madVR - high quality video renderer (GPU assisted)

[pose]

madshi, v0.91.7 works as fast as v0.91.5 for me on RX470. Image upscaling doubling menu looks a bit confusing. NGU Anti-Alias looks great. :) I noticed that some content which NGU sharp quadruples, NGU Anti-Alias is doubling instead with the same "let madvr decide" settings. Is that normal behavior?

[Anima123]

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?

Just found that the AR with the downscaling is the one to blame. Tried bicubic 60 with or without AR in downscaling make huge difference. The bicubic 60 without AR is as fast as old versions, when AR is in play, the stuttering began to present.

Edit: Quality-wise, NGU-AA is quite impressive even at low level with natural movies. The prominent one is that the edge of human faces is sharp and non-aliased with NGU-AA, which is pleasing to watch.

[madshi]

Just found that the AR with the downscaling is the one to blame. Tried bicubic 60 with or without AR in downscaling make huge difference. The bicubic 60 without AR is as fast as old versions, when AR is in play, the stuttering began to present.
Ah, interesting. Well, I've changed the AR filter from "strict" to "relaxed", which should result in a sharper looking result. But the "relaxed" AR filter is slightly slower. If you want to go back to "strict", you can setup the downscaling algo in the "image downscaling" section with "strict" AR, and then in the "image upscaling" section you can use the option "use image downscaling settings". That should get you back to NGU-pixart performance levels, on the cost of some sharpness.

Edit: Quality-wise, NGU-AA is quite impressive even at low level with natural movies. The prominent one is that the edge of human faces is sharp and non-aliased with NGU-AA, which is pleasing to watch.
Glad to hear that! NGU-AA "low" has changed quite a bit compared to the first NGU pixart test builds. It now has somewhat less anti-aliasing capabilities than before, but looks much more natural with normal film content, while still being better at AA than NGU Sharp.

I noticed that some content which NGU sharp quadruples, NGU Anti-Alias is doubling instead with the same "let madvr decide" settings. Is that normal behavior?
Sounds weird. Can you make a screenshot of the OSD (Ctrl+J) and of your "image upscaling" settings page, please?

[andybkma]

madshi, after playing around with mVR v91.4 I realized that NGU-Low for Image Upscaling was the best I can do without stuttering and dropped frames on my laptop's GT650M. Which of the new mVR settings is comparable to the quality and load use of what seems to be the now defunct NGU-Low? Thanks much...

[madshi]

There's nothing defunct. NGU-Low still exists, just the downscaling after doubling got a bit slower. See one post above for more information about that.

[Anima123]

It would be nice and more logical to me to keep the downscaling AR strict for low.

Or a lot of people would be confused on this.

BTW, AR relax is not 'a bit' slower, you sure there're no bug there in latest changes? The sporadically stuttering is quite suspicious to me.

[madshi]

Have you tried my suggestion of "use image downscaling settings"? If so, do you get the same speed as with the pixart test builds? If you do, there doesn't seem to be a bug. Maybe your GPU simply doesn't like the relaxed AR downscaling algo, for some weird reason.

It's worth considering strict AR for "low" doubling quality. Or maybe even for all Anti-Aliasing quality levels, because Anti-Aliasing isn't as sharp to begin with. I'll consider it for the next build.

[andybkma]

There's nothing defunct. NGU-Low still exists, just the downscaling after doubling got a bit slower. See one post above for more information about that.

madshi, sorry but I can't find a "NGU-Low setting" inside the Image Upscaling options in the new mVR version 91.7. I upgraded from 91.4 to 91.7 with same settings and now instead of NGU-Low for Image Upscaling it automatically bumped me down to Lanczos 3. And on the right side under the new heading "Doubling" I only have options for NGU Anti Alias and NGU Sharp with another two NGU's greyed out. There is no option available to me for NGU-Low as I said....

[huhn]

they are renamed: https://forum.doom9.org/showpost.php?p=1799592&postcount=42866

[andybkma]

they are renamed: https://forum.doom9.org/showpost.php?p=1799592&postcount=42866

Trying to read between the lines here... am assuming new name for NGU-Low is NGU-Soft but unusable in 91.7?

[pose]

Sounds weird. Can you make a screenshot of the OSD (Ctrl+J) and of your "image upscaling" settings page, please?
https://imgur.com/a/X3Nzj

[tranfa]

it's not that easy.


no.

