madVR - high quality video renderer (GPU assisted)

[Ver Greeneyes]

It's like 1803 no longer allows MadVR to use the system at full power.

Yeah, that's what I noticed too. I'm not sure if it's related to GPU usage or GPU memory use, but any complex algorithms including super-xbr cause my queues to stop filling up despite the processing times being very low. That's on a GTX 1080, so hardly a weak GPU.

[huhn]

I'm not sure I can, it's the only one available (https://i.imgur.com/sTTicQl.gif).

does the refreshrate change in FSE? it is clearly running at 59p

Alright, here's after default reset (https://i.imgur.com/HY6KUei.jpg). Don't mind those few drops/repeats shown, they only happened when opening the file and switching to fullscreen, they don't happen during playback.

can you now add NGU sharp for the scaler don't touch anything else.

Yes, turning it on/off makes no difference.
sorry i forgot AMD doesn't support it.

That's the thing. It shouldn't even be 6 ms. Before 1803 and now in FSE, this file plays at around 10-11 ms, which is when everything works right with no drops.
It's like 1803 no longer allows MadVR to use the system at full power.
the ms increase that's more than strange. you are not using 2 screen by any chance?

@Ver Greeneyes

don't use optimal power saving setting in the nvidia control center and the problem should disappear which is sadly the default setting.

[Msarc]

does the refreshrate change in FSE?
No.

can you now add NGU sharp for the scaler don't touch anything else.
All default with NGU Sharp Med (https://i.imgur.com/vnbygWU.jpg)
All default with NGU Sharp High (https://i.imgur.com/5CREf7A.jpg)

you are not using 2 screen by any chance?
No, just one.

[huhn]

that's messed up i try may r9 270 later.

edit: and with later i mean days i guess at week end.

[Ver Greeneyes]

@Ver Greeneyes

don't use optimal power saving setting in the nvidia control center and the problem should disappear which is sadly the default setting.
Nah, I've had it set to Adaptive for MPC-HC for ages. This started after the update to 1803.

[Siso]

Nah, I've had it set to Adaptive for MPC-HC for ages. This started after the update to 1803.

Adaptive make rendering spikes, if you want no spikes, use prefer maximum performance.

[XTrojan]

Anyone knows what gives the best details between software decoding and DXVA2 Copyback decoding on LAV filters?

Decoding HEVC ~ 80GB files puts my 8700k CPU at load, even though it's clocked at 5GHz

[huhn]

only dxva2 native has lower image quality and that not a decoding issue so you what ever works best for you.

[XTrojan]

only dxva2 native has lower image quality and that not a decoding issue so you what ever works best for you.

Obviously DXVA2 Copyback is superior than software if image quality is equal between them, the question is if it is, I heard some people getting more artifacts with DXVA2 copy-back

[HelmedHorror]

I'm switching from MPC-HC to MPC-BE, but MPC-BE doesn't seem to want to cut out the black bars the way MPC-HC does. I set up the options between the two to be as identical as possible, and of course madvr's settings are the same between the two.

Is it a known issue that MPC-BE doesn't cooperate well with madvr's black bar features? I did find this old discussion (https://forum.doom9.org/showpost.php?p=1741608&postcount=3325) between madshi and someone else on the MPC-BE thread regarding madvr black bar support, but it's kind of over my head in the technical details. Also it's old.

Wasn't sure if I should ask this here or in the MPC-BE thread...

[chros]

It's working here with BE using the latest beta (always worked).

[huhn]

Obviously DXVA2 Copyback is superior than software if image quality is equal between them, the question is if it is, I heard some people getting more artifacts with DXVA2 copy-back

they are bit identical for codecs like h264 and newer.
i still don't see dxva2 as obviously superior but let'S leave it at this.

[ajp_anton]

How does the smooth motion on/off logic work?

"only if there would be motion judder without it..." - Smooth motion is turned off on VFR video, how can madVR think there won't be judder?
"... or if the display refresh rate is an exact multiple of the movie frame rate" - Isn't this a good example of when SM is *not* needed?

[XTrojan]

they are bit identical for codecs like h264 and newer.
i still don't see dxva2 as obviously superior but let'S leave it at this.

I meant that it's superior since it doesn't consume massive amounts of CPU when decoding large HEVC files. This is assuming MadVR doesn't hog 99% of the GPU already, but I doubt it requires more than 5% GPU power on my 1080 TI with DXVA2 Copy-back.

[Ver Greeneyes]

IIRC DXVA is done on a separate chip (or a dedicated section of logic on the main chip) so it shouldn't use the rest of the GPU (except for copying data back and forth).

