madVR - high quality video renderer (GPU assisted)

[huhn]

i get ~28000 on my 1060 i will do a better test later but relative to my 960 this is actually matching.
960 ~5073
1060 ~10073
on paper a 960 has 2500 GF and a 1060 4.600 GF taking the boost behaviour and the architectural difference this could be spot on.

[okc_smoker]

This may be a pipe dream, but it would be great to have some sort of automated test bench. Something like 3DMark or Passmark. Just something a user would run and it would spit out scores for, say, 720p -> 4k, 1080p -> 4K, Tone Mapping, etc., using the appropriate scalers and everything else set to default.

[huhn]

3 more runs:
27854
28234
28214

working flawless for me right now but it has to work at least for the majority too.

[toki]

madVR isn't creating profiles when putting in user-created settings bins. I cannot get it to change

Edit: had to revert to default before adding in.

Now if someone could tell me how to keep the seekbar from turning blue with no markers when going into fullscreen.

[Asmodian]

Now if someone could tell me how to keep the seekbar from turning blue with no markers when going into fullscreen.

Disable FSE.

[toki]

Disable FSE.Thank you. I tried your max and low powered settings and I couldn't even handle low power on a 1060ti : (

now, how can I ungrey perfrom tone and gamut mapping in less accurate color space?

[Asmodian]

Ah yes, my low power settings do not consider the performance hit from HDR. :o

I am not sure what you mean by your question, sorry. Are you talking about an option from the hdr test builds?

[ng4ever]

That is easy. :)

[toki]

Ah yes, my low power settings do not consider the performance hit from HDR. :o

I am not sure what you mean by your question, sorry. Are you talking about an option from the hdr test builds?it's option on the last official build under trade performance. I just put it back with last beta build. That fixed it.

Now, I've trid multiple combos for image tweaking. The only thing that seems to work without causing stutter is ngu anti alias (medium), downscaling, ssid (100), image upscaling dvxa. Every other combination bumps it too high and starts skipping. So far, lol. I'm sure I'll find something wrong with it.

[toki]

I meant to say 1660ti for the card I'm using.

[Plutotype]

Just a remark with custom modes "Test settings" button in madvr. I hit it accidentally and since then, TV went black and there is no signal. Now I dont a have a second monitor to fix it. Cool.

[huhn]

power button.

[x7007]

I have 1080 GTX now. I might move to AMD, any specific issues with Madvr and AMD drivers/Cards? or it's just same settings as Nvidia now

[ryrynz]

AFAIA the Navi cards are golden when it comes to MadVR, let us know how well it works compared to the 1080 for 24hz and NGU performance.

[oldpainlesskodi]

I have 1080 GTX now. I might move to AMD, any specific issues with Madvr and AMD drivers/Cards? or it's just same settings as Nvidia now

Same here, I just pulled the trigger on a RX5700 XT...will be here later today...

[Plutotype]

power button.

I had to connect a PC monitor to be able to fix the issue. Restart/hard shutdown didnt help with the GPU outputing to my TV. Probably bad timings stayed baked there.

Btw with the old driver 383.13 I was able to get 23.976Hz with frame drops/repeats every 4 hours. But there was Windows 10 update couple of days ago, which must change something and suddenly, the custom timings in madvr were not working anymore ( stats were showing 23.971Hz which is default Nvidia ). I have upgraded to Nvidia driver 431.60 and played with the madvr custom resolutions to optimize it. The best I could get was 23.97734Hz, which results into frame repeats every 12 minutes.

Is there a Nvidia driver version which can give me 23.976 with the recent version of Windows 10?

Thanks

[mclingo]

I have 1080 GTX now. I might move to AMD, any specific issues with Madvr and AMD drivers/Cards? or it's just same settings as Nvidia now

There are two minor issues I am aware of.

1 - I have an issue where colours/reds are blown out with SDR material sub 60hz, I have tried any of the new drivers so I dont know whether its fixed. You will only see it if you are setup for HDR / 10 bit though. This can easily be fixed though on my system by setting up madvr calibration to BT2020.

