madVR - high quality video renderer (GPU assisted)

[madshi]

Originally Posted by aufkrawall

Nope, it would just be nice if the edge smoothing strength for direct quadrupling NGU would be as strong as for 2x doubling.

I'm not completely sure what you mean. Basically you seem to say that edge smoothing strength for direct quadrupling is less strong as it is for 2x doubling. But is it? And what exactly do you mean with "edge smoothing"? Do you mean anti-aliasing? Or something else?

Originally Posted by Georgel

Not entirely sure if NGU AA low is equal with NNEDI 16 neurons quality wise, and I'm not entirely sure if it looks exactly the same so it might still be okay to keep it around if it's not too much effort. I mean, NNEDI looks good by itself, but NGU looks pretty good as well.

NGU AA low is better in some images than NNEDI-16 and worse in other images. NGU AA med should beat NNEDI-16 in most images and still be faster.

Originally Posted by Georgel

This new version doesn't have PixArt installed within, no?

What do you mean? The NGU PixArt algo was renamed to NGU Anti-Alias, as mentioned in the v0.91.6 announcement post.

Originally Posted by Warner306

I think the new image upscaling/doubling settings are a mess and difficult to interpret if you are a noob...way too many drop-down menus.

There are too many drop-down menues for my taste. As you might have noticed, I had tried to reduce the number of options, but many users weren't happy with my cuts, so I've brought most settings back. However, to make things easier for noobs, most settings have a "let madVR decide" setting now, and also default to that setting.

FWIW, what you're seeing now is the "expert" version of the settings dialog. When I reach v1.0, there may be a noob view with limited options. But I've not decided that yet.

Originally Posted by nevcairiel

No layout in the world can suddenly make noobs not be noobs.
If they don't even understand the basic concepts being offered there (ie. what doubling means), it doesn't matter how the options look. The "let madVR decide" modes are already much easier then before.

Agreed.

Originally Posted by jkauff

madshi, I've been a professional UI designer for 25 years. If you ever want to bounce ideas off me, or get an opinion on a design, feel free to reach out. I'd be glad to help.

Thanks. But it's not the right time for that now. Right now I'm trying to do only what's absolutely necessary, in as short time as possible.

Originally Posted by Telion

And please add a promised hotkey for the settings dialog, madshi. I hope it won't take much efforts.

All in good time.

Originally Posted by dioxholster

I dont know what madvr settings to use. example, super-xbr ? Using amd 470 card.

If you like sharp, for good quality sources try "NGU Sharp". For bad quality sources, try "NGU Anti-Alias". It's less sharp, but more forgiving.

Originally Posted by ABDO

NGU AA need more performance than NGU pix art 3.

It's not supposed to!! You're the 2nd user who reports this, and I don't know why you have that experience. Can you try to find out more about that? NGU AA is supposed to have exactly the same speed as NGU pixart 3. Except for the "high" quality preset, which wasn't available with NGU pixart.

[madshi]

madVR v0.91.7 released

http://madshi.net/madVR.zip

* fixed: doubling twice with "normal" chroma quality resulted in weird colors
* fixed: "downscaling after doubling" dropbox was missing an item

[jmonier]

I'm trying out 91.6 replacing 91.5 on a GTX1060 6gb and trying to have the same settings as before: NGU high for luma and NGU Sharp med for chroma (was NGU med). With the same material I'm getting 13-14 ms times vs 7-8 ms with 91.5 and the GPU utilization is 80% vs 50% (as reported by GPU-Z). Am I doing something wrong? Could "let madVR decide" have a greater effect now? Or is there something else that has changed that much in 91.6?

EDIT: I went back to it and now 91.6 is about the same as 91.5. This is something that I've been seeing sporadically before but haven't been able to reproduce when I want to.

[madshi]

Weird! Which GPU? Can you make a screenshot of the same movie showing the rendering time difference, with the OSD turned on, with both v0.91.5 and v0.91.6? Please also create an empty file "showrendersteps" in the madVR folder before creating the screenshots.

[Magik Mark]

madshi,

Which NGU variant would you recommend that would perform similarly as of neddi3 64 neurons?

[Damien147]

I think a benchmark like SVP for noobs to automatically set settings is a good idea for v1.0.Also is 1.0 gonna be a paid version?Just curious.8 years is a long time and you need something like a ''prize'' for your work.I don't need it but I'll buy it because I feel grateful.Maybe a donate button!That works too.