Not sure it's a subject that will bring us OT, but I'd be really grateful to have more clarity around these concepts and the chain needed...

For instance, coming to 4:4:4, I'm in a weird situation in which the .png file that Madshi graciously made available shows a clear 4:4:4 if read on the browser, and doesn't once passing through Potplayer + MadVR...I think it's an issue with the internal Potplayer decode, but I'm not 100% certain...

[madshi]

Trying to read between the lines here... am assuming new name for NGU-Low is NGU-Soft but unusable in 91.7?
You're confusing the algo name with the quality level. "NGU", "NGU Anti-Aliasing" and "NGU Sharp" are all algorithm names. "Low" is the quality level. In v0.91.7, "NGU" was renamed to "NGU Sharp". There's an extra dropdown box which allows you to select the quality level.

https://imgur.com/a/X3Nzj
Crazy! Can you find any pattern to when NGU AA is used vs Sharp?

[andybkma]

You're confusing the algo name with the quality level. "NGU", "NGU Anti-Aliasing" and "NGU Sharp" are all algorithm names. "Low" is the quality level. In v0.91.7, "NGU" was renamed to "NGU Sharp". There's an extra dropdown box which allows you to select the quality level.


Thanks madshi, ya that's exactly what happened. And you are as patient as a saint :-)

However new NGU Sharp-Low is using much more gpu load/higher gpu temps than the previous NGU-Low that was in 91.4... what can I do about this? All I did was change the Image Upscaling in the new version 91.7 to NGU Sharp Low that was previously NGU-Low in 91.4. Intentional to make it much more intensive?

[madshi]

And you are as patient as a saint :-)
I'm trying to. But in order to not test my any further, why don't you read the last 20 posts in this thread or so? :p

[pose]

Crazy! Can you find any pattern to when NGU AA is used vs Sharp?
It happens with every video with 4:3 aspect ratio and resolution 640x480+.

Edit. Actually x480 seems to be the culprit.

[huhn]

Not sure it's a subject that will bring us OT, but I'd be really grateful to have more clarity around these concepts and the chain needed...

For instance, coming to 4:4:4, I'm in a weird situation in which the .png file that Madshi graciously made available shows a clear 4:4:4 if read on the browser, and doesn't once passing through Potplayer + MadVR...I think it's an issue with the internal Potplayer decode, but I'm not 100% certain...

feel free to make a new thread but i usually don't care about potplayer and i don't have a clue how potplayer is working internally. you may just scale it i don't know.

[andybkma]

I'm trying to. But in order to not test my any further, why don't you read the last 20 posts in this thread or so? :p

Lol, I've been reprimanded by madshi :p

Am looking forward to your new NGU versions soft and extra extra soft for old laptops haha thanks again for your help today and have great night :-)

[madshi]

It happens with every video with 4:3 aspect ratio and resolution 640x480+.
I'm not sure if I have such a video at hand, could you maybe upload a very short/small cut of such a video?

Am looking forward to your new NGU versions soft and extra extra soft for old laptops
NGU Soft will have the same speed as NGU Sharp.

[pose]

I'm not sure if I have such a video at hand, could you maybe upload a very short/small cut of such a video?
I think i figured it out. Anything with vertical resolution of 450+ pixels gets doubled by NGU AA and quadrupled by NGU sharp. Anything below x451 gets quadrupled by both NGU AA and NGU sharp. Aspect ratio doesnt matter. You can check it by creating image with desirable resolution in paint and then changing file's extension to .mkv.

[madshi]

I think i figured it out. Anything with vertical resolution of 450+ pixels gets doubled by NGU AA and quadrupled by NGU sharp. Anything below x451 gets quadrupled by both NGU AA and NGU sharp. Aspect ratio doesnt matter. You can check it by creating image with desirable resolution in paint and then changing file's extension to .mkv.
I don't like the idea of spending hours, trying different image resolutions and different aspect ratios, in the desparate attempt to reproduce the problem. So once more: Could you please upload a small chunk of a video with which you can reproduce the problem?

[pose]

I don't like the idea of spending hours, trying different image resolutions and different aspect ratios, in the desparate attempt to reproduce the problem. So once more: Could you please upload a small chunk of a video with which you can reproduce the problem?
I mean i already tested everything, but sure, here you go samples.zip (https://dl.dropboxusercontent.com/u/19076649/samples.zip).