[Clammerz]

When madVR can't connect to the internet (ethernet working, internet offline), it takes several seconds longer to start up and also produces ~6 seconds of black before the video shows properly.
Are you sure that's madVR causing the delay?
It seems like it may be either Windows SmartScreen checking, or your media player (perhaps version or other telemetry checking).
The last time I checked madVR doesn't attempt any network connection unless you have the LAN option set (which isn't impacted by internet connectivity).

[brazen1]

Isn't madVR using a network connection to send crash reports which is the telemetry of your system?

[nevcairiel]

Isn't madVR using a network connection to send crash reports which is the telemetry of your system?

Only if it crashes and only if you actually press the Send button. This would not interact with playback at all.

[brazen1]

That's good to know. Thank you nevcairiel.

[e-t172]

How does the smooth motion on/off logic work?

"only if there would be motion judder without it..." - Smooth motion is turned off on VFR video, how can madVR think there won't be judder?
"... or if the display refresh rate is an exact multiple of the movie frame rate" - Isn't this a good example of when SM is *not* needed?

The madVR options thread (https://forum.doom9.org/showthread.php?p=1709764#post1709764) should have the answers to your questions.

[huhn]

it is doing 4:2:0 -> 4:2:2 -> 4:4:4 with "simple" scaler.
why this is the case well i guess you have to ask madshi.
doing it in 1 shader should be faster if i remember it correctly Shiandow was testing this.

[Asmodian]

"only if there would be motion judder without it..." - Smooth motion is turned off on VFR video, how can madVR think there won't be judder?

madVR makes this decision at startup and will use whatever value for the frame rate it is told by LAV, it doesn't analyze timestamps.

[Weirdo]

You have all Windows 10 focus assist (https://www.windowscentral.com/how-use-focus-assist-reduce-distractions-windows-10-april-2018-update) features disabled? It fix "frozen video" for me.
You were right, thanks. It's strange because Focus Assist was supposedly disabled, but problem went away only after unticking all options.

[Asmodian]

Yes, focus assist seems oddly behaved. It needs to be completely disabled or it damages performance even if it is not supposed to be doing anything. :confused:

[Stereodude]

Can we get the ability to go negative (below 0) in custom video levels in madVR? For example, being able to go to -16 would fix incorrectly encoded black levels on a laptop or PC system that maps 0 to black and 255 to white.

[huhn]

control+alt+shift+i

maybe there is a tag name too

[Stereodude]

control+alt+shift+i

maybe there is a tag name too
That affects both ends though right? From my quick test it seems to. I've seen content where the peak white levels are correct, only the blacks are mis-mapped. Hence the request.

[Clammerz]

Can we get the ability to go negative (below 0) in custom video levels in madVR? For example, being able to go to -16 would fix incorrectly encoded black levels on a laptop or PC system that maps 0 to black and 255 to white.

Have you tried:
* added tag "blacklevel=%value%", value range [-50, +50]

If you CTRL+F for "tag" in the changelog you'll see others that may be of use.

[oldpainlesskodi]

Hi fellows,

I'm sure I am not the only one following progress over on the AVS forum....waiting patiently.

Can anyone else with a HDR set confirm at this stage that madVRhdrTestFinalColor4.zip is producing a picture with far more depth, contrast, and better tone mapping than the latest version (today)? Maybe I am misunderstanding the latest builds in terms of what it is trying to achieve - are the latest builds purely about trying to restore detail, or is Madshi, using the results of Color test builds, and now on to restore details?

Forgive me if it's a stupid question, but on a HDR set, madVRhdrTestFinalColor4.zip is producing the best image of all test HDR to SDR builds to date (100, 100, 100, 75, 50, 25, off, off)

K

[KoKlusz]

Latest 3 test builds are only making changes to the detail restoration algorithm, the rest of tonemapping pipeline is not affected, as far as i know.

[ajp_anton]

The madVR options thread (https://forum.doom9.org/showthread.php?p=1709764#post1709764) should have the answers to your questions.I think I got it now, but they are not that well explained in the options panel. Still doesn't address the VFR issue.

madVR makes this decision at startup and will use whatever value for the frame rate it is told by LAV, it doesn't analyze timestamps.But doesn't LAV know about it being VFR? At least mediainfo says it's VFR, so it should be known without even looking at the timestamps?

[huhn]

yes they could go and read the timestamps before starting playback and could choice the refreshrate and if or if not SM should be used.

but this sounds like a lot of work.

just the info if it is VFR is not enough. the first frame could be reported with a shorter frame time and the rest is CFR but the file would still be reported back as VFR. this is a real "problem" BTW. but real VFR is really uncommon.

you could add a tag to force SM. sorry i don't know the name of them but pretty sure there is one.

