madVR - high quality video renderer (GPU assisted) [Archive] - Page 756

foozoor

5th May 2016, 22:00

The latest madVR build (v0.90.17) is from March 25.

It would be great to have some news about madshi and madVR on the forum.
However madshi is still active on his bug tracker.

Will keep this open for some future time. OSD will be redesigned at some point.
Right now I'm concentrated on algorithms and stuff, so no time for OSD at the moment.

:)

QBhd

5th May 2016, 22:02

Was there a resent update a week ago?
...

madVR has no auto-update. Not sure who/what is messing with your system, but something sounds sketchy to me

QB

70MM

6th May 2016, 00:51

madVR has no auto-update. Not sure who/what is messing with your system, but something sounds sketchy to me

QB
Seems strange to me, there is no one else in the house to play with my system!

How do I find out then what these "other" new settings are for on the sharpness page that wernt there before the update? Where do I go to get the explanations of the new settings?

oddball

6th May 2016, 01:17

I still have the issue with MadVR on Windows 10 with AMD. I can only run it in they way I normally run it by using DX11 mode. If I set it to anything else I get a blank video on my HDTV using FSE. If I turn off delay by number of frames in FSE settings it works. But then if I switch back to windowed mode on the primary display MPC-HC crashes/hangs with the busy circle (but I can still get into MadVR settings when this happens). Also if I enabled delay playback in DX11 mode MPC-HC just hangs with the busy circle as soon as it goes FSE on the HDTV.

I have tried rolling back drivers, MadVR and MPC etc with no effect. I am beginning to wonder if something changed in one of the Windows 10 updates recently to cause this. I never had this issue before I don't think.

Any advice?

Warner306

6th May 2016, 01:28

Seems strange to me, there is no one else in the house to play with my system!

How do I find out then what these "other" new settings are for on the sharpness page that wernt there before the update? Where do I go to get the explanations of the new settings?

Anti-bloating is the only new setting. This reduces the fatness created by line sharpeners. Not needed for crispen edges, which is already "skinny."

At 1080p -> 1080p, I like sharpen edges (0.5) + 50% AB + AR. Others may prefer crispen edges (0.5) + AR.

har3inger

6th May 2016, 02:21

The big theory is called Super-Resolution as I understand it and it uses informations from previous frames in order to increase details but I have yet to see it used IRL and especially disclosed by intel agencies so again either it's all a big theoretical myth or they refuse to share the results possibly due to NDA's. Not something that will happen in real-time using mVR anyway so case closed I guess, NNEDI3 is as good as it's gonna get for our GPU's.

One technique doesn't use previous frames, but different frames taken of the same subject under same lighting at similar angles. You sub everything up together for amazing effects. The technique doesn't really apply to video much, but hot damn if it doesn't perform black magic.

http://www.sciencedirect.com/science/article/pii/S0032063315003591

Otherwise, lots of video can be cleaned up if you know the artifacts to be from optics rather than from signal compression. Also entirely useless for movies, unless you want to make an anti-shaking camera filter.

http://blogs.adobe.com/jnack/2011/10/adobe-demos-amazing-deblurring-tech-new-video.html

On topic: I've been trying out the super-xbr antibloat option a bit more. For low-res sources (480p or lower), sxbr AB is fantastic. It has a drastic effect on reducing line bloat. For 720p and higher, its effects are less visible and don't always outweight the weird ringing artifacts it causes. The performance cost of AB vs no AB is also pretty substantial at the moment, but I can't see why that couldn't change in the future.

70MM

6th May 2016, 02:22

Anti-bloating is the only new setting. This reduces the fatness created by line sharpeners. Not needed for crispen edges, which is already "skinny."

At 1080p -> 1080p, I like sharpen edges (0.5) + 50% AB + AR. Others may prefer crispen edges (0.5) + AR.

Thank you Warner...
So at 1080p > 1080p you never use sharpen edges + crispen edges at the same time?

Asmodian

6th May 2016, 05:29

Anti-bloating is the only new setting. This reduces the fatness created by line sharpeners. Not needed for crispen edges, which is already "skinny."