[madshi]

I mean i already tested everything, but sure, here you go samples.zip (https://dl.dropboxusercontent.com/u/19076649/samples.zip).
Which of those should I use to reproduce the problem? I don't need 20 files, just one with which I can reproduce the problem. Reproducing is the key.

One thing that irritates me right now is that in your 2 OSD screenshots, *everything* (source resolution, AR and scaling factor) seems to be identical, except the algorithms used by madVR. That doesn't really fit to your findings that the problem depends on the video resolution, does it?

Can you tell me under which circumstances exactly the problem occurs? Right now I have trouble reproducing it at all.

[pose]

Which of those should I use to reproduce the problem? I don't need 20 files, just one with which I can reproduce the problem. Reproducing is the key.

One thing that irritates me right now is that in your 2 OSD screenshots, *everything* (source resolution, AR and scaling factor) seems to be identical, except the algorithms used by madVR. That doesn't really fit to your findings that the problem depends on the video resolution, does it?

Can you tell me under which circumstances exactly the problem occurs? Right now I have trouble reproducing it at all.
Try anything x451. NGU AA will double while NGU sharp will quadruple. Try anything below below x451. Both algorithms will quadruple (NGU AA double twice). My monitor resolution is 1080p.

Edit. Followup example (https://i.imgur.com/zNc4Sy2.jpg). I think we maybe misunderstood each other because of my broken English or something. The original question was about "let madvr decide" logic for NGU AA and NGU sharp with some video files. If i chose AA it will double image. If i choose NGU sharp it will quadruple same image.

[MisterXDTV]

I have a question, guys: how can I make MadVR the default video renderer for DirectX/Avisynth?

I am trying to encode an HDR sample to SDR and I would like to use the MadVR conversion for the encoding...

How can I do that?

[sneaker_ger]

madVR doesn't have any output pin so you can't use it for encoding. (I'm ignoring screen capturing, that would be insane.)

[Damien147]

Is there any conflict/benefit by using YV12 chroma upsampling shader with NGU SHARP (med)?

[MisterXDTV]

madVR doesn't have any output pin so you can't use it for encoding. (I'm ignoring screen capturing, that would be insane.)

So right now, how can you encode/convert an HDR clip to SDR?

[jmonier]

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.

As I said in my edit, I was premature in blaming the problem on a difference between 91.5 and 91.6. It was, instead, something that I've been seeing for some time with different versions. It just pops up from time-to-time. It seems that it will start when I start up the player (Zoomplayer) and continue until I exit the player. Then it may or may not show up the next time.

I'll try to get screenshots of each condition when they occur but my recollection is that the only difference is in the times (and I do have "showrendersteps" enabled).

I often don't know that it has happened because I'm trying out different settings and it just looks like these are the times for that setting. Then later I look at the times again during a more extensive test and they are less or more so it's hard to keep track of what is happening.

[madshi]

Try anything x451. NGU AA will double while NGU sharp will quadruple. Try anything below below x451. Both algorithms will quadruple (NGU AA double twice). My monitor resolution is 1080p.

Edit. Followup example (https://i.imgur.com/zNc4Sy2.jpg). I think we maybe misunderstood each other because of my broken English or something. The original question was about "let madvr decide" logic for NGU AA and NGU sharp with some video files. If i chose AA it will double image. If i choose NGU sharp it will quadruple same image.
I understood your original report in that way that you selected NGU Sharp but madVR used NGU Anti-Aliasing instead, or maybe the other way round. So that's not really what you're saying?

NGU Sharp has direct quadruple ability which NGU Anti-Aliasing does not. Direct quadrupling is very fast (much faster than doubling twice), so if you select NGU Sharp, and if you have a scaling factor only slightly larger than 2.0x, then NGU Sharp will activate quadrupling, because it's not much slower than doubling. But when using NGU Anti-Aliasing, it will only double in that situation because the only other alternative would be to double twice, which is a LOT slower.

Does that explain what you're seeing?

Is there any conflict/benefit by using YV12 chroma upsampling shader with NGU SHARP (med)?
The chroma upsampling shader is stone age, it shouldn't be used together with any of the madVR chroma upscaling algorithms. It just makes the chroma channel softer.