2. The is a bug in MADVR when using 3D, if your desktop is 4k and you play a 3D MVC movie, when it comes out of the movie and drops back to 4k you get HDMI loss, there are lots of ways around this.

Some people have reported banding in some situations, i've never seen any ever, I have a 2015 4k LG EF950v OLED.

I'm currently using a RX 580, its good for NGU high but I usually run it at medium or low in to keep heat and noise down, i generally find there is no distinguishable differce between high and low for me anyway.

[mclingo]

Same here, I just pulled the trigger on a RX5700 XT...will be here later today...

i'd be really interested to know how you think this compares to a 1060 / 1080 in your opinion when you get chance, these seem to be the most popular cards on here for NVIDIA.

[oldpainlesskodi]

Will post some observations here tomorrow.

[RealSnoopyDog]

How complicated is it, to add support for HLG (Hybrid Log Gamma)? This is another HDR format which is used with DVB transmissions to save bandwidth. And who needs to support it? Only madVR or the graphics card as well?

[nevcairiel]

How complicated is it, to add support for HLG (Hybrid Log Gamma)? This is another HDR format which is used with DVB transmissions to save bandwidth. And who needs to support it? Only madVR or the graphics card as well?

If madVR supported it, it could at least make use of the HDR data to create a better SDR image.

If you want the HLG data actually send out to the screen without madVR processing it to ordinary SDR, the graphics card would probably need to accept it as a valid mode. Luckily HLG has no extra metadata, so that might be easier.

[Alexkral]

For SDR output you can make a 3DLUT with DisplayCal.

Edit: You have the option of using the SMPTE 2084 roll-off with an HLG profile, but that probably won't work.

[huhn]

madVR doesn't read HLG yet and doesn't like HLG 3D LUT.
i just "convinced" madVR that this 3D LUT is fine just use it and it worked fine with an HLG source.

[YukonTrooper]

Is there any known cause for presentation glitches to accelerate over time? I've tried almost every option in the general settings page to no avail. Rendering times are ~10ms and there's no repeated or dropped frames.

[jkauff]

You don't say what presentation glitches are occurring. You also need to tell us something about your system, such as what graphics card are you using, and how much video memory does it have?

[oldpainlesskodi]

Will post some observations here tomorrow.

Initial findings (summary)

Like:

1st ever time using an AMD card...love the black levels (picking up details in shadows and dark scenes that were never visible with the 1080) and PQ (appears cleaner,and yet more detail at the same time), and control over just about every aspect of PQ.

Using Wattman (auto overclock), the card is happy and stable at 2130. About 30% faster that my overclocked 1080 strix (render times).

Dislike:

Control over just about every aspect of PQ (for us types...oh well, even more obsessing)

Software not as intuitive as Nvidia.

Seems to throttle at a lower threashold (nivida on 23p, seems to be about 39 vs 32 ms)

HDR is better on Nividia because of the API links in Madvr (I assume) being able to tell the display its a Bt2020 signal.

Verdict - 1080 up on ebay over the next few days, but still have lots more testing and tuning to do.

[huhn]

there is no difference in black level this has been proven countless times using measurement hardware you are imagine things or your level settings are messed up.

can you make a screen with ShowRenderSteps would be nice to see AMD back with some NGU capable cards.

[nevcairiel]

IIRC, AMD has (or used to have) some image improvement processing options enabled by default. Make sure you're not falling into a processing trap.

[ng4ever]

Just ordered RX5700 XT a few minutes ago.

Should be here in a hour. Can't wait.

[DMU]

oldpainlesskodi
Could you check HDR passthrough at 8 bit madVR output?

[ng4ever]

oldpainlesskodi
Could you check HDR passthrough at 8 bit madVR output?

I will check that for you.

Just got the video card around 30 minutes ago.

[DMU]

ng4ever, thank you.
I managed to output HDR-passthrough at 8 bit madVR output on my Vega using a profiles bug (http://bugs.madshi.net/view.php?id=611), but there is doubt that statistics show the truth.

[el Filou]

IIRC, AMD has (or used to have) some image improvement processing options enabled by default.Isn't madVR supposed to bypass all those? (as long as you're not using DXVA processing in Trade quality for performance, of course)

[YukonTrooper]

You don't say what presentation glitches are occurring. You also need to tell us something about your system, such as what graphics card are you using, and how much video memory does it have?I think I discovered the issue.

I started getting the presentation glitches when using my LG C9 @ 119Hz. I also have a 165Hz G-Sync display connected to the same system. Doing some searching, apparently there can be issues with two high refresh displays, despite using the "Second screen only" option in Windows.

I was able to reduce the presentation glitches by lowering the presented frames to low as 2, although they still occurred. I then tried disabling presented frames and using the backbuffer at default 3 without any presentation glitches. However, I only tested ~5 minutes of playback before going to bed. I'll try extended playback tonight and see if the backbuffer method is indeed truly the fix.

Still, any input is welcome. Maybe there's an optimal setup somebody is aware of I can jump straight to. I also still need to try 1 presented frame and much higher presented frames. Apparently very high presented frames fixed presentation glitches for someone in a similar scenario.

Also, I'm on Windows 1803 with a GTX 1070.

[SamuriHL]

Man that sounds really similar to the issues we get on RTX cards. Try a present frames of 1 and see if that clears it up. Alternatively, turn D3D11 presentation off and use D3D9 and see if the problem goes away. I'd be curious how that works for you.

[el Filou]

There's a note in NVIDIA's driver doc that stutter problems can arise when using multiple monitors with different refresh rates or GSync/non-GSync, maybe that's a case of that.

[Asmodian]

I found disabling G-sync entirely helped a lot. Mixing G-sync and non-G-sync displays has always caused issue for me if I left G-sync enabled. madVR and G-sync are not friends in general. :(

I also think matching refresh rates across multiple displays is a good idea, if you are running a C9 @ 119 Hz you might want to run the G-sync display at that too, with G-sync disabled. :(

Windows DWM and GPU drivers still aren't good at multi-displays refresh rates and madVR does not play well with G-sync.

All that said trying D3D9 Windowed Overlay is definitely something to test, it is the most OS independent option.

[okc_smoker]

Here are a few screenshots using my new RX 5700 (*not* XT). I'm debating returning it and picking up one of the aftermarket XT's that are just now coming out, but we'll see. I've been really impressed with this card, so it may not be worth shaving a few ms per frame.

Note that the rendering times in the screenshots are actually higher than the actual average for some of them (I'm assuming due to the action of taking the screenshot). You can also ignore the dropped frames as I was doing a lot of stopping and starting. I'll post the actual render times with each picture. Also, motion blur, jpeg compression, yada, yada...

This is my first time posting screenshots, so let me know if there's a better way to do this.

Bohemian Rhapsody (720p -> 2160p) (~26ms)

Avatar (1080p -> 2160p) (~24ms)

Planet Earth II (4k HDR w/ tone mapping) (~30 ms)

[huhn]

just use alt print madVR will not even notice with the screenshoot function madVR renders the image differently.
and upload them somewhere else.

[okc_smoker]

just use alt print madVR will not even notice with the screenshoot function madVR renders the image differently.
and upload them somewhere else.

Great, thanks. I'll do that in the future. I initially linked to Imgur, but they were WAY too large and I'm sure that would have been irritating. Some forum software will show a smaller, expandable image when you embed, but apparently not this one.

[huhn]

just use a link and don't embed them.

[seiyafan]

Bohemian Rhapsody (720p -> 2160p) (~26ms)

Avatar (1080p -> 2160p) (~24ms)

Planet Earth II (4k HDR w/ tone mapping) (~30 ms)

Curious what are your upscaling settings?

[seiyafan]

Initial findings (summary)

Like:

1st ever time using an AMD card...love the black levels (picking up details in shadows and dark scenes that were never visible with the 1080) and PQ (appears cleaner,and yet more detail at the same time), and control over just about every aspect of PQ.

Using Wattman (auto overclock), the card is happy and stable at 2130. About 30% faster that my overclocked 1080 strix (render times).

Dislike:

Control over just about every aspect of PQ (for us types...oh well, even more obsessing)

Software not as intuitive as Nvidia.

Seems to throttle at a lower threashold (nivida on 23p, seems to be about 39 vs 32 ms)

HDR is better on Nividia because of the API links in Madvr (I assume) being able to tell the display its a Bt2020 signal.

Verdict - 1080 up on ebay over the next few days, but still have lots more testing and tuning to do.

How's the thermal compared to 1080, does 5700 xt run hot?

[huhn]

it uses more power so it produces more heat. if it runs "hot" or is loud depends only on the cooler.

the stock cooler is the worst all partner card cooler tests i have seen looked very good.

[fireattack]

Hi everyone, I encountered a problem with manual deinterlacing.

If the video has proper interlaced tag, MadVR can detect and deinterlace it just fine.
However, if I want to force interlacing some videos that are marked wrong by pressing ctrl+alt+shift+D, or by adding "deint=on" to the filename, it simply doesn't work.
The OSD does show "deinterlacing on", but no deinterlacing is going on and the final result is still interlaced.

I tested with multiple players and this bug persists. I use LAV Filters to decode, if it matters.

Below are two screenshots, one with auto deinterlacing, one with force interlacing (by adding "deint=on" to filename). You can see from the video canvas that the latter one doesn't work.

https://i.imgur.com/FOIS7sk.jpg https://i.imgur.com/II24ZDJ.jpg

[nevcairiel]

If the video has proper interlaced tag, MadVR can detect and deinterlace it just fine.
However, if I want to force interlacing some videos that are marked wrong by pressing ctrl+alt+shift+D, or by adding "deint=on" to the filename, it simply doesn't work.
The OSD does show "deinterlacing on", but no deinterlacing is going on and the final result is still interlaced.

I tested with multiple players and this bug persists. I use LAV Filters to decode, if it matters.

Below are two screenshots, one with auto deinterlacing, one with force interlacing (by adding "deint=on" to filename). You can see from the video canvas that the latter one doesn't work.


Actually it is performing deinterlacing, but its using the wrong field order (a piece of information the decoder would also convey when the file is tagged properly) - and with the wrong order, deinterlacing produces even worse artifacts then not deinterlacing.

Unfortunately I don't think there is a filename tag for it, but there is Ctrl+Alt+Shift+F to switch the field order during playback.

[huhn]

what are your lavfilter settings?

the only situation where deinterlacing doesn't work even when forced for me is when deinterlacing is disabled in lavfilter which is still a bug in madVR, deint=film works which is strange.

forcing deinterlacing on progressive sources works fine.
but

[huhn]

Unfortunately I don't think there is a filename tag for it, but there is Ctrl+Alt+Shift+F to switch the field order during playback.

it doesn't care if the bug is triggered madVR shows that it is switching the field order but it is still not deinterlacing.

and even if the field order is wrong you would not see his image you would get a total mess in motion but the still image should be fine not like his waved one.

[fireattack]

what are your lavfilter settings?

the only situation where deinterlacing doesn't work even when forced for me is when deinterlacing is disabled in lavfilter which is still a bug in madVR, deint=film works which is strange.

forcing deinterlacing on progressive sources works fine.
but

Yeah I do use deinterlacing = disabled in these testings.

So it's an acknowledged bug I assume?

[huhn]

no never heard of before don't change the deinterlace lavfilter setting in general especial don't use disabled it if you want deinterlacing.

feel free to add it:
http://bugs.madshi.net
don't miss the part about the lavfilter setting so he can easily reproduce the issue.