madVR - high quality video renderer (GPU assisted)

[James Freeman]

I think the new "display peak luminance" feature only works if your display has a static and unchangable minimum light level (MLL) of the display.
Meaning, it only truly works if you have a FALD (full-array local dimming) display.
If your display can do 4,000nits but keeps the contrast ratio at 1000:1 it simply does not differ from any other SDR TV.

OR, I might be very wrong!
..and madVR actually compress the HDR content to 1000:1 and you should turn your SDR display brightness to max (400nit) to utilize the new feature...?

I need some HDR demo clips, anyone?

[Plutotype]

What about Dolby Vision support in madVR? Is it something technically impossible to implement?
Thanks

[madshi]

Originally Posted by nevcairiel

Note however that not all UHD Blu-rays will use HDR, in fact most will not.

I'm not sure about that. I've some content that is mastered at 1200 nits with DCI-P3. I could imagine that something like this might be suitable for any old and new movies. So it's quite possible that all (or most) movies might get HDR transfers. But I don't really know, we'll have to wait and see...

Originally Posted by fairchild

How do we know what to set the display peak luminance to?

FWIW, there's no exact science for how these HDR -> SDR conversions work. So the numbers in the settings dialog are rough estimates. Things also depend on the ambient light level in the room and how your display is calibrated. E.g. with a front projection setup, although the luminance is much lower than e.g. 600 nits, you might still get the best image quality by setting your display to 600 or 900 nits. So my recommendation is to simply play some HDR content and let you eyes decide which nits setting in madVR looks best in your setup.

Originally Posted by James Freeman

How can you really know in what range the studio mastered their movie? Seen where?

Install a nightly LAV build, then press Ctrl+J and the mastered nits level is displayed.

Originally Posted by James Freeman

Judging by what I see with madTestPattern, how did you know what region is more important to compress and what to clip?
With 1200nit mastered content (presumably), the whole range is compressed.
With 4000nit, again practically the whole range is compressed.
Doesn't that mean that the movie will be too dark overall just so that we will be able to see the super bright parts?

As mentioned before, there's not an exact science to this. The content has more range than any display, so there's way around compressing it somehow somewhere. How much to compress and maybe to clip depends on the video encoding luminance and the display peak luminance. I've selected values for each combination which with real world material looked "good" to me.

Originally Posted by TheElix

I have switched to Random Dithering and it seems to fix crashes. But at around the same time when it crashed previously now the video hangs up. It appears like the video goes in an endless loop of several frames while the audio keeps playing.

Oh well. Have you tried installing a different GPU driver version? Do you have any funny GPU related software installed/running? If so, try disabling it.

Originally Posted by TheElix

Here's the log with the debug OSD: https://yadi.sk/d/QDP2CRqHmDvVi Maybe you can tell why it doesn't go into FSE?

Will look later when I'm back at my development PC.

Originally Posted by TheElix

I've downloaded several test videos and tried the new HDR feature. I am currently using a professional NEC P241W monitor which is true 10 bit through DisplayPort cable and is capable of >Adobe RGB coverage and has a preset for DCI color space. I've set the maximum brightness which is 350 cd/m2 for this display and set the white point to 6500K. I used 400 cd/m2 in madVR settings. Well, it works. But I have horrible amount of dropped frames even though rendering is well under 10 ms. It looks like I cannot truly appreciate the difference now since I'm not getting 10 bit without FSE.

Which queue is the first one empty? Probably the decoder queue? 4K HEVC decoding is really demanding!

Originally Posted by TheElix

Also, you said that we may choose the maximum luminance between 0-10000 cd/m2, but the lowest I can choose now is 400 cd/m2. Is it intentional?

No, 0 - 10000 is the range each pixel can have, not the range you can choose in the settings. The settings you can choose is which peak luminance your display has. If your display has a peak luminance of 0 nits then its broken because it doesn't display anything at all.

Originally Posted by James Freeman

I think the new "display peak luminance" feature only works if your display has a static and unchangable minimum light level (MLL) of the display.
Meaning, it only truly works if you have a FALD (full-array local dimming) display.
If your display can do 4,000nits but keeps the contrast ratio at 1000:1 it simply does not differ from any other SDR TV.

OR, I might be very wrong!
..and madVR actually compress the HDR content to 1000:1 and you should turn your SDR display brightness to max (400nit) to utilize the new feature...?