So right now, how can you encode/convert an HDR clip to SDR?
tritical at some point released a t3dlut filter. Maybe you could use a DisplayCAL created HDR -> SDR conversion 3dlut with t3dlut? But I don't really know if that would work.

As I said in my edit
FWIW, editing a post after I already replied to it usually means I don't ever see/notice the edit, because the forum doesn't send notifications about edited posts. It's better to write a new post in that situation.

[Georgel]

Glad to hear that! NGU-AA "low" has changed quite a bit compared to the first NGU pixart test builds. It now has somewhat less anti-aliasing capabilities than before, but looks much more natural with normal film content, while still being better at AA than NGU Sharp.


I see now!

This might be why I noticed that there as a small difference between NGU AA and Pixart early samples.

At any rate, I think it looks good and should work just fine with most materials.

Comparing NNEDI at 256 neurons and NGU AA at very high, NGU AA looks a bit better, some edges are even cleaner / more natural while it doesn't seem to introduce any new ringing or any other artifact compared to NNEDI.


The image looks crisp, the corners are rounded and I actually prefer NGU AA for RL content as well, not only anime.

[sneaker_ger]

So right now, how can you encode/convert an HDR clip to SDR?
Take a look at fmtconv (https://forum.doom9.org/showthread.php?t=166504).

[Dreamland]

I can't set NGU "Standard settings" on "Chroma Upscaling" , automatically jump on NGU"Sharp settings"
https://t9.pixhost.org/thumbs/62/38076223_appunti01.jpg (https://pixhost.org/show/62/38076223_appunti01.jpg)
is my problem?
thx a lot for great work, Madshi !

[madshi]

Comparing NNEDI at 256 neurons and NGU AA at very high, NGU AA looks a bit better, some edges are even cleaner / more natural while it doesn't seem to introduce any new ringing or any other artifact compared to NNEDI.

The image looks crisp, the corners are rounded and I actually prefer NGU AA for RL content as well, not only anime.
Glad to hear that!

I can't set NGU "Standard settings" on "Chroma Upscaling" , automatically jump on NGU"Sharp settings"
That's as intended, because "NGU Standard" and "NGU Soft" are not available yet. They will be introduced soon in a future version.

[pose]

So that's not really what you're saying?

No... Sorry.
NGU Sharp has direct quadruple ability which NGU Anti-Aliasing does not. Direct quadrupling is very fast (much faster than doubling twice), so if you select NGU Sharp, and if you have a scaling factor only slightly larger than 2.0x, then NGU Sharp will activate quadrupling, because it's not much slower than doubling. But when using NGU Anti-Aliasing, it will only double in that situation because the only other alternative would be to double twice, which is a LOT slower.
Now i understand the concept and it makes sense but now the thing is when everything under "doubling" is set to "let madvr decide" there is literally 0 difference between two in performance in my tests when quadrupling is forced. Here is example (https://i.imgur.com/KGxOLYR.jpg). Ive tested many videos of different resolutions. It happens because madVR decides to quadruple chroma compared to NGU AA. However, if chroma is manually set to normal for both, then Sharp is like 30% faster than AA.

[madshi]

Now i understand the concept and it makes sense but now the thing is when everything under "doubling" is set to "let madvr decide" there is literally 0 difference between two in performance in my tests when quadrupling is forced.
Which is as expected. NGU Sharp and NGU Anti-Aliasing are very near in performance. NGU AA is maybe a tiny little bit slower, but the difference is really small.

The key difference is that I can do "direct quadrupling" (meaning 400% zoom in one rendering step) with NGU Sharp, but I can't do that with NGU Anti-Aliasing at the moment. Which is why NGU Sharp is more eager to (direct) quadruple than NGU AA is, because NGU AA would have to double twice, which is very slow.

Here is example (https://i.imgur.com/KGxOLYR.jpg). Ive tested many videos of different resolutions. It happens because madVR decides to quadruple chroma compared to NGU AA. However, if chroma is manually set to normal for both, then Sharp is like 30% faster than AA.
"Normal" chroma quality comes practically for free when using NGU Sharp, while it doesn't come for free when using NGU AA. So when using "normal" chroma quality, NGU Sharp has a speed advantage over NGU AA.

