madVR - high quality video renderer (GPU assisted)

[blu3wh0]

Originally Posted by Warner306

You can set madVR to tone map and then pass the result to the display in PQ with HDR metadata. But you would have to find a target nits that disables the display's own tone mapping. And, even then, the display might still alter the input in unwanted ways. You would have to experiment with this setting in the hdr menu (process by pixel shader math).

HDR -> SDR can output to any brightness or color gamut. The problem is your SDR calibration would have to balance both formats. SDR is most accurate close to 100 nits, while HDR is most accurate close to 1,000 or 4,000 nits. So it isn't easy to balance the two with a reasonably bright display.

Thanks, that's why I would prefer to keep the two separated. But you say SDR can output to any brightness or color gamut, but how would that actually work on my TV in SDR mode while maintaining peak brightness and bt.2020? I can't say that's something I've ever thought to try or calibrate towards. Would this result in an image with the same peak brightness and color gamut that HDR mode and source could provide (aside from the fact that tone-mapping would be different, but I would have control over it)?

[Warner306]

Originally Posted by blu3wh0

Thanks, that's why I would prefer to keep the two separated. But you say SDR can output to any brightness or color gamut, but how would that actually work on my TV in SDR mode while maintaining peak brightness and bt.2020? I can't say that's something I've ever thought to try or calibrate towards. Would this result in an image with the same peak brightness and color gamut that HDR mode and source could provide (aside from the fact that tone-mapping would be different, but I would have control over it)?

You get the brightness of your display's SDR calibration, whatever that might be. It would not be as bright as HDR mode, unless you turn the backlight all the way up to the maximum.

The gamut would have to also be shared between the two formats unless your display has a setting to display each gamut accurately without a conversion.

You could create two profiles:

If (HDR) "BT.2020"
else "BT.709"

[blu3wh0]

Sorry, that's not quite what I meant and this might be a TV specific question. With this conversion, can I make my SDR mode perform like the HDR mode, minus the tone-mapping that would be done by MadVR now, if I calibrate it to max brightness and output in bt.2020? Or would using HDR mode still be the best option, especially on my TV? I ask because HDR mode is a bit finicky and I have a lot less control over the calibration.

[Warner306]

Originally Posted by blu3wh0

Sorry, that's not quite what I meant and this might be a TV specific question. With this conversion, can I make my SDR mode perform like the HDR mode, minus the tone-mapping that would be done by MadVR now, if I calibrate it to max brightness and output in bt.2020? Or would using HDR mode still be the best option, especially on my TV? I ask because HDR mode is a bit finicky and I have a lot less control over the calibration.

You could calibrate SDR mode to the maximum brightness. But your gamma curve would likely be less accurate, and this would harm the HDR -> SDR conversion. SDR content would also look terrible.

[blu3wh0]

I can have different TV profiles for normal SDR, so I'm not concerned about that. But it sounds like this conversion isn't really designed for improving tone-mapping in HDR mode on a TV that does it decently enough.

[Warner306]

Originally Posted by blu3wh0

I can have different TV profiles for normal SDR, so I'm not concerned about that. But it sounds like this conversion isn't really designed for improving tone-mapping in HDR mode on a TV that does it decently enough.

You should try the process by pixel shader option and see if it works harmoniously with your TV. Without the ability to disable the display's tone mapping, it is hard to provide tone mapping for a display that enforces its own PQ curve. A 3D LUT can make adjustments working with the display's HDR mode, but it is a fixed curve for all videos.

It is a better option for a projector. Some still choose to use HDR -> SDR by sacrificing some brightness with HDR content for madVR's tone mapping. It partly depends on the brightness of your display in HDR mode.

[huhn]

Originally Posted by Siso

I'll do some testing with the doubling, is SSIM 1D 100%+ scale in linear light and AR-relaxed a good downscaler?

i personally don't like bicubic 150 but that should be more than good enough. there is the possibility to scale chroma and luma differently which may not work with super xbr so NGU low maybe has the upper in this case.

[mclingo]

Originally Posted by Warner306

A noisy fan is common.....

yeah already doing that with argus monitor, useful little all in one fan controller tool, my GPU runs at 20% until it hits 60 degrees, it steadily climbs from there.

I've also logged a ticket with saffire to see if i have a fault with the fans or whether I can buy any quieter versions, its a current card so hoping I can.

cheers,


EDIT - got my figures way off, must have been thinking about another card. My current sapphire rx 580 nitro+ 4gb is only silent under 40% fan, if I run MADVR at NGU sharp high for both upscale and chroma it taxes my card about 70% GPU.

I've found that if I set me fan curve to to run at 40% fan right up to 80 degrees it remains at a stable 75 degrees whilst running fairly silent, this was with a 4k HDR movie which seems to tax the card much more than upscaling from 480p to 4k for some reason... go figure...