Huh? No. Contrast got nothing to do with it. If you have a perfect black level, contrast automatically calculates to infinite, regardless of peak white luminance. And MLL got nothing to do with it, either.

Originally Posted by Plutotype

What about Dolby Vision support in madVR? Is it something technically impossible to implement?

I neither have any such content nor any technical spec. Also there are no decoders available for that. But my best guess is that it might be compatible to "conventional" HDR. I don't really know, though.

[James Freeman]

Originally Posted by madshi

Huh? No. Contrast got nothing to do with it. If you have a perfect black level, contrast automatically calculates to infinite, regardless of peak white luminance. And MLL got nothing to do with it, either.

Then what is the difference between 1,200 and 400 nit with 1000:1 display?
So this "peak brightness" selector only works if you have perfect blacks (stable and unchangable MLL where only the peak brightness changes, aka FALD or OLED display)??

[madshi]

I'm not sure why you keep trying to limit the "peak brightness selector" to specific conditions under which it might work or might not work. There is no such limit. And as I said, the display contrast got nothing to do with it.

HDR content that is mastered on e.g. a 4000 nits display should ideally be displayed on a 4000 nits display. If your display cannot do 4000 nits then madVR has to compress the highlights, otherwise the image would be much too dim overall. The difference between 1200 nits and 400 nits in the madVR "peak luminance" option is how much compression madVR applies to the luminance channel. The dimmer your display is the more compression madVR has to apply to make the image bright enough overall.

Why don't you simply download some HDR content and try different settings for that madVR option and see which works best for you? I think for most people with today's display technology, the default setting of 600 nits should work well. If the image is too dim/dull for your taste, try 400 nits. If the image is too bright, try 900 nits or higher.

[James Freeman]

I understand that it compresses the range, what I don't understand is whether you assume that the display has infinitely low black leve and the only thing that changes is the peak brightness level?

The only free content that is available now is a couple of seconds of Life of Pi and Exodus right?

[madshi]

We're talking about digital processing. madVR outputs video content with a black level value of 0 (or 16 when using TV levels). Whether or not the display has infinitely low black levels or not isn't something madVR has to worry about.

Basically what madVR does is modifying the "gamma curve" (in words that might make more sense to you), to account for the difference in peak luminance capabilities between the mastering display and your display.

[James Freeman]

Got it, I'm kinda' slow today.
Sorry about that.

BTW, it would by nice to bypass this feature for reference, or just to see how the image mapped across standard power curve gamma.

[nevcairiel]

Originally Posted by James Freeman

BTW, it would by nice to bypass this feature for reference, or just to see how the image mapped across standard power curve gamma.

Just switch to EVR or something. It looks terrible, totally lifeless and dull.

[har3inger]

Originally Posted by mogli

Yes, before the current version repeated frames were reported and visual stutter occured at that very moment. Now there's only the latter.

Note that this only seems to happen when fps is almost the same as the display rate. It took me some time to remember that the interlaced NTSC DVDs (actual video, not film) will have there frames doubled by DXVA deinterlacing, therefore running 59 fps at 60 Hz. This seems to be hard for FRC. (No way around starting to use ReClock, it seems?)

madVR never did report such repeated frames, at least not for me. When viewing a 24 fps movie at 60 Hz with FRC on I never got even one repeated frame reported in any version. Or do you want to say, this was by accident, because each and every frame was unique or blended, so that there never was any actual repeated frame?

Just as a test, try lowering the GPU load as much as possible (bilinear scalers, check on all the trade perf/quality options) and see if you still have stutter. Back on my old laptop, I experienced something similar, where the OSD wouldn't report any problems or glitches, but my playback would stutter and repeat frames nonstop with smooth motion on. This happens when I'm just riding the edge of the GPU's capability without dropping frames in the OSD. It's still the case with my new computer if I push things too hard.

When you're playing FRC 59 hz at 60 hz display, smooth motion has to blend pretty much every single 60 hz frame, which adds substantially more load compared to 24hz at 60 hz. Yeah, it would be much harder, plus should look much worse with so much frame ghosting. You may as well leave it off or use reclock for that. FRC works best when the display rate is at least double the source framerate.

[mogli]

Originally Posted by madshi

What fill state do the queues have when that stuttering occurs? How many frames are you presenting in advance? Or did you disable the "present several frames in advance" option? FWIW, it's recommended to use that option because it increases presentation reliability.

8 frames in advance; what happens is that the second number of the present queue in the OSD shortly goes up from 7 to 8 when the repeated frame occurs (both with FRC on or off), the other numbers being stable;

[madshi]

Hmmm... Windowed or FSE mode? Try FSE, it's usually more reliable.

[TheElix]

Originally Posted by madshi

Oh well. Have you tried installing a different GPU driver version? Do you have any funny GPU related software installed/running? If so, try disabling it.

I'm on latest Crimson 15.11.1 version, but will try to downgrade as much as possible.

Originally Posted by madshi

Which queue is the first one empty? Probably the decoder queue? 4K HEVC decoding is really demanding!

The present queue is the first one. Here's the picture: https://yadi.sk/i/a1t0Pc0NmE45r

[madshi]

The decoder queue is almost empty, that's the bottleneck. Your decoder is too slow.

[Thunderbolt8]

is it possible to take screenshots which contain subtitle lines with madvr? from what I can see they arent saved with the picture.

[madshi]

Originally Posted by TheElix

Here's the log with the debug OSD: https://yadi.sk/d/QDP2CRqHmDvVi Maybe you can tell why it doesn't go into FSE?

Yes. The debug log reads:

D3D11 fullscreen windowed (8 bit)
covered by some windows
madVR window [madVR] {0,0,1920,1200}
covered by StrokesPlus.exe window [STROKESPLUS] {0,0,1920,1200}

So basically StrokesPlus.exe has a fullscreen window which covers the madVR rendering window. That's the reason why madVR can't go FSE.

Originally Posted by Thunderbolt8

is it possible to take screenshots which contain subtitle lines with madvr? from what I can see they arent saved with the picture.

I think there was a way, but I don't remember. Maybe someone else can help out? FWIW, screenshotting will be revisited/improved in some future version, but not very soon.

[sneaker_ger]

Originally Posted by Thunderbolt8

is it possible to take screenshots which contain subtitle lines with madvr? from what I can see they arent saved with the picture.

http://forum.doom9.org/showpost.php?...ostcount=19673

[Aktan]

Originally Posted by madshi

madVR only drops frames if either the GPU is too slow or if the refresh rate it too low. Please make a screenshot of the Ctrl+J madVR menu when you get those frame drops, then maybe we can help.

Generally you should get better quality when letting madVR deinterlace. So it might make sense to try that again and post a screenshot of Ctrl+J with that setup, too.

I think it's more like the madVR isn't requesting the frames from the capture source. Here is the screenshot of Ctrl+J in Potplayer using the capture source: http://i.imgur.com/8KwExdw.jpg and here is the screenshot of Ctrl+J in Potplayer using a captured AVI from the same capture source: http://i.imgur.com/7anTyFl.jpg. Here are a few things I've noticed:

-Graphedit using madVR as a render shows a black screen only when directly connected to capture source. This doesn't happen playing back a file.
-Both Potsplayer and MPC-HC show the queue being pretty much empty all the time when playing a capture source.
-You are right, the deinterlacing from madVR works great (no need for ffdshow deinterlacing) ONLY if I playback an AVI of the capture source. I think this is mostly due to the queue being empty though...
-Using Graphedit and VM9 and ffdshow to set interlace flag to force DXVA deinterlacing, everything works perfectly with capture source.

I'm pretty sure it's not my GPU as I have 1080i clips that play fine with deinterlacing on. Also the fact it works fine with an AVI of the same capture source shows it's not maxing the GPU at all. I think it is a hint that madVR shows black for Graphedit with capture source but not files.

Now I do realize having a huge decoding queue would introduce input lag for gaming like in this example, but if I were to watch just analog TV (which actually doesn't exist anymore) or just say a VCR, I would still have this problem. I could also just lower the queue if I do live game.

[dansrfe]

Is it possible for madVR to adjust display backlight over DDC/DI based on profile rules for HDR/SDR media? That way, like calibration curves, backlight can be adjusted automatically as well.

[rack04]

I have exhibited random crashes in the past couple of weeks. I have updated madVR, LAV Filters, MPC-BE and graphic drivers to the latest versions but I can't narrow down the culprit. I'm hoping this log provide some kind of clarity. It is a large file so please forgive me.

http://www.filedropper.com/madvr-log