At 1080p -> 1080p, I like sharpen edges (0.5) + 50% AB + AR. Others may prefer crispen edges (0.5) + AR.

I hope you don't mind if I steal your description of anti-bloat? :D

Uoppi

6th May 2016, 08:19

Georgel

6th May 2016, 15:20

1) Is there consensus if any of the new features (like anti-bloating) is "needed" when only using moderate superres (chroma/luma at 1 or max 2) and no other sharpening? (For luma I use NNEDI3 at 32 or above whenever possible, but for a couple of profiles I've had to resort to Super XBR.)

2) Would you use superres at 1 or 2 for a) SD -> 1080p (mostly good quality DVD rips) and b) 720p -> 1080p? I would think 2 for a) and and 1 for b) and not the other way around, right?

Just asking for some quick tips because I have zero hours left in my tweaking quota (says wife) after having spent tens of hours with madVR already, LOL. After upgrading to v0.90.17 I haven't touched any of the settings and things look OK to me with the new settings at default.

I do not fully understand your question.

A consensus can only be given for users having the exact same setup and videos usually. This being said, the consensus is that AB helps a lot and makes things better, but everyone found their own setting that works for them.

I have a hard time reading the rest of the question.

Super res has it's uses too, usage of AB is good for all algorithms so far.

regiregi22

6th May 2016, 17:08

Betroz

6th May 2016, 18:25

Hi guys, is there any guide regarding the recommended settings on madvr latest versions?
I have a GTX 970.

With my GTX 970 card I have :

- Chroma upscaling : NNEDI3 128
- Image downscaling : SSIM 2D 100%, AR relaxed, LL
- Image doubling : Double & quad Luma & Chroma at always super-xbr, anti-bloating 25
- Image Upscaling Jinc AR
- Upscaling Refinement : Sharpen Edges 4.0, Enhance Detail 1.0, Lumasharpen 0.5, AR, SuperRes 3 AR, anti-bloat 50%.

Warner306

6th May 2016, 19:00

Thank you Warner...
So at 1080p > 1080p you never use sharpen edges + crispen edges at the same time?

They both do the same thing, so those shaders would be competing with each other. It is better to choose one or the other, not both.

Warner306

6th May 2016, 19:01

Hi guys, is there any guide regarding the recommended settings on madvr latest versions?
I have a GTX 970.

It's like two years ago since I haven't updated madvr/mpc-hc, seems like quite some options have been added.

Try this link (http://forum.kodi.tv/showthread.php?tid=259188&pid=2238974#pid2238974) from the guide in my signature.

Warner306

6th May 2016, 19:02

I hope you don't mind if I steal your description of anti-bloat? :D

Sure, I'm not sure if it is 100% accurate. But you can use it. I've used many of your descriptions.

Warner306

6th May 2016, 19:05

With my GTX 970 card I have :

- Chroma upscaling : NNEDI3 128
- Image downscaling : SSIM 2D 100%, AR relaxed, LL
- Image doubling : Double & quad Luma & Chroma at always super-xbr, anti-bloating 25
- Image Upscaling Jinc AR
- Upscaling Refinement : Sharpen Edges 4.0, Enhance Detail 1.0, Lumasharpen 0.5, AR, SuperRes 3 AR, anti-bloat 50%.

Too much sharpening for my tastes. Lots of redundant shaders.

huhn

6th May 2016, 19:17

sharpen edges and crisped edge (thin edges and enhance details) are not designed to work on the same part of the image so you can use them all together at the same time.

http://forum.doom9.org/showpost.php?p=1745696&postcount=34110

Uoppi

6th May 2016, 19:17

I do not fully understand your question.

A consensus can only be given for users having the exact same setup and videos usually. This being said, the consensus is that AB helps a lot and makes things better, but everyone found their own setting that works for them.

I have a hard time reading the rest of the question.

Super res has it's uses too, usage of AB is good for all algorithms so far.

OK, I'll leave AB disabled because video looks good already and I've got minimal time to go through testing all AB permutations (that's why I was asking for some starting points).