I'm not sure right now: Is everything cleared up now, or do you think there's still a bug somewhere?

[pose]

Which is as expected. NGU Sharp and NGU Anti-Aliasing are very near in performance. NGU AA is maybe a tiny little bit slower, but the difference is really small.

The key difference is that I can do "direct quadrupling" (meaning 400% zoom in one rendering step) with NGU Sharp, but I can't do that with NGU Anti-Aliasing at the moment. Which is why NGU Sharp is more eager to (direct) quadruple than NGU AA is, because NGU AA would have to double twice, which is very slow.


"Normal" chroma quality comes practically for free when using NGU Sharp, while it doesn't come for free when using NGU AA. So when using "normal" chroma quality, NGU Sharp has a speed advantage over NGU AA.

I'm not sure right now: Is everything cleared up now, or do you think there's still a bug somewhere?
No, no. No bugs. I am still confused as of why AA is not as eager to quadruple (double twice) as NGU Sharp when using default settings (let madvr decide), because performance ends up to be the same. Even though you explained me everything. Maybe there will be some difference on a 4K display and different content, i dunno. Just some food for thought.

[madshi]

No, no. No bugs. I am still confused as of why AA is not as eager to quadruple (double twice) as NGU Sharp when using default settings (let madvr decide), because performance ends up to be the same. Even though you explained me everything. Maybe there will be some difference on a 4K display and different content, i dunno. Just some food for thought.
It's possible that in some situations madVR thinks that direct quadrupling is possible with NGU Sharp and consequently activates quadrupling, because it's so fast, only to find out later that direct quadrupling can't be used, after all, and thus doubles twice instead. That would actually classify as a little bug, maybe. However, I may decide to remove "doubling twice" in a future version completely (not decided yet, I'll ask for feedback on that when all the NGU algos can do direct quadrupling), so I'm not sure if I want to analyze this in detail right now.

[Werewolfy]

I only ever included it for "educational capacity", as you say. I don't think it's useful for anything else, even for pixart, unless you absolutely need to check every single pixel separately in pixart, but why you need that? Check out this one:

http://screenshotcomparison.com/comparison/202526

Why would you ever want nearest neighbor, if you can get the quality shown in that screenshot comparison, when using madVR?


I'm confused. You're linking to some algorithms which distort the pixels but you say you don't want distored-shaped pixels?

Why exactly do you need pixels to be exactly the way nearest neighbor produces them? It seems weird to me. I can understand that you don't want blurry, but you don't need to have blurry.


Huh?

http://screenshotcomparison.com/comparison/202526

To my eyes nearest neighbor is not sharper in that screenshot comparison. It's just dramatically more aliased.

To be honest I didn't notice that nearest neighbor was missing because I use it very rarely but I like using it when I watch old video games. I understand that any upscaling filters is better but when I play 2D video games, I never activate any upscaling filters and when I watch videos of these games, I don't want any upscaling filters because it's not the way I've played them.

So I would vote to add it again. If you want to remove one upscaling option, why not "spline"? Does someone really use it?

[BluesFanUK]

I wish I could use the new settings, they look good, but the noise it makes on the GPU (Palit 980t9 SJ) is unbearable. :(

[pose]

It's possible that in some situations madVR thinks that direct quadrupling is possible with NGU Sharp and consequently activates quadrupling, because it's so fast, only to find out later that direct quadrupling can't be used, after all, and thus doubles twice instead. That would actually classify as a little bug, maybe. However, I may decide to remove "doubling twice" in a future version completely (not decided yet, I'll ask for feedback on that when all the NGU algos can do direct quadrupling), so I'm not sure if I want to analyze this in detail right now.
I think in the end we still did not understand each other :D. Lets move on. Ill just use profile groups and force AA doubling twice. On a side note i want to say that NGU AA very high looks more natural and better than NNEDI256 with live action movies to me. Before i only tested with anime. Great job and thank you :rolleyes:

[Damien147]

The chroma upsampling shader is stone age, it shouldn't be used together with any of the madVR chroma upscaling algorithms. It just makes the chroma channel softer.


I use evr(cp) for high bitrate hevc and I like to set and forget.I will remove it.Thank you.

[madshi]

To be honest I didn't notice that nearest neighbor was missing because I use it very rarely but I like using it when I watch old video games. I understand that any upscaling filters is better but when I play 2D video games, I never activate any upscaling filters and when I watch videos of these games, I don't want any upscaling filters because it's not the way I've played them.

So I would vote to add it again. If you want to remove one upscaling option, why not "spline"? Does someone really use it?
I really don't understand how anybody could want nearest neighbor when looking at these results:

http://screenshotcomparison.com/comparison/202526

Anyway. If I could remove spline, that would solve the problem, but in my experience, whenever I remove something, somebody will cry. You see, I *thought* I would be safe removing such a dramatically bad algorithm like nearest neighbor, but even for that some users are crying. So you can bet with spline there would also be some.

Enough people have asked for nearest neighbor to return, so I'd like to do that, although I can absolutely not understand it, but I currently don't have the space necessary in the settings dialog, so right now I see no easy solution.

I wish I could use the new settings, they look good, but the noise it makes on the GPU (Palit 980t9 SJ) is unbearable. :(
Then you might want to get a better GPU. If you do some research, it seems that there are GPUs that have no audible fan noise nor any high pitched buzzing noise or coil whine. IIRC there was even a GPU manufacturer who was willing to accept RMAs on GPUs which produced coil whine. You can probably exchange the GPU via Ebay without too much additional cost.

Ill just use profile groups and force AA doubling twice.
Or modify the "activate quadrupling drop down", that should also work.

On a side note i want to say that NGU AA very high looks more natural and better than NNEDI256 with live action movies to me. Before i only tested with anime. Great job and thank you :rolleyes:
Glad to hear that!

[BluesFanUK]

Then you might want to get a better GPU. If you do some research, it seems that there are GPUs that have no audible fan noise nor any high pitched buzzing noise or coil whine. IIRC there was even a GPU manufacturer who was willing to accept RMAs on GPUs which produced coil whine. You can probably exchange the GPU via Ebay without too much additional cost.

Not sure it's worth it, unless the 1080 Ti turns out to be substantial enough to upgrade. Strangely it only seems to be an issue with MadVR, everything else is fine, games included.

Perhaps i'll just stick my headphones on. :sly:

[Werewolfy]

I really don't understand how anybody could want nearest neighbor when looking at these results:

http://screenshotcomparison.com/comparison/202526

Anyway. If I could remove spline, that would solve the problem, but in my experience, whenever I remove something, somebody will cry. You see, I *thought* I would be safe removing such a dramatically bad algorithm like nearest neighbor, but even for that some users are crying. So you can bet with spline there would also be some.

Enough people have asked for nearest neighbor to return, so I'd like to do that, although I can absolutely not understand it, but I currently don't have the space necessary in the settings dialog, so right now I see no easy solution.


Don't misunderstand me, those results are great. But it's just not the way I've played the game.

Does someone use Spline? Or could madshi remove it?

Well... it worths the try :)

[BluesFanUK]

Spline looks ok, doesn't use much resources either.

[huhn]

in the old madVR version i was using spline 3 ART instead of lanczos 3 which is not possible anymore after doubling.

removing spline 3 for NN is a bad joke...

[clsid]

MPC-HC has a nearest neighbor option for EVR-CP. That could be used instead of madvr for such videos.

[Xaurus]

There is a bug still in 91.7 where you have a random profile, it displays a high or higher rendering time than "frame" but in reality it is lower.
It seems to happen with high frame rate material more often than not.

For example, I am watching a 60 fps hls stream (twitch) and it has a 16.67 ms frametime.
When I load the stream the rendering time is shown as around 17 ms but there are 0 frame drops, should be plenty of frame drops if that was correct.
While streaming, I then go and change the chroma upscaling from "super-xbr" to some random algorithm, for excample "Cubic", hit apply, then select "super-xbr" again and hit apply.
Now the rendering time is showing roughly 11.5 ms which is in line with the GPU usage and zero frame drops.

This is with 720p/1080p -> 4k.

[Arm3nian]

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.

Because when you double, sharpen edges is applied, and when you double once more, sharpen edges is again applied. But when you quadruple in one step, sharpen edges is applied only once. I get this from the fact that doubling twice looks similar to quadrupling directly, in my sample. If anything, quadrupling directly looks better. But with sharpen edges and refine after every step turned on, doubling twice clearly looks sharper.