But fix the black screen with dxva2(native) 10bit hevc amd rx4x0 first.

[ryrynz]

Originally Posted by Magik Mark

Which NGU variant would you recommend that would perform similarly as of neddi3 64 neurons?

You know there's only two versions of NGU right? And you read madshi's release post a couple of pages back right?.. It's not hard to put two and two together.

[har3inger]

The anti alias NGU is quite good, and IMO is probably good enough to replace nnedi3. I'm more interested to see how the direct 4x for it will look, since that's the most relevant application for speed/performance ratio on 480p and lower sources.

Did you change something in the NGU sharp scaler? It runs a smidgeon (5-10%) slower than it used to in 91.5.

Edit: Yep. Reverted to 91.5, and lost 3ms worth of render time from 36ms when using NGU high. Unfortunately, this is the difference between perfectly playable and stuttering mess for me. (91.6 is slow. 91.7 is similar, or even a tiny bit slower than 91.6)

[dioxholster]

Originally Posted by jmonier

I'm trying out 91.6 replacing 91.5 on a GTX1060 6gb and trying to have the same settings as before: NGU high for luma and NGU Sharp med for chroma (was NGU med). With the same material I'm getting 13-14 ms times vs 7-8 ms with 91.5 and the GPU utilization is 80% vs 50% (as reported by GPU-Z). Am I doing something wrong? Could "let madVR decide" have a greater effect now? Or is there something else that has changed that much in 91.6?

EDIT: I went back to it and now 91.6 is about the same as 91.5. This is something that I've been seeing sporadically before but haven't been able to reproduce when I want to.

I dont know, but it feels like the framerate is more sped up with the new one, or it could just be me.

[Neo-XP]

Originally Posted by madshi

I don't understand how that would work. If you select one of the doublers in the "image upscaling" settings dialog, you can't select Lanczos 3, anymore. It's either Lanczos 3 or e.g. NGU in the current settings dialog. So how does "same as upscaling algo" make any sense? It would then be NGU again for chroma doubling. Or am I missing something?

It would be the same as the upscaling algo after doubling, as in v0.91.1.
Something like "same as upscaling after doubling algo". Would that be possible ?

Originally Posted by madshi

Anyway, it doesn't matter too much, NGU Soft/Standard/Sharp will all change again soon.

I can't wait to test that

[AngelGraves13]

I hope NGU AA will get direct 4X soon.

[Arm3nian]

How does NGU quadrupling work with "refine the image after every 2x step / only once" options? If I have sharpen edges at 4.0, refine the image only once, and NGU on "double again" then it seems to be at 4.0. Same with direct quadruple. But if it's "refine the image after every step" then it looks like 8.0. And if NGU is on direct quadruple and I select refine after every step then it overrides it and uses NGU twice. So if you could select sharpen edges at 8.0 then you could use NGU direct quadruple with the same look as doubling twice and refining the image twice right now. But then I guess that would go up to 16 in the new version?

[Anima123]

NGU-AA low consumes more GPU power than previous NGU pixart, the Intel Graphics 630 can hardly handle 720p -> 1024p @ 23.976 fps fluently, the playback stuttered from time to time, maybe due to throttling?

[HillieSan]

Originally Posted by leeperry

OK, so both 10 & 12 bit 4:2:2 seem possible in HDMI 2.0b but do the RX480 drivers provide an option to do that? FWIR only nvidia allow this kind of kludge in RGB?

Actually that link seems to imply that it would even be possible with HDMI 1.4, it's all pretty dang confusing and I haven't seen any option to output 4:2:2 RGB in the 13.12 AMD drivers.

There is an option in the Crimson driver where you can select 4:4:4, 4:2:2 or 4:2:0 and 8, 10, or 12 bits. You can only selects 8 bits for 4:4:4. This is the same for nVidia. It's the movie that will determine if 8, 10 or 12 bits are send to the TV/Projector.

[madshi]

Originally Posted by Magik Mark

Which NGU variant would you recommend that would perform similarly as of neddi3 64 neurons?

On a pure speed level, I think "NGA Anti-Alias High" should be faster than NNEDI3-64 (except on Polaris, maybe), while "Very High" might be slightly slower. But why don't you simply benchmark yourself?

If you're asking about quality instead of speed, that's what I actually want to hear from you!!

Originally Posted by har3inger

Did you change something in the NGU sharp scaler? It runs a smidgeon (5-10%) slower than it used to in 91.5.

Edit: Yep. Reverted to 91.5, and lost 3ms worth of render time from 36ms when using NGU high. Unfortunately, this is the difference between perfectly playable and stuttering mess for me. (91.6 is slow. 91.7 is similar, or even a tiny bit slower than 91.6)

I've changed the downscaling algo after doubling/quadrupling. Do you see the same slowdown if you do a straight 2x upscale without any downscaling involved? If yes, please do what I suggested here:

https://forum.doom9.org/showpost.php...ostcount=42902

Originally Posted by Neo-XP

It would be the same as the upscaling algo after doubling, as in v0.91.1.
Something like "same as upscaling after doubling algo". Would that be possible ?

Oh, I see. I'm not sure if anybody would understand that concept, though. I mean one dropdown box referring to another is kinda weird.

Originally Posted by Arm3nian

How does NGU quadrupling work with "refine the image after every 2x step / only once" options? If I have sharpen edges at 4.0, refine the image only once, and NGU on "double again" then it seems to be at 4.0. Same with direct quadruple. But if it's "refine the image after every step" then it looks like 8.0. And if NGU is on direct quadruple and I select refine after every step then it overrides it and uses NGU twice. So if you could select sharpen edges at 8.0 then you could use NGU direct quadruple with the same look as doubling twice and refining the image twice right now. But then I guess that would go up to 16 in the new version?

I can't really follow your thoughts. Where do you get the idea from that madVR would do 8.0 or 16.0 in your case? As you said yourself, if you select "refine after every step", quadrupling is turned off because it can't work in that case.

Originally Posted by Anima123

NGU-AA low consumes more GPU power than previous NGU pixart, the Intel Graphics 630 can hardly handle 720p -> 1024p @ 23.976 fps fluently, the playback stuttered from time to time, maybe due to throttling?

I've changed the downscaling algo after doubling/quadrupling. Do you see the same slowdown if you do a straight 2x upscale without any downscaling involved? If yes, please do what I suggested here:

https://forum.doom9.org/showpost.php...ostcount=42902

[Anima123]

Originally Posted by madshi

I've changed the downscaling algo after doubling/quadrupling. Do you see the same slowdown if you do a straight 2x upscale without any downscaling involved? If yes, please do what I suggested here:

https://forum.doom9.org/showpost.php...ostcount=42902

Before I do as what you're asking, need to tell that the rendering time goes from less than 10 ms up to crazy 3x ms average and 6x ms peak, when the stutter began. And after stuttering for some time, the rendering time would go back to normal.

With that kind of case, would it help to do the screen-shot, and when should I do it?

[huhn]

Originally Posted by HillieSan

There is an option in the Crimson driver where you can select 4:4:4, 4:2:2 or 4:2:0 and 8, 10, or 12 bits. You can only selects 8 bits for 4:4:4. This is the same for nVidia.

it's not that easy.

It's the movie that will determine if 8, 10 or 12 bits are send to the TV/Projector.

no.

[madshi]

Originally Posted by Anima123

Before I do as what you're asking, need to tell that the rendering time goes from less than 10 ms up to crazy 3x ms average and 6x ms peak, when the stutter began. And after stuttering for some time, the rendering time would go back to normal.

With that kind of case, would it help to do the screen-shot, and when should I do it?

Are the rendering times identical, when comparing NGU-AA vs NGU-pixart, except for those crazy 3x slowdowns? Or are the rendering times also slower when ignoring those 3x slowdowns?

[Anima123]

Originally Posted by madshi

Are the rendering times identical, when comparing NGU-AA vs NGU-pixart, except for those crazy 3x slowdowns? Or are the rendering times also slower when ignoring those 3x slowdowns?

Under 10 ms for both of your algorithms, except for the crazy period of time.

[madshi]

So the real problem for you are just the crazy slowdowns? No idea how to debug that, unfortunately... And you can reproduce it 100% that you get those slowdowns with NGU-AA, but not with NGU-pixart? Does it also happen if you just do a simple 2x upscale without any downscaling involved, and without 4x?