[oldpainlesskodi]

Latest 3 test builds are only making changes to the detail restoration algorithm, the rest of tonemapping pipeline is not affected, as far as i know.

Ok thanks, that's what I was hoping for.

EDIT:

Just re-read what you said. Not sure the latest version is using the same tonemapping as madVRhdrTestFinalColor4.zip, as they look very different on my setup, to my eyes anyway.

[huhn]

that's messed up i try may r9 270 later.

edit: and with later i mean days i guess at week end.

sorry my r9 270 is working like a charm.
but my windows has a lot of changed settings to make it usable.

[kostik]

Previously D3D11 fullscreen windowed was showing only 8 bit but now , OSD shows me D3D11 fullscreen windowed (10 bit) .

Both in HDR and SDR modes.

I am not sure whether nvidia 398.98 hotfix fixed it or the new Win10 18214 build or maybe the latest madVR test build.

[KoKlusz]

Ok thanks, that's what I was hoping for.

EDIT:

Just re-read what you said. Not sure the latest version is using the same tonemapping as madVRhdrTestFinalColor4.zip, as they look very different on my setup, to my eyes anyway.

Well, I wrote that just minutes before latest version dropped, which did changed how madVR reads metadata.

[Siso]

Well, I wrote that just minutes before latest version dropped, which did changed how madVR reads metadata.

Where is this thread in the AVS forum?

[videoh]

https://www.avsforum.com/forum/24-digital-hi-end-projectors-3-000-usd-msrp/2954506-improving-madvr-hdr-sdr-mapping-projector.html

[ryrynz]

Previously D3D11 fullscreen windowed was showing only 8 bit but now , OSD shows me D3D11 fullscreen windowed (10 bit) .


Test build fixed it. Surprised madshi didn't release an interim build or quick fix release here, he's pretty dedicated to avsforum and playing with HDR right now so I guess that's why.

[ajp_anton]

yes they could go and read the timestamps before starting playback and could choice the refreshrate and if or if not SM should be used.

but this sounds like a lot of work.

just the info if it is VFR is not enough. the first frame could be reported with a shorter frame time and the rest is CFR but the file would still be reported back as VFR. this is a real "problem" BTW. but real VFR is really uncommon.

you could add a tag to force SM. sorry i don't know the name of them but pretty sure there is one.
If there's a lot of "fake" VFR, then that's a problem for sure. But VFR isn't as rare as you think. All of my recent smartphones record VFR video. Sometimes frames are skipped or delayed for whatever reasons... CPU throttling because of overheating, background tasks, who knows. Learned it after many hours of pulling my hair when trying to re-encode my videos using tools that assume CFR.

[svengun]

Test build fixed it. Surprised madshi didn't release an interim build or quick fix release here, he's pretty dedicated to avsforum and playing with HDR right now so I guess that's why.

Where are these test builds available ?

[ryrynz]

Been linked multiple times.. with the last like a page ago..

[svengun]

Been linked multiple times.. with the last like a page ago..

I checked the last two pages , I can't find it , can you give me a hint ?

[videoh]

You need to follow this thread:

https://www.avsforum.com/forum/24-digital-hi-end-projectors-3-000-usd-msrp/2954506-improving-madvr-hdr-sdr-mapping-projector.html

madshi links new test builds there. Scan backwards from the last post to find recent builds. E.g., see post #1601.

[huhn]

i hope people read the >projector< part.

[SamuriHL]

Once the project is done, in theory it'll work on any display and could be better than the HDR on some TV's. But it's still very early stages.

[cyberscott]

i hope people read the >projector< part.

True but madshi did implement at least one fix on the test builds in that forum: 10 bit full screen is (non-exclusive)working again. It was stuck in 8 bit after the Windows 1803 update.

I’m glad to see him actively working on madVR even if not within this forum. :)

[Asmodian]

It was not due to the Windows 1803 update. It was due to the Nvidia drivers returning poor quality when using HDR + Windowed 10 bit at the same time, so madshi disabled Windowed 10 bit as a quick and dirty "fix".

[KoKlusz]

Once the project is done, in theory it'll work on any display and could be better than the HDR on some TV's. But it's still very early stages.

It's already working on any display, and it's much, much better than tonemapping of my TV.

[SamuriHL]

It's already working on any display, and it's much, much better than tonemapping of my TV.

Yup, I've been following the AVS thread with great interest. I'm glad to see the work and testing being done. I've not had time to contribute to the testing, unfortunately, but, I'm very interested to see where it goes.