As for the second question, I don't know how to word it better. In essence, would one generally use higher strength of superres the bigger the upscaling factor or vice versa. If the answer is the classic "use your eyes", I'll just leave superres at 1 for everything and stay happy. :D

aufkrawall

6th May 2016, 19:46

Betroz

6th May 2016, 22:17

Too much sharpening for my tastes. Lots of redundant shaders.

Ok, what do you recommend that I change/remove?

Warner306

6th May 2016, 22:21

The higher the upscaling factor, the less SuperRes strength you need to see a distinct effect of it.
Matter of personal taste if you want to have a stronger or weaker effect. High SuperRes strength looks less artificial than other sharpen filters, however you can get problems with aliasing, mostly when the source is bad.

I find the opposite; the higher the scaling factor, the higher the value of SuperRes is appropriate because of how soft the image becomes.

Warner306

6th May 2016, 22:26

Ok, what do you recommend that I change/remove?

At 2x (1080p -> 2160p), I would use SuperRes (3) without anti-bloating because SuperRes really doesn't bloat all that much if the image is not oversharpened. Nor does it need anti-ringing. enhance details (1.0) is another shader that is not redundant. To me, SuperRes at the right strength provides plenty of sharpening on its own.

sharpen edges and LumaSharpen are likely pushing the image towards being oversharp and are redundant. If you like the image really sharp, I might add sharpen edges with anti-bloating but not LumaSharpen on top of that.

ryrynz

6th May 2016, 23:25

With my GTX 970 card I have :

- Chroma upscaling : NNEDI3 128

I've said it before and I'll say it again, good lord why?
Did you read anything in the forum before setting that, or even do a visual comparison between 16 or 32 neurons vs 128? Such a waste of resources.

aufkrawall

6th May 2016, 23:36

I find the opposite; the higher the scaling factor, the higher the value of SuperRes is appropriate because of how soft the image becomes.
I don't see where your latter sentence would contradict my statement.

Warner306

7th May 2016, 02:33

I don't see where your latter sentence would contradict my statement.

You're right; both are true.

Uoppi

7th May 2016, 06:27

You're right; both are true.

So if I understood you and aufkrawall correctly:

The effect of SuperRes may be easier to spot when scaling SD -> 1080p vs. 720p -> 1080p using an identical SuperRes setting (1, 2, 3 etc.)

But because of the softness of SD, a higher SuperRes setting may be required to counter the softness (but with the increased the risk of artefacts). And 720p on the other may become oversharp, so a lower setting might be more appropriate.

Notice the "mays" and "mights". I do realize sharpening is about personal preference. I have sharpness turned to 0 on my Samsung TV; my eyes are sensitive to too much sharpness and start to "hurt".

That's why I tend to use SuperRes at 1, use no other sharpening and direct my available resources towards NNEDI3 processing. Also when downscaling, I use lower values, mostly bicubic 75% and for UHD -> 1080p also 2D SSIM 25%.

Betroz

7th May 2016, 08:28

I've said it before and I'll say it again, good lord why?
Did you read anything in the forum before setting that, or even do a visual comparison between 16 or 32 neurons vs 128? Such a waste of resources.

If rendering time is below 30ms, why not? All my content is 23 fps, so my GTX 970 can handle it. Image doubling at NNEDI3 kills performance more.

huhn

7th May 2016, 08:32

If rendering time is below 30ms, why not? All my content is 23 fps, so my GTX 970 can handle it. Image doubling at NNEDI3 kills performance more.

but image doubling will have an effect on image quality unlike nnedi3 chroma at higher neurons.

Betroz

7th May 2016, 08:51

but image doubling will have an effect on image quality unlike nnedi3 chroma at higher neurons.

I know, but super-xbr anti-bloat 25% is good for image doubling too, and the fans on my 970 card doesn't spin up as much as they do with NNEDI3. A good compromise I think. Chroma Upscaling 128 is nice to have with 1080p movies that run 1:1 (no image scaling).

QBhd

7th May 2016, 08:53

but image doubling will have an effect on image quality unlike nnedi3 chroma at higher neurons.

