course). But they are all different variations of the cubic interpolation. That C-R happens to be exactly the same as Bicubic50 is just the way it is.

[FDisk80]

Hi guys, i am having a problem that i think is related with madvr.
Sometimes when i am playing a tv show and i skip to the next episode, it just shows a black screen, and even returning from fullscreen i can't play the episode. i have to close mpc and open the episode again. this do not happen when i am not in fullscreen.
Maybe is related to the change in refresh rate, i use the option to change refresh rate in madvr.
madvr version 0.89.19.
When i am not in fullscreen i can skip all episodes in a season without any problem, but in fullscreen it sometimes give me this blackscreen.
Using windows 10.

I have the same problem if I use MPC-BE with internal filters.
But if I use MPC-BE with LAV filters the problem goes away.

Just tested and it's the same problem also with the latest madVR version.

[Warner306]

Why is there NNEDI3 for Chroma Upscaling and Image Doubling/Quadrupling, but not Image Upscaling?

I am also a bit perplexed about the whole Bicubic thing. Are you saying that all the Image Downscaling options (and perhaps others?) are just renamed version of Bicubic? Catmull-Rom = Bicubic50, SSIM = Bicubic125, Jinc/NNEDI3/others = some other Bicubic???

After 345 posts, I have a hard time believing any of your questions are serious? Do you really want an answer or are you trying to flood this forum with useless posts?

You must know what NNEDI3 is -- image doubling, not image upscaling. And of course not all of the downscaling options are Bicubic; they are no different than the image upscaling algorithms. Many probably use very similar approaches to image scaling, with similar math -- but they are not the same.

[clsid]

madshi,
MPC-HC menu > View > Statistics shows the wrong framerate when using madVR. With all other renderers it displays the playback framerate of the video (video framerate * playbackspeed). In case of madVR it shows the screen refresh rate.

According to U78 the info is gathered using "get_AvgFrameRate()" method provided by the "IQualProp" interface of the filter graph.

[nevcairiel]

madshi,
MPC-HC menu > View > Statistics shows the wrong framerate when using madVR. With all other renderers it displays the playback framerate of the video (video framerate * playbackspeed). In case of madVR it shows the screen refresh rate.


Actually it shows the actually rendered frame rate on other renderers. I think madVR does the same thing, except that in some of the rendering modes its going to present a new frame for every VSYNC, hence showing screen refresh.

So I wouldn't necessarily call it wrong. If you use a mode that only presents every frame once, ie. D3D9 or Overlay, it shows what you might expect.

[clsid]

I use D3D9. Disabling "present several frames in advance" indeed gives the framerate value I expected.

[ryrynz]

SSIM = Bicubic125?

C'mon man. You've been here a while now. LOL.

[markanini]

I've found some setting which I think substantially improves low-res/bitrate sources, but looks good for good sources too:
Image enhancements: Enhance detail 1.0 (pre-resize)
Image doubling: NEDI
Image upscaling: Lanczos3 AR sigmoidal
Upscaling refinement: SuperRes 3

[x7007]

Madshi ? there is no possible way that CUVID will work with NNEDI3 with newest drivers ?


Should I use Chroma Upscaling NNEDI3 32 neurons + Activate SuperRes + Upscaling refinement + SuperRes refine the image only once after upscaling is complete.

[ryrynz]

Madshi ? there is no possible way that CUVID will work with NNEDI3 with newest drivers ?


Why you still using it?

[x7007]

Why you still using it?


I wonder why it wouldn't work, we might need it on the future. CUVID can decode things that DXVA2 can't, as LAV Filter allows.

[huhn]

I wonder why it wouldn't work, we might need it on the future. CUVID can decode things that DXVA2 can't, as LAV Filter allows.

that's not really correct.

adding something like DXVA mpeg4 ASP decoding is simply not worth the time. and you gain absolutely nothing by using DXVA2 for such codecs.

nvidia broke it and they don't seem to care at all. in theory DXVA2 can do everything important CUVID can do.

[x7007]

that's not really correct.

adding something like DXVA mpeg4 ASP decoding is simply not worth the time. and you gain absolutely nothing by using DXVA2 for such codecs.

nvidia broke it and they don't seem to care at all. in theory DXVA2 can do everything important CUVID can do.