[KoKlusz]

Originally Posted by Warner306

You get the brightness of your display's SDR calibration, whatever that might be. It would not be as bright as HDR mode, unless you turn the backlight all the way up to the maximum.

It's still won't be enough. Consumer level oleds peak around 480 nits when not using internal HDR mode. I guess this is for power efficiency reasons or/and to prevent burn-in.

[Manni]

Originally Posted by foobar2000

actually currently no video card could run the highest setting of madvr-upscaling, like upscaling a 480p video to 4k, even the 1080ti drops a lot frames, my friend tests it with a common anime video (x264, 500MB, 20mins)

I have no idea what you're talking about, I upscale 480p video to 4K with my 1080ti and I don't drop any frame. Of course, I don't use the same settings as for 1080p or 4K, but that's what profiles are for.

In any case, this has NOTHING to do with the build you linked to, which is an experimental build for HDR conversion with pixel shader.

I'm only saying, please post here about official builds, and there if you can help the work in progress instead of wasting our time.

If you have any questions about upscaling 480p content, feel free to ask here but don't use an experimental build that had a bug to make a point. It's ridiculous.

[foobar2000]

We could say the highest upscaling section in madvr surely is a monster, 1080ti basically deals with every games but it can't 100% smoothly run the ultimate upscaling setting at all, not even close. My friend just pre-ordered the 2080ti, but he doesn't think the new card could break the ice, madvr is already another 3DMark...

[nussman]

Yes madVR has some demanding filters for video processing. So what? Some algo’s for video processing even can not be done in realtime at the moment. So please stop posting nonsense ...

[el Filou]

foobar2000 is a troll

[Asmodian]

Originally Posted by foobar2000

actually currently no video card could run the highest setting of madvr-upscaling, like upscaling a 480p video to 4k, even the 1080ti drops a lot frames, my friend tests it with a common anime video (x264, 500MB, 20mins)

We are already very close with the 1080 Ti. My Titan X (Pascal) can run 480p30 -> 4K with NGU AA very high chroma upscaling, debanding & RCA, doubling twice with NGU Sharp very high, upscaling with Jinc, smooth motion, and ED1 dithering. My card is watercooled and overclocked, and I have soldered extra resistors to the current monitoring circuit to double the available power, but madVR didn't go above 120% power too often even before that.

720p -> 4K is harder, I need to use NGU high for the quadrupling step. However, NGU high is a lot faster than very high for a very small difference in quality, a 2080 Ti might be give more headroom but it wouldn't improve quality noticeably for anything <60fps.

[ryrynz]

Originally Posted by Asmodian

We are already very close with the 1080 Ti. My Titan X (Pascal) can run 480p30 -> 4K with NGU AA very high chroma

Why NGU Very High for chroma? you know that's taboo around here. I actually didn't notice it was you posting this at first haha.

When the time comes for Avisynth/Vapoursynth you guys upscaling 480ish anime are gonna love Awarpsharpen, blows the pants off NGU sharp.
I have a fairly simple script that does wonders for 720 and 480 animated content and you mix that with NGU AA and it's just amazing ^.^

[Asmodian]

I always use NGU high for chroma when actually watching things.

But the GPU is easily capable of using very high for chroma if desired (chroma upscaling isn't very hard with 480p anyway). I have a lot of headroom with my normal 480p30 settings. I also usually use NGU AA. I don't like Awarpsharpen or NGU Sharp that much, they can change the look of the lines too much. Old SD Anime is supposed to look like old SD Anime.

Using RCA high + NGU AA very high quadrupling can get a little tricky though, sometimes RCA High + NGU Sharp looks better than NGU AA to me (it takes a bit more GPU power too).

[Siso]

Originally Posted by huhn

i personally don't like bicubic 150 but that should be more than good enough. there is the possibility to scale chroma and luma differently which may not work with super xbr so NGU low maybe has the upper in this case.

I did some testing last night with doubling using NGU low-standart for both luma and chroma+error diffusion 1 and for downscaling ssim 1D-ll+ar.

Rendering times for 2.40:1 movies are ~17 ms GPU load ~ 42%, for 1.85:1 are ~20 ms GPU load ~ 50%. The question is it really good idea to use error diffusion 1 for dithering, which bumps up the rendering times plus 3-4 ms, or to use ordered dithering?

[ryrynz]

Difference is minimal but if you got nothing else to spend the available card performance on then I'd use it. ~20 is still on the low side for 24fps.

[Sunset1982]

I've got a question for those of you with AMD crads...

Is there anything I need to know when I want to change my nvidia and use a amd rx480 instead?

I need the more powerful gtx1070 in my gaming rig.

Is 23,976 working with amd? what about hdr switching?

[Siso]

Originally Posted by ryrynz

Difference is minimal but if you got nothing else to spend the available card performance on then I'd use it. ~20 is still on the low side for 24fps.

Then I guess, I'll leave error diffusion 1 on, and basically these are my optimal settings for GTX 1050 ti. Will test more to see if there are dropped or delayed frames.