Exactly, spare those Chroma Neurons and spend them on NNEDI3 Image Doubling. I only bump my Chroma high when there is nothing left to give on Image. When I set up profiles, I always start with Image Doubling and see how high it can go, then if there is room to spare do I add to Chroma. I am sure Betroz could easily drop to 32 Chroma and get some NNEDI3 goodness out of Image Dougling

I know, but super-xbr anti-bloat 25% is good for image doubling too, and the fans on my 970 card doesn't spin up as much as they do with NNEDI3. A good compromise I think. Chroma Upscaling 128 is nice to have with 1080p movies that run 1:1 (no image scaling).

If the fans are not spinning up for Chroma 128, I highly doubt that it's even being used... Can you post a Screen shot of the OSD with that particular setup?

QB

Betroz

7th May 2016, 09:48

If the fans are not spinning up for Chroma 128, I highly doubt that it's even being used... Can you post a Screen shot of the OSD with that particular setup?
QB

It is, but Chroma 128 doesn't tax my card as hard as Image doubling, at least not when super-xbr is used for doubling/quad. I do have tried diffferent settings... ;)

Sunset1982

7th May 2016, 14:05

wanezhiling

7th May 2016, 14:55

1070 is even stronger than titan x.

Uoppi

7th May 2016, 15:02

I currently use Super XBR 75 for doubling with some profiles where I can't afford NNEDI3. I'm considering whether I should try switching to Super XBR anti-bloat 25 because many seem to like it.

Any pointers as to the strengths/weaknesses of each, i.e. with what material would I best see their benefits or drawbacks? I've understood they should look pretty close.

BluesFanUK

7th May 2016, 15:34

just read the new presentation of the new geforce cards. 1070 as strong as 980ti and 1080 stronger than titan... seems we will get a good jump in performance. Now lets see what amd got and then we'll see how much benefits madvr will get from this more powerfull gpu's when upsaling 1080p to 4k on my machine. :devil:

Neither card is a substantial enough upgrade over a 980Ti to get more performance out of MadVR. Nvidia have provided no real proof other than one's apparently the equivalent of a Titan X and the other can out perform 980SLI.

The real performance cards are due out next year. The next Ti should allow some serious high settings for pushing 1080p@60 on to a 4K screen.

Speaking of which, i'm actually glad I downgraded my monitor back to 1080 (the 4K one I had was too hazy for my liking). I can use pretty much any setting I like now with my 980Ti and I can barely notice any difference. 4K is a gimmick. Beneficial to gamers perhaps, but you'd need a good 60-70" screen to appreciate the difference for video watching.

On another note, IMO, NNEDI3 is a complete waste of resources and I think Super-XBR looks miles better. I had all the settings ramped up on one of my profiles for 720p using NNEDI3, GPU was at about 99% with no dropped frames and no real improvement on screen. Changed to Super-XBR, less resource heavy and sharpens my images enough to look 1080. Guess it really is down to personal taste.

nevcairiel

7th May 2016, 15:38

Neither card is a substantial enough upgrade over a 980Ti to get more performance out of MadVR. Nvidia have provided no real proof other than one's apparently the equivalent of a Titan X and the other can out perform 980SLI.

The current estimates (from their presentation information and the core/clock ratio) go around 15-20% over a TitanX/980Ti at stock, with easy 20% OC headroom, which personally I find substantial. Only benchmarks will reveal the real information either way however, and anything else is just speculation, both positive and negative.

Ver Greeneyes

7th May 2016, 16:01

1070 is even stronger than titan x.They said that at the end, but I think that was taking their VR viewport trick into account (since that adds another ~40% performance in VR). Still, the 1070 won't exactly be a pushover either.

Stereodude

7th May 2016, 16:16

The current estimates (from their presentation information and the core/clock ratio) go around 15-20% over a TitanX/980Ti at stock, with easy 20% OC headroom, which personally I find substantial. Only benchmarks will reveal the real information either way however, and anything else is just speculation, both positive and negative.
But who knows if the new architecture will actually be faster or not for madVR. Nvidia pretty much had AMD's number in the last gen when it came to gaming performance, but AMD's cards offered better madVR performance. A Nvidia card that was effectively the equal in gaming of a given AMD model had considerably higher rendering times with the same madVR options.

nevcairiel

7th May 2016, 16:33

But who knows if the new architecture will actually be faster or not for madVR. Nvidia pretty much had AMD's number in the last gen when it came to gaming performance, but AMD's cards offered better madVR performance. A Nvidia card that was effectively the equal in gaming of a given AMD model had considerably higher rendering times with the same madVR options.