Understood. I just thought it can be easy to restore the support from one side at least or the "other" side.

Could you help me with the other question ? if we are on the subject on NNEDI3 , I just don't know if I need to enable SuperRes on both Chroma Upscaling and refinement.

[huhn]

i can only tell you want i do.

you should ask your own eyes. that's what matter.

nnedi3 on chroma is totally useless in my eyes.

[ryrynz]

i can only tell you want i do.

you should ask your own eyes. that's what matter.

nnedi3 on chroma is totally useless in my eyes.


Could you help me with the other question ? if we are on the subject on NNEDI3 , I just don't know if I need to enable SuperRes on both Chroma Upscaling and refinement.


Yeah anything over nnedi3 32 neurons is wasted performance AFAIC. It's a nice sharp chroma upscaler but it most situations you won't notice a difference.

Jinc AR and Reconstruction soft are my fave pics with SuperRES being a luxury if you can afford it.

Luma should be where you spend the majority of your available performance for the simple fact that those changes are a lot more tangible.

[Asmodian]

I am also a bit perplexed about the whole Bicubic thing. Are you saying that all the Image Downscaling options (and perhaps others?) are just renamed version of Bicubic? Catmull-Rom = Bicubic50, SSIM = Bicubic125, Jinc/NNEDI3/others = some other Bicubic???

Only Catmull-Rom and Bicubic50 are equivalent. Every other scaling option is unique.

Catmull-Rom is a special call out for Bicubic50 because it is the sharp theoretically optimal option for bicubic. Similar theoretical sweet spots for bicubic are SoftCubic100 and Mitchell-Netravali, Mitchell-Netravali is bicubic with B=1/3 and C=1/3. In madVR b = softness and c = sharpness. The proposal (https://de.wikipedia.org/wiki/Mitchell-Netravali-Filter) was any set of values where B + 2 * C = 1, with 1/3 and 1/3 as a particularly good balance between artifacts and sharpness. I believe optimal meant minimizing aliasing and ringing with varying sharpness. However, this work didn't have madVR's, or any, anti-ringing filter.

[leandronb]

I have the same problem if I use MPC-BE with internal filters.
But if I use MPC-BE with LAV filters the problem goes away.

Just tested and it's the same problem also with the latest madVR version.

I am using LAV filters, ac3filter and madvr.
Updated madvr and mpc and after 3 episodes no problem, i will try with other episodes later.

[adhara]

Question:
Is CatmullRom still the best algorythm for downscalling stuff ?
Or should I move to Jinc for this ?

I'm outputting source video to 2160p with a 970GTX video card.

[ryrynz]

Question:
Is CatmullRom still the best algorythm for downscalling stuff ?


There is no best, only what you prefer. Jinc AR is good, so is SSIM 1D & 2D and many of the sharper bicubic scalers (100-150).
Many have updated their downscaling preferences to those, depending on the content.

[AngelGraves13]

madshi, will you be extending the GPU queue to maybe 32 or 64 eventually? Same question for frames to present in advance.

I've noticed higher values produce smoother motion with smooth motion turned on. I'm guessing it has to do with how many frames are blended together.

[nevcairiel]

madshi, will you be extending the GPU queue to maybe 32 or 64 eventually? Same question for frames to present in advance.

I've noticed higher values produce smoother motion with smooth motion turned on. I'm guessing it has to do with how many frames are blended together.

Present in advance can't be increased, thats a limit of the Direct3D API.
And personally I seriously doubt that increasing the queues increases the smoothness, as long as the value is high enough for the algorithm to work. It won't blend infinitely many frames, just the exactly right amount.

[James Freeman]

Fellows I have to express my findings.
Super-xbr 150 does wonders to Chroma, it makes the very lossy 4:2:0 appear as though it is 4:4:4.
Reconstruction (sharp/placebo) does something very similar but a lot harder on the gpu than super-xbr.

[hdboy]

I'm an avid madVR user but I was slightly bemused with all the effort to support 3D, now that it appears that Samsung and LG are about to pull the plug was it and is it really worth the effort for what appears to be a dead format in its current guise ...
I'm think LG may keep it alive a bit longer. Their passive 3D on OLED is getting rave reviews for being the best 3D ever.

[James Freeman]

Yeah, it's quite ironic that Nev and Madshi implemented support for 3D right at the end of 3D life cycle in the TV industry (unexpectedly I assume).
Nevertheless it's awesome for whom loves full resolution frame-packed 3D and madVR together.

[nevcairiel]

I'm think LG may keep it alive a bit longer. Their passive 3D on OLED is getting rave reviews for being the best 3D ever.

Unfortunately their OLEDs cost an arm and a leg. Wonder if their 2016 S-UHD screens are any cheaper and still good.

[FDisk80]

Yeah, it's quite ironic that Nev and Madshi implemented support for 3D right at the end of 3D life cycle in the TV industry (unexpectedly I assume).
Nevertheless it's awesome for whom loves full resolution frame-packed 3D and madVR together.

I'm calling it now. VR Movies! It's 3D and surround.
Just think about it. You can watch the same movie over and over and always find something new that you didn't see before because you just didn't turn your head to look at it before.

Nev and Madshi get to work :D

[James Freeman]

I'm calling it now. VR Movies! It's 3D and surround.
Just think about it. You can watch the same movie over and over and always find something new that you didn't see before because you just didn't turn your head to look at it before.

Try NNEDI3 256 on that... :D

[FDisk80]

Try NNEDI3 256 on that... :D

Ha! (http://new2.fjcdn.com/thumbnails/comments/5528430+_0b4695e6a754cee126803a8de1301b9c.gif)

[thewildsun]

I have the same problem if I use MPC-BE with internal filters.
But if I use MPC-BE with LAV filters the problem goes away.

Just tested and it's the same problem also with the latest madVR version.

I have the same issue in PotPlayer and Windows 10. It doesn't make a difference if I use the latest LAV nightlies or PotPlayer's internal decoders. The only way I've been able to work around this issue is to uncheck D3D11 in madvr. Has anyone found a way to avoid the issue while keeping D3D11?

[CarlosCaco]

Fellows I have to express my findings.
Super-xbr 150 does wonders to Chroma, it makes the very lossy 4:2:0 appear as though it is 4:4:4.
Reconstruction (sharp/placebo) does something very similar but a lot harder on the gpu than super-xbr.

What a coincidence, I was messing with the configuration and tried the placebo and sharp, was very very clear the diference from soft, só i tried xbr on 150 because performance hit and achieve great results, so i came to the fórum and reads your post, have the same finding! Great!

By the Way madshi thanks for the great improvements on madvr

Ssim 2d is detailed clean sharp a great addition to the renderer
Great job!

B

[JarrettH]

Doesn't reconstruction sharp and placebo create a green halo for you guys? It's not subtle. Soft is the only choice for me. :cool:

[CarlosCaco]

Doesn't reconstruction sharp and placebo create a green halo for you guys? It's not subtle. Soft is the only choice for me. :cool:

I don't see any green halo...

Madshi there is a way to reduce the perfomance hit on reconstrution placebo and sharp?

[hu1kamania]

Doesn't reconstruction sharp and placebo create a green halo for you guys? It's not subtle. Soft is the only choice for me. :cool:

I definitely see ringing with reconstruction and sxbr.

I find nnedi3 to be the sharpest, and it's comparable to the performance hit taken by reconstruction sharp AR and placebo AR.

[JarrettH]

And by halo, I mean ringing. When I was watching an episode of Family Guy it was quite apparent. Basically any cartoons. The ringing happened to look like an opposing colour

[Uoppi]

And by halo, I mean ringing. When I was watching an episode of Family Guy it was quite apparent. Basically any cartoons. The ringing happened to look like an opposing colour

So if NNEDI3 is too taxing, would you say Jinc AR is the next best option for chroma?

Currently, I use reconstruction soft (previously I used Jinc AR) for chroma when NNEDI3 can't be used. I'm seeing ringing in some cartoons too but not sure if it's just the shitty SD source.

[Manni]

Fellows I have to express my findings.
Super-xbr 150 does wonders to Chroma, it makes the very lossy 4:2:0 appear as though it is 4:4:4.
Reconstruction (sharp/placebo) does something very similar but a lot harder on the gpu than super-xbr.

Are you saying Super-XBR 150 gives a better result than NNEDI3 32? Is this with bluray?

[James Freeman]

Are you saying Super-XBR 150 gives a better result than NNEDI3 32? Is this with bluray?

I painted a test pattern with thin and thick lines of blue and red circles in 4:4:4.
Then video captured the screen with CamStudio (screen recording software).
Then compressed the RGB (4:4:4) video through x264 to YCbCr 4:2:0 lossless.

I simply compared the original 4:4:4 video with the 4:2:0 video and chose a chrome upscaler which looks closest to the original 4:4:4 video.

To my eyes Super-XBR 150 almost recreates the thin chroma data without touching much the thick circles.

[ryrynz]

Are you saying Super-XBR 150 gives a better result than NNEDI3 32? Is this with bluray?

All neuron counts with NNEDI3 for chroma look basically identical. It's pretty much a waste using more than 16 neurons on it for chroma IMO as
I Rolled a 16 neurons screenshot and a 256 neuron screenshot back and forward and I can't tell the difference viewing it 1 foot away.

I've noticed that when anti-ringing is enabled with super-xbr and to a lesser degree Jinc (both with AR enabled and not along with some other scalers) can bring green tinges and possibly other colors if the source white contains some inside, which isn't ideal. This is only noticeable upon close inspection however.

The below screenshot comparison shows how bad this can be but you'd be hard pressed to notice it while watching.

super-xbr 150 vs super-xbr 150 AR (http://screenshotcomparison.com/comparison/162790)

Also for comparison NNEDi3-16 taps vs Reconstruction sharp (http://screenshotcomparison.com/comparison/162795)

I'm actually really liking super-xbr 150 for chroma... I wonder if it'd be possible to make a combination of it and reconstruction?
So it looks like I've got a new preference for chroma..

James can you share your videos? Thanks for testing it BTW, I wouldn't have considered it useful for chroma if you hadn't mentioned it.

[Manni]

I painted a test pattern with thin and thick lines of blue and red circles in 4:4:4.
Then video captured the screen with CamStudio (screen recording software).
Then compressed the RGB (4:4:4) video through x264 to YCbCr 4:2:0 lossless.

I simply compared the original 4:4:4 video with the 4:2:0 video and chose a chrome upscaler which looks closest to the original 4:4:4 video.

To my eyes Super-XBR 150 almost recreates the thin chroma data without touching much the thick circles.

Thanks, I'll give it a try.

Do you notice an improvement compared to NNEDI3 with real content, especially bluray chroma upscaling?

[l0rdraiden]

I'm using madVr and I have like 100 presentations glitches or more per second (no dropped frames)
Where is the issue?
I have tried everything, I have used the SVP4 installer, the KCP installer, I have installed and configure everything manually, I have change all the imaginable settings in madvr, I have tried with MPC BE/HC, I have spend hours in looking for a solution in google and still i have tons of presentation glitches despite I can not notice any degradation in the video.

I'm on win10 x64 i54670k and a raedon HD 7970 with latest version of the video drivers Crimson hotfix 1.1 (same issue with the stable version). I didn't have this issue on win 8.1 as far as I can remember.

Please can someone help me to troubleshot this problem?

Thanks in advance.

[QBhd]

I painted a test pattern with thin and thick lines of blue and red circles in 4:4:4.
Then video captured the screen with CamStudio (screen recording software).
Then compressed the RGB (4:4:4) video through x264 to YCbCr 4:2:0 lossless.

I simply compared the original 4:4:4 video with the 4:2:0 video and chose a chrome upscaler which looks closest to the original 4:4:4 video.

To my eyes Super-XBR 150 almost recreates the thin chroma data without touching much the thick circles.

Is this S-XBR 150 AR for chroma? or just plain S-XBR 150 for chroma? Is ask this, because on ryrynz's screenshot comparisons AR looks a little nasty.

QB

[chros]

I've noticed that when anti-ringing is enabled with super-xbr and to a lesser degree Jinc (both with AR enabled and not along with some other scalers) can bring green tinges and possibly other colors if the source white contains some inside, which isn't ideal. This is only noticeable upon close inspection however.
Interesting, thanks. Can you make 1 more test with the same image using Jinc/AR? Thanks

[Manni]

All neuron counts with NNEDI3 for chroma look basically identical. It's pretty much a waste using more than 16 neurons on it for chroma IMO as
I Rolled a 16 neurons screenshot and a 256 neuron screenshot back and forward and I can't tell the difference viewing it 1 foot away.

I've noticed that when anti-ringing is enabled with super-xbr and to a lesser degree Jinc (both with AR enabled and not along with some other scalers) can bring green tinges and possibly other colors if the source white contains some inside, which isn't ideal. This is only noticeable upon close inspection however.

The below screenshot comparison shows how bad this can be but you'd be hard pressed to notice it while watching.

super-xbr 150 vs super-xbr 150 AR (http://screenshotcomparison.com/comparison/162790)

Also for comparison NNEDi3-16 taps vs Reconstruction sharp (http://screenshotcomparison.com/comparison/162795)

I'm actually really liking super-xbr 150 for chroma... I wonder if it'd be possible to make a combination of it and reconstruction?
So it looks like I've got a new preference for chroma..

James can you share your videos? Thanks for testing it BTW, I wouldn't have considered it useful for chroma if you hadn't mentioned it.

Thanks, I use NNEDI3 32 because I have the power to spare without raising fan noise. At 64, it raises the fan noise significantly for no visible improvements.

I kept 32 because I read a while ago that only 16 could be worse. I'm so far from maxing the CPU/GPU with 32 that really it doesn't make any difference to keep 32.

In any case, I'm going to try Super XBR 150 for chroma. No AR I guess given the artifacts with AR.

[huhn]

this shows the color boosting from reconstruction and the ringing i reported some weeks ago.

super XBR 150 can go wrong. super XBR 75was the only save choice i found.

[tObber166]

Hmm, something's wrong with DXVA Upscale colors

DXVA vs Spline
http://screenshotcomparison.com/comparison/162815


DXVA vs Jinc
http://screenshotcomparison.com/comparison/162814


It seems to only affect upscale, not DXVA downscale.
Anyone else who experienced this?


Using madvr 0.90.10
Nvidia 361.91

[markanini]

this shows the color boosting from reconstruction and the ringing i reported some weeks ago.

super XBR 150 can go wrong. super XBR 75was the only save choice i found.

How does superXBR75 compare to reconstruction?

[Unr3aL]

I'm using madVr and I have like 100 presentations glitches or more per second (no dropped frames)
Where is the issue?
I have tried everything, I have used the SVP4 installer, the KCP installer, I have installed and configure everything manually, I have change all the imaginable settings in madvr, I have tried with MPC BE/HC, I have spend hours in looking for a solution in google and still i have tons of presentation glitches despite I can not notice any degradation in the video.

I'm on win10 x64 i54670k and a raedon HD 7970 with latest version of the video drivers Crimson hotfix 1.1 (same issue with the stable version). I didn't have this issue on win 8.1 as far as I can remember.

Please can someone help me to troubleshot this problem?

Thanks in advance.
Are you, by any chance, running your screen at 120Hz or something? If yes try to use 60 Hz and see if that works...

MadVR never used to play nice with 120Hz in my case.

Greetz, Unr3aL67

[Werewolfy]

@madshi

First, I'd like to thank you and Shiandow for SSIM downscaling. I've never been happy with downscaling in the past, always found it too blurry and now it's sharp and natural :) I don't agree with some statements that it looks artifical, it definitely doesn't.

I spent a long time playing with your upscaling refinement options but I can't find a correct setting for every contents. I admit that I'm not a fan of sharpening filters, it always sharpens the flaws in the picture (even if your edge enhancement is the best I've seen!) and the only one I can use is the Reality Creation from Sony. The thin edges option is quite good too on DVD anime but I don't want it on movies so it's still not a set and forget setting.

I have a question. I see that there is an option to move the subtitles. Would it be possible to allow the user to change the color and the font too? I know it's easy with .srt files but I can't find a way to do it with DVD and Blu-Ray subtitles. I found that your option to move the subtitles works with Blu-Ray so I immediately thought to ask you if it would be possible. It's not just a cosmetic change, it allows the local dimming on Edge Led not to get very bright when subtitles are displayed. I also have a simulation of HDR on my display called X-tended Dynamic range and it works wonderfully... but not with white bright subtitles. Dark yellow subtitles are really better with my TV.