madVR - high quality video renderer (GPU assisted)

[cremor]

cremor
Do you use a 23 Hz mode of any kind (standard or optimized)? Other refresh rates than 23 Hz don't have problems in short intervalls of frame drops or repeats, as far as I know.

No, I just use 100 (25*4) and 119 Hz (23,976*5). Since this is a PC monitor, I don't like lower refresh rates because of sluggish mouse movement etc.

[Siso]

Guys, for scaling factor 1.3 what do you recommend? Image upscaling: Jinc+upscaling refinement (crispen edges or adaptive sharpening), or image doubling: NGU Sharp medium or high?

[zaemon]

Guys, for scaling factor 1.3 what do you recommend? Image upscaling: Jinc+upscaling refinement (crispen edges or adaptive sharpening), or image doubling: NGU Sharp medium or high?



Go for NGU if you can afford it.

[Siso]

Go for NGU if you can afford it.

I can afford NGU Sharp High, but my thought is - is it worth the gpu load...

[tp4tissue]

Go for NGU if you can afford it.

I'm a huge NGU fan.

For < 1080p files, NGU makes a huge difference even at a distance.

For >= 1080p files, If you're sitting < 5 feet from the monitor/ tv . NGU still looks sharper.

However, for > 5 feet, It's almost impossible to tell NGU vs Nearest neighbor (1080 to 4K) unless you're running a test pattern.

And I tested myself for NGU vs Lanc (1080 to 4K) @ 5 feet, I got it wrong 50% of the time, So I definitely can not tell at that distance for 1080p files. ESPECIALLY for moving image,

Overall, my point is, DO consider being green when possible. Lanc looks very good at 5+ feet.

[Asmodian]

I agree for a ratio of 1.3x. At that low of an upscale NGU+SSIM100 v.s. Jinc or Lanczos is very subtle in almost all viewing conditions. NGU is only useful for ratios below 1.5x when you use electric heat anyway.

[nussman]

@Siso was asking for scaling factor 1.3, Nearest neighbor for integer scaling is a complete different story.
Anyway if you have to ask here then NGU is probably not worth it? You have to test it, because differences depend on your content and then decide on your own.

[Siso]

@Siso was asking for scaling factor 1.3, Nearest neighbor for integer scaling is a complete different story.
Anyway if you have to ask here then NGU is probably not worth it? You have to test it, because differences depend on your content and then decide on your own.

I'm watching 1080p good encodes on the monitor which is 21:9 (2560x1080p) I can manage NGU Sharp High with my GTX 1050ti, but I was wondering if I can use image upscaling settings instead of image doubling. The visual difference is not that obvious between them.

[nussman]

Of course you can. I am using Jinc/DXVA upscaling most of the time and „NGU high quality settings“ for a special movies night. ;)

[Siso]

Of course you can. I am using Jinc/DXVA upscaling most of the time and „NGU high quality settings“ for a special movies night. ;)

NGU High is good, but does it really need the ~24 ms rendering time on my card...

[ryrynz]

What sort of question is that? Take it or leave it.
Change chroma or drop back a notch if 24ms is "too high" for you.

[tp4tissue]

NGU High is good, but does it really need the ~24 ms rendering time on my card...

If you're on the small screen yes, because you're really close.

On 1080 to 4K from 5 feet w/ a 40-60 inch, it's very hard to tell between Nearest and NGU (highest), Nearly impossible to tell between NGU and Lanc

[tp4tissue]

What sort of question is that? Take it or leave it.
Change chroma or drop back a notch if 24ms is "too high" for you.

I think he means, is the visual improvement worth the Processing Effort.

[ryrynz]

I've not seen one comment on here protesting it ever.
It's his decision to make is he seriously just ticking things and saying "is this good?" He's been on this forum for quite a while.. Anyone would think this was a new account.

[Siso]

I've not seen one comment on here protesting it ever.
It's his decision to make is he seriously just ticking things and saying "is this good?" He's been on this forum for quite a while.. Anyone would think this was a new account.

I see nothing wrong asking other forum members for their opinion.

[ryrynz]

I kinda like NGU Standart, NGU Sharp seems kinda artificial to me

:confused:

[Siso]

:confused:

I did some more testing and I stopped on NGU Sharp. ;)

[Warner306]

I'm a huge NGU fan.

For < 1080p files, NGU makes a huge difference even at a distance.

For >= 1080p files, If you're sitting < 5 feet from the monitor/ tv . NGU still looks sharper.

However, for > 5 feet, It's almost impossible to tell NGU vs Nearest neighbor (1080 to 4K) unless you're running a test pattern.