The Pascal architecture doubles the parallelism by making SM units smaller and doubling their count (64 vs 128 cores per SM, compared to Maxwell, Kepler had 192 cores per SM for comparison), which should help any tasks that are incredibly parallel, which image scaling should fall under. But in the end we won't know until we know.

har3inger

7th May 2016, 18:27

BluesFanUK, try out bilateral chroma with superchromares 2. I personally prefer it to all other algorithms (including nnedi3), and it costs way less than nnedi3.

sat4all

7th May 2016, 19:54

fellas, what's your debanding settings for bluray's remux, is it better to keep it off?

Stereodude

7th May 2016, 20:34

The Pascal architecture doubles the parallelism by making SM units smaller and doubling their count (64 vs 128 cores per SM, compared to Maxwell, Kepler had 192 cores per SM for comparison), which should help any tasks that are incredibly parallel, which image scaling should fall under. But in the end we won't know until we know.
I'm hopeful they're better this generation. I will probably replace the GTX 650 in my desktop PC with a GTX 1070. Getting big jump in madVR performance would be appreciated since the GTX 650 basically can't do anything beyond bicubic driving my 2550x1440 display.

fellas, what's your debanding settings for bluray's remux, is it better to keep it off?
I use it on low, med during fades myself. I use that setting for all content.

aufkrawall

7th May 2016, 22:56

fellas, what's your debanding settings for bluray's remux, is it better to keep it off?
BDs are 8 bit and often badly mastered, so there's a good reason to use deband filter at least at low setting if you don't mind very minimal detail loss (I doubt it can be noticed without concentratedly starring at image comparisons.).
When you use fade-detection, I recommend not to disable the performance/quality trade-off option for it, as you will need longer CPU and GPU queues else.

huhn

7th May 2016, 23:09

"banding" let's call it steps i guess could can be intentional like in the shinsekai yori ED.

so >I< i don't have it on by default.

SpoCk0nd0pe

8th May 2016, 02:39

The Pascal architecture doubles the parallelism by making SM units smaller and doubling their count (64 vs 128 cores per SM, compared to Maxwell, Kepler had 192 cores per SM for comparison), which should help any tasks that are incredibly parallel, which image scaling should fall under. But in the end we won't know until we know.

Yes, it seems NVidia has been trying to fix their shader architecture starting with Maxwell. We have to see if Pascal can close the gap to AMD.
Is half precision enough for NNEDI3? If yes, Pascal should be very fast at that.

CarlosCaco

8th May 2016, 06:47

I find a strange but good behave on madvr
When i m downscaling a 1080 source i getting some great extra performance if set chroma to reconstruction soft, all other algorithms have greater performance hits when downscaling 1080 to 1366x768 even mitchel nevali is slower than reconstuction soft on this scenario, getting almost less 10 ms using reconstruction soft, the same thing if uncheck don t use linear light for dithering i get better render times that if that option is checked... Someone know why this?

huhn

8th May 2016, 09:16

could be GPU "powerstate".

Uoppi

8th May 2016, 11:56

SSIM downscaling is not working when deinterlacing - it always reverts to Bicubic150. It says so in the OSD and also switching between 1D and 2D makes no difference in rendering times.

Looking a some older posts, it looks like at one point SSIM may not have been working at all after doubling (?) but it does work now apart from when deinterlacing (dxva) is going on. I'm just wondering if this behavior is to be expected or not.

Piyoko

8th May 2016, 12:51

A question regarding HDR, why does it currently not work and how would it theoretically work in the future? If I have file encoded in HDR and I'm connected to an HDR TV where does the information that it's an HDR stream get stuck?

Let's assume HDR10 and not the more complex Dolby Vision. As I understand it all of the characteristics are globally defined just like a gamma curve so theoretically the GPU-makers could just provