madVR - high quality video renderer (GPU assisted)

[Qaq]

Originally Posted by nevcairiel

The audio renderer provides the reference clock, so it is indeed quite off-topic.

GPU + video driver provide image upscaling, chroma upsampling, gamma correction etc etc... And lets not forget madVR provided even video decoders for a while.

[ryrynz]

Hey madshi can you explain why chroma is listed twice on the OSD when I have NNEDI active? The chroma is using my luma upscaling resizer, as well as the chroma upscaling resizer, why's that?

[Anatasia]

Originally Posted by tobindac

massive 'green' artifact here on 64bit



though I don't see much use for the version yet since I always use SVP on top which is still 32bit.

GPU is an R9 290 latest stable drivers.

Confirmed this green artifact with NNEDI 3 chroma upscaling and double/quadrouble Chroma resolution; darker screen (look like scanlines) with double/quadruple Luma resolution even with the newest 15.4 beta driver. These issues only occurred with MadVR 64 bit, my VGA card is R9 290X and in Windows 8.1 environment.

Hi Guys,
my intent is to identify the best graphic card to use with MadVR and outputting a 2160p signal to a UHD TV.
I don't know if there is a specific thread about this....

My fisrt question is:

1)If we set the pc to output a RGB 4:4:4 2160p@24hz signal (in the control panel of the grapich card drivers) there is differences in the quality of the signal if we use an AMD card (with HDMI 1.4 I suppose) and a NVIDIA GTX970 (with HDMI 2.0 I suppose)?
In other words may be that the AMD cards do some kind of downsampling?

2)I know that madVR can support displays with panels higher of 10bit. Is It possible to know if a TV is capable to accept and represent a 12bit signal from a HTPC?
There is a method to know If a UHD TV is capable to accept signals @8/10/12bit? MadVR is capable to report this info?

thank you all

About Picture quality:

If I do not use "use full screen exclusive mode" I've some loss of Picture quality?

thank you

[huhn]

Originally Posted by ryrynz

Hey madshi can you explain why chroma is listed twice on the OSD when I have NNEDI active? The chroma is using my luma upscaling resizer, as well as the chroma upscaling resizer, why's that?

first of all madVR always does 4:2:0 -> 4:4:4 first that would be the first chroma resize step and that is was chroma upscaling does and nothing more.

when image doubling is used without chroma the chroma doesn't match the luma anymore and the image upscaler or down scaler is used to fix this.

have a look here: http://forum.doom9.org/showthread.php?t=171787
even advanced user may found something new here.

[huhn]

Originally Posted by Kirk Lazarus

1)If we set the pc to output a RGB 4:4:4 2160p@24hz signal (in the control panel of the grapich card drivers) there is differences in the quality of the signal if we use an AMD card (with HDMI 1.4 I suppose) and a NVIDIA GTX970 (with HDMI 2.0 I suppose)?
In other words may be that the AMD cards do some kind of downsampling?

if your TV support 4:4:4 in the first place you can get 4:4:4 with an AMD 1.4 HDMI card up to 30 HZ.

2)I know that madVR can support displays with panels higher of 10bit. Is It possible to know if a TV is capable to accept and represent a 12bit signal from a HTPC?
There is a method to know If a UHD TV is capable to accept signals @8/10/12bit? MadVR is capable to report this info?

madVR output is 8 bit. it can handle input with higher bit deep.
for 10 bit a fullscreen directx 10 or newer surface is needed. the high quality dither makes up for this for now.

the current version can't do that.
and it's not known what current driver do with a 10 bit input signal but it's know that they can output 1080p in up to 12 bit.

Originally Posted by Kirk Lazarus

About Picture quality:

If I do not use "use full screen exclusive mode" I've some loss of Picture quality?

thank you

if you don't get dropped, repeated or delayed frames everything should be 100% fine with windowed mode.

[ryrynz]

Originally Posted by huhn

first of all madVR always does 4:2:0 -> 4:4:4 first that would be the first chroma resize step and that is was chroma upscaling does and nothing more.

when image doubling is used without chroma the chroma doesn't match the luma anymore and the image upscaler or down scaler is used to fix this.

have a look here: http://forum.doom9.org/showthread.php?t=171787
even advanced user may found something new here.

Is it not possible to do a single resize of the chroma with the chroma upscaler from the get go? Or if it absolutely has to be done in two stages why not use the specified chroma upscaler again??>

Anyone else getting frame drops when using 24Hz but not with 60Hz? I've reverted drivers and madVR versions but no change.

[nevcairiel]

The chroma resizer is only ever used for the 420->444 step, since its probably somewhat specialized for a clean 2x upscale. All other scaling is done with the image resizer.

[ryrynz]

Would it make sense to have a specialized chroma image resizer for this step?

[huhn]

Originally Posted by ryrynz

Would it make sense to have a specialized chroma image resizer for this step?

some filter need RGB and are better used before luma scaling.
how to use nnedi3 chroma on something that isn't x2?
they are specialized in 2x resizing how do you special general resizing?

[huhn]

Originally Posted by ryrynz

Is it not possible to do a single resize of the chroma with the chroma upscaler from the get go?

of cause it is possible to scale directly to the target chroma.

Or if it absolutely has to be done in two stages why not use the specified chroma upscaler again??>

nnedi3 can't be used this way and the image scaler would scale the chroma in this case anyway on top of it usually the better scaler used for luma.

Anyone else getting frame drops when using 24Hz but not with 60Hz? I've reverted drivers and madVR versions but no change.

i'm currently running 23p at 23p and i didn't get dropped frames.
if you are on windows 7 are you sure you haven't broken the desktop composition? happen to me all the time...
can you make a screen?

sorry i kinda repeat my self here from the last post. happens i guess.

[nevcairiel]

Originally Posted by ryrynz

Is it not possible to do a single resize of the chroma with the chroma upscaler from the get go?

IIRC, madVR used to do that in the beginning, but quality and performance reportedly improved when it moved to the current model.

Unless you have an actual problem, just accept how it works internally? madshi probably put some thought into how it would work best.

[Arm3nian]

Originally Posted by huhn

at 60 HZ the effect should be minimal to the eyes. but if you use any kind of frame interpolation in your TV madVR SM is not a good idea. but i wouldn't be shocked if some TV exposes some easily spot able issue with SM.
on an LCD without frame interpolation there should be any problems they have a motion clearness of only 300p anyway you can't get less sharpness then this.

Frame interpolation works fine with SM.

Where do you get that from? Frame interpolation works fine with SM.

it's rare that a screen has flawed 23p/24p but it happens. so this screen maybe has a problem with that mode. some people even say that about some OLED screens

Which further proves my point that SM works for everyone and offers the same benefits. You also forget that not every display can be set to 23/24.

MPC-BE audio renderer can resample audio to match the video clock like reclock?
MPC-HC MPC Audio renderer is broken it's even labeled broken move your mouse over and read: "MPC Audio renderer is broken, do not use" the BE version should be better. a new MPC-HC audio renderer is planned BTW.

MPCBE doesn't need to resample audio to match the video clock. I meant it resamples the sample rate, bit depth, etc, things an audio renderer does. And yes MPCHC audio renderer is broken, that's literally what my post said.

[e-t172]

Originally Posted by nevcairiel

The audio renderer provides the reference clock, so it is indeed quite off-topic.

But in the case of Reclock, it's basically a hack. Reclock has to guess what the refresh rate is and resample the audio based on that. If it guesses wrong (for example it's using the wrong screen), the clock won't be synced to the video, and therefore frame drops/repeats will occur.

It would be cleaner from a DirectShow architecture perspective to have madVR provide the DirectShow clock for Reclock, because as a video renderer, it's madVR's job to know the screen's refresh rate. Therefore, if you want to slave the audio clock to the video clock, then it's better to have the video renderer provide the clock, and then slave the (resampling) audio renderer to that clock.

That said, I'm mostly nitpicking here.

Originally Posted by Arm3nian

MPCBE doesn't need to resample audio to match the video clock.

Assuming Smooth Motion isn't used, and a multiple refresh rate is used, there are two possible scenarios:
- You slave the video clock to the audio clock (default case), in which case you get frame drops/repeats; OR
- You slave the audio clock to the video clock (Reclock case), in which case you need to resample the audio.

You can either get one or the other. There is no free lunch. Unfortunately that's the way it is on computers. Set-top boxes don't have this problem because they have the same hardware clock for both video and audio.

[Arm3nian]

Originally Posted by e-t172

Assuming Smooth Motion isn't used, and a multiple refresh rate is used, there are two possible scenarios:
- You slave the video clock to the audio clock (default case), in which case you get frame drops/repeats; OR
- You slave the audio clock to the video clock (Reclock case), in which case you need to resample the audio.

You can either get one or the other. There is no free lunch. Unfortunately that's the way it is on computers. Set-top boxes don't have this problem because they have the same hardware clock for both video and audio.

Why would you need reclock if a multiple refresh rate is used? Reclock is used when you don't have a multiple refresh rate, but can get close. If your screen refresh rate can't come close to the source material then reclock is useless. In other words, use SM.

Frame repeats are also highly exaggerated.

[e-t172]

Originally Posted by Arm3nian

Why would you need reclock if a multiple refresh rate is used? Reclock is used when you don't have a multiple refresh rate, but can get close.

Of course, that's what I meant. As I said, it's impossible to get an exact multiple anyway, because that would mean using the same clock, which is not possible on a PC. I guess I failed to mention the case where you can get so close that it doesn't really matter, but you need to do a lot of tinkering and trial&error to get to that point.

[huhn]

Originally Posted by Arm3nian

Frame interpolation works fine with SM.

how to frame interoplate a blended frame? of cause you can use it with SVP but for a simple reason. SVP runs before SM.

Which further proves my point that SM works for everyone and offers the same benefits. You also forget that not every display can be set to 23/24.

i know and i use it on my gaming PC. but plasma user report visible ghosting problems. and i'm still very possive that blended frames create problems for TV frame interpolation.

MPCBE doesn't need to resample audio to match the video clock. I meant it resamples the sample rate, bit depth, etc, things an audio renderer does. And yes MPCHC audio renderer is broken, that's literally what my post said.

and how do you do 25 hz -> 24 hz ? reclock is simple not useless. i told other "you could use SM at 60 HZ and not reclock" but if people have reason to use reclock or problems with SM i simple have nothing to add there.
i'm 99% sure i can find the SM using display in a blind test and that in a short time.

[flashmozzg]

Where does madVR save freeze reports? I can't find them, trying to make one because it freezed again. Found them on my desktop... Silly me =)

Originally Posted by madshi

Can you guys please create a freeze report (Ctrl+Alt+Shift+Break) and upload it somewhere (not here) and link here, so that I can look at it?

https://mega.co.nz/#!JRlGBaKL!G1W2ME...AjLJ-ltYSb6yWQ

Originally Posted by huhn

@XMonarchY

64 bit has better decoding performance. so it can matter a lot for software decoding.

in term of madVR performance there shouldn't be any real differences.

and about your issue we need way more information that just "if i got in FSE mode i get dropped frames"

resolution? what settings what are the renderer times, FPS, and way more questions.

Obviously 64bit has better decoding, but using ReClock is quite important for those who want perfect sound+video sync.., which is only possible with ReClock, which is and always will be x86 (32bit)...

I use the latest MPC-HC 1.7.8 x86, ReClock 1.8.8.4 x86, madVR 0.87.21 (reset older version to default and properly uninstalled, then installed new version and reset to defaults) and the latest LAV Filters 0.64. My LAV Video filter uses nVidia CUVID (for all 9 boxes underneath), Enabled Adaptive H/W Interlacing for 25/30p Films with High Quality Processing. The rest is default.

I use high quality settings:
- Rec.709 3DLUT (ArgyllCMS 1.7.0 non-buggy Beta dev. release + dispcalGUI 2.9.0.7 Beta)
- Reduce Banding Artifacts = enabled for lower quality videos
- Chroma Upscaling = NNEDI3 64 neurons
- Image Doubling and Quadrupling (both Luma and Chroma) = "Always, If Upscaling Needed" = 32 neurons for each of 4 boxes
- Image Upscaling = Jinc 3 tap + "Activate Anti-Ringing Filter"
- Image Downsaling = Catmull-Rom with "Activate Anti-Ringing Filter" & "Scaling in Linear Light"
- Smooth Motion = Off
- Dithering = Error Diffusion - Option 1, Use Colored Noise = NOT selected/ticked, Change Dither for Every Frame = NOT selected/ticked
- Trade Quality for Performance = all boxes are UN-ticked
- The rest = default (aside from FSE)

Framerate drops occur in H.264 .MKV films recorded @ 1080p with high quality. My rig specs are in my sig (3770K @ 4.8Ghz, GTX 980 4GB, 1080p monitor, SSD, Windows 8.1).

Of course I would get rid of framerate drop by reducing NNEDI3 from 64 neurons to 32 neurons, but that is not a solution because when FSE is not enabled, no frames are dropped with NNEDI3 set to 64 neurons.

Its no big deal, but I am just curious. I figured FSE would actually be better as far frame dropping goes...