And I tested myself for NGU vs Lanc (1080 to 4K) @ 5 feet, I got it wrong 50% of the time, So I definitely can not tell at that distance for 1080p files. ESPECIALLY for moving image,

Overall, my point is, DO consider being green when possible. Lanc looks very good at 5+ feet.

This chart appears to be fairly reliable and is probably a good guide for perceived resolution:

Resolution: Screen Size vs. Viewing Distance (https://photos.smugmug.com/AVSFORUM-2/i-9MVJNTS/0/f90eecbf/O/8K%20RESOLUTION%20DISPLAYS%20WORTH%20IT%20CHART%20%28hi-res%29.jpg)

[tp4tissue]

This chart appears to be fairly reliable and is probably a good guide for perceived resolution:

Resolution: Screen Size vs. Viewing Distance (https://photos.smugmug.com/AVSFORUM-2/i-9MVJNTS/0/f90eecbf/O/8K%20RESOLUTION%20DISPLAYS%20WORTH%20IT%20CHART%20%28hi-res%29.jpg)

My test was for difference between NGU Lanc and Nearest @ Greater or Less than 5 feet.

Where with a 1080p file, expensive upscaling is a bit wasteful.


As for choice of Movie Resolution. 4K is ALWAYS better than 1080p at ANY Distances due to 4:2:0. it's the only way to get 1080p chroma.

[Warner306]

Things are different with a projector-sized screen. Projector owners see a lot of benefit with 1080p -> 4K upscaling.

[tp4tissue]

well yea, projected is usually much larger, and bigger pixels.

Although from what I've seen of NON-native 4K projectors, the pixel shifting has an effect similar to analog upscaling (scanning), In that case, I'd say again NGU may not be so necessary, since it's not producing the intended NGU output anyway.

For most TV sizes/ distances, its' ok to go Green.

[Warner306]

Yes, you'd want a native 4K projector if you placed a premium on upscaling. Even then, many are using very high settings for image upscaling and derive a lot more visual improvement than say a 55" 4K TV.

[Bl4ze87]

Sorry for the question, but what Nvidia driver versione is better?
I ve a 1050ti

[madjock]

Bl4ze87

https://forum.doom9.org/showthread.php?t=176013

[SirMaster]

Anyone have any suggestions for this issue?

I am not sure where to ask it, but I think it's probably a Windows or maybe NVidia thing?

I set up a HTPC for a friend, but for some reason, whenever a video opens or closes (Like starting a video in MPC-HC or then closing MP-HC itself) it causes his display to have to re-acquire it's signal.

This is pretty annoying as he has an older JVC (RS600) with a near 20-second sync time.

Now, I am completely familiar with refresh rate switching and such, but I have all of that completely disabled. I have the HTPC set for 3840x2160 23Hz RGB 12-bit (tried 8-bit as well). Refresh rate switching is disabled in madVR and disabled in MPC-HC.

Any ideas of how I can have it so that the JVC doesn't need to re-sync every time I open and close a video?

Well, I figured out the screen re-sync issue.

It was a combination of software and hardware.

It happened when using in madVR calibration tab "Report BT2020 to display" enabled and when sending the video through a Marantz 8802.

If we send the video to the JVC RS600 directly, it never blanks with "Report BT2020 to display" enabled, or if we send the video through the Marantz 8802 with "Report BT2020 to display" disabled it also never blanks.

But send "Report BT2020 to display" enabled into the Marantz and it blanks every time opening and closing a video through madVR.

Weird!

[svengun]

Well, I figured out the screen re-sync issue.

I also have that option enabled and send my signal through a Marantz SR6010, but I don't have that issue, also use an RX2070 (latest driver)

So double weird :-)

[Manni]

Well, I figured out the screen re-sync issue.



Is your Marantz set to HDMI enhanced in the HDMI config? Otherwise it reduces your bandwidth to 10Gb/s for compatibility.

I have an X8500H and HDMI Enhanced needs to be enabled to enable the full 18Gb/s bandwidth.

If you don't enable BT2020 when using 12bits, the native gamut isn't selected and your calibration is all wonky.

It needs to be enable in 12bits to get the correct gamut in SDR BT2020.

Not a problem with 8bits.

[SirMaster]

Is your Marantz set to HDMI enhanced in the HDMI config? Otherwise it reduces your bandwidth to 10Gb/s for compatibility.

I have an X8500H and HDMI Enhanced needs to be enabled to enable the full 18Gb/s bandwidth.

If you don't enable BT2020 when using 12bits, the native gamut isn't selected and your calibration is all wonky.

It needs to be enable in 12bits to get the correct gamut in SDR BT2020.

Not a problem with 8bits.

It is set to Enhanced.

Also, to solve the problem we ended up just outputting 2 cables from the HTPC. One is run straight to the JVC for video, and the other is run straight to the Marantz for audio. Lip sync is then set manually to line up correctly.

This way we can have "Report BT2020" enabled without any issues, because we are using 2160p 23Hz 12-bit RGB on the RS600.

[Manni]

It is set to Enhanced.

Also, to solve the problem we ended up just outputting 2 cables from the HTPC. One is run straight to the JVC for video, and the other is run straight to the Marantz for audio. Lip sync is then set manually to line up correctly.

This way we can have "Report BT2020" enabled without and issues, because we are using 2160p 23Hz 12-bit RGB on the RS600.

As you know, using 12bits on the JVCs isn't recommended because it forces to YCC422 behind the renderer's back.

Worth trying 8bits and see if you still have these issues. You won't get any banding (if madVR is set to 8bits dithering) and you'll get better chroma.

[SirMaster]

As you know, using 12bits on the JVCs isn't recommended because it forces to YCC422 behind the renderer's back.

Worth trying 8bits and see if you still have these issues. You won't get any banding (if madVR is set to 8bits dithering) and you'll get better chroma.

Yes, but again this is a friends HT that I am configuring and it's an RS600, so 12-bit RGB works fine.

Yes on my own HT with my NX5 of course I use 8-bit with madVR error diffusion dithering and I don't need the report BT2020. But it also does not cause re-sync through my Denon X4200W anyways for if JVC does ever fix 12-bit.

[Manni]

Yes, but again this is a friends HT that I am configuring and it's an RS600, so 12-bit RGB works fine.

Yes on my own HT with my NX5 of course I use 8-bit with madVR error diffusion dithering and I don't need the report BT2020. But it also does not cause re-sync through my Denon X4200W anyways for if JVC does ever fix 12-bit.

12bits doesn't work fine on the rs600, it does exactly the same re chroma (forces YCC422 with RGB or YCC444 12bits), but you can't use 8bits without getting the magenta bug, so not an option indeed unless you only use 60p, you have to live with the forced YCC422 (or select YCC422 to start with).

[SirMaster]

12bits doesn't work fine on the rs600, it does exactly the same re chroma (forces YCC422 with RGB or YCC444 12bits), but you can't use 8bits without getting the magenta bug, so not an option indeed unless you only use 60p, you have to live with the forced YCC422 (or select YCC422 to start with).

Oh, I thought I remember you saying 12-bit RGB worked on your RS500.

So what are you saying for how should he run his HTPC to RS600 then for tone-mapped HDR? What should we set in nVidia, YCC422 so that it's not happening behind madVR's back?

[Manni]

Oh, I thought I remember you syaing 12-bit RGB worked on your RS500.

So what are you saying for how should he run his HTPC to RS600 then for tone-mapped HDR? What should we set in nVidia, YCC422 so that it's not happening behind madVR's back?

Yes, 12bits "works" with the rs500, and I had no choice because of the magenta bug in 8bits.

You should test chroma with various options. Given that the RS600 will force YCC422 with any colorspace in 12bits, I would suppose that selecting this in the driver would be best, but when I do this with the rs2000 it's still a worse result than using 8bits.

[miroslav22]

Just started using Madvr, very impressed!

I just had a couple of questions I'm struggling to find answers to. Apologies if I'm not fully understanding this!

The issues I have are:

I want to use madvr to upscale TV at 50hz to 2160p, however switching to 50hz only works if I set the colour output in the Nvidia control panel to 4:2:0 chroma/luma sampling OR 8 bit precision (due to the bandwith limitations of HDMI 2.0). For some reason I can only select 8/12 bit for precision and not 10 bit. My understanding is that HDMI 2.0 has enough bandwidth for full RGB/10 bit output at 50hz so it's annoying the option doesn't seem to be available. (Is there a reason for this?)

But I also want HDR output for films at 23p. So I guess I am forced to set the Nvidia control panel to output 12 bit and not 8 bit? If this is the case I am also forced to use 4:2:0 chroma/luma to allow 50hz to work.

So I guess my question is are you 'stuck' with what you set in the Nvidia control panel for all refresh rates, or can/does Madvr change the colour precision/subsampling output dynamically, along with the refresh rate? I'm running in DX11 FSE mode

ie Can/does madvr switch the output to 10/12bit RGB at 2160p23 and 8bit RGB (or even 12bit 4:2:0) at 50hz?

If not then is the quality difference of always outputting 4:2:0 vs RGB negligible for HTPC use?

[Asmodian]

Why is 12 bit important to you? 4:2:0 at 12 bit is much lower quality than RGB at 8 bit.

I suggest you simply use 8 bit RGB all the time. madVR cannot change the bit depth but if you set 23Hz as 12bit the driver will use 12 bit when madVR switches to 23Hz.

Please do some visual tests, if 12 bit looks the same as 8 bit to you please do not obsess about getting 12 bit. 12 bit really is the least important improvement and not worth sacrificing anything else for.

[miroslav22]

Thanks for the reply... Sorry I was under the impression that the display needed to be configured to 10 or 12 bit for HDR to work. Is that not the case? This was the only reason I was trying to use it.

If it does require it then it seems I must choose between having HDR available (but using 4:2:0) or full RGB, as I can't have both (due to needing to switch to 50hz for TV)

[Asmodian]

No, HDR does not need 10 bit, at least on Nvidia with old and new drivers. There were some in the middle that did require 10 bit.

AMD does require 10 bit for HDR. :(

[miroslav22]

Many thanks I understand much better now. I'd assumed the extra bits were needed to specify the greater luminance for HDR but I've realised this isn't how it works after a bit more reading.

You're right, 10 -> 8-bit dithered by madvr looks almost identical to 10-bit native close up, and is completely indistinguishable at viewing distance. Certainly not worth worrying about!

Thanks again!

[tp4tissue]

To Be fair, it looks almost identical, because most panels do 10-bit badly or are not true 10 bit designs.

[Asmodian]

It would look almost identical even on a very good 10 bit display. The difference between well dithered 8 bit and well dithered 10 bit is simply very subtle.

[Daicon]

Hi,

i am noob for here.
Have this problem:

Add images:


Same configuration settings TV



PLEX TV Oled c8 - HDR:
https://d2aztkdj0ezvrk.cloudfront.net/items/04180C181e2d2J3a3O2c/Image%202019-05-07%20at%207.51.40%20PM.png




TV OLED c8 - Mode HDMI / MODE PC / MODE blueray
HTPC - Vega 11 - MAdvr HDR

https://d2aztkdj0ezvrk.cloudfront.net/items/352y0c2D3P3w0g301o3u/Image%202019-05-07%20at%207.50.11%20PM.png

The HDR logo is activated on TV. (this is not the problem)

Why is it lost in color?
Has anyone had problems with Vega 11?

THANKS
:-S

[Asmodian]

Are you sure it isn't your TV? That looks like a very different white point and you do not mention what mode you are using for the PLEX. Try Expert Dark Room for both. Are you using madVR's tone mapping or just passthrough?

Sorry, I have no experience with Vega.

[DMU]

Daicon
Are you sure it isn't your TV? That looks Are you using madVR's tone mapping or just passthrough?
It is also important the FSE (exclusive) or FSW (windowed) you use for Vega. What OS?

PS: No one noticed the image changes a change in Windows10 1903 with HDR passthrought?

[Alexkral]

That's usually the difference between 709 and 2020. You're using passthrough, or "output video in HDR format" (or 2020 in calibration settings). PLEX is converting to 709.

[Asmodian]

With 2020 displayed as if it was 709 I usually get obviously lower contrast. I don't have a lot of experiance trying to judge photos of displays but the contrast seems similar on both. It looks like a white point and tone mapping difference to me, but I could be wrong.

[Alexkral]

With 2020 displayed as if it was 709 I usually get obviously lower contrast.

That's if the PQ transfer function is not applied to the HDR 2020, but in this case it is.

EDiT: But you're right, it seems a different color temperature.

[Klaus1189]

Anybody else tested 1903 and encountered some issues?

[SamuriHL]

Nope. Been running 1903 since it dropped in insider builds a while ago. No issues for me so far.

[Daicon]

Are you sure it isn't your TV? That looks like a very different white point and you do not mention what mode you are using for the PLEX. Try Expert Dark Room for both. Are you using madVR's tone mapping or just passthrough?

Sorry, I have no experience with Vega.
Sure.
The TV have same settings and same mode (Technicolor) with Plex and HTPC Madvr.

just passthrough.
:-(


Daicon

It is also important the FSE (exclusive) or FSW (windowed) you use for Vega. What OS?

PS: No one noticed the image changes a change in Windows10 1903 with HDR passthrought?

Windows 10.
I activated FSE, but dont have any effect different. :-S

That's usually the difference between 709 and 2020. You're using passthrough, or "output video in HDR format" (or 2020 in calibration settings). PLEX is converting to 709.
MMM
I have researched a little, and it seems that the problem continues that way ...

That is to say.
When I deactivate "passthrough", and I select in "calibration" 709, the image (SDR) is very similar in color to the HDR of PLEX (but without the brightness / etc of the HDR, of course.)
When selecting 2020, the image (SDR) becomes more muted, as it happens to me in HDR.

The problem is that I select the options that I select, when HDR is activated in full screen, I always get the SAME result as the 2nd image I have set.

Any solution?
What am I doing wrong...??

[DMU]

Windows 10.
I activated FSE, but dont have any effect different. :-S
What version of AMD drivers and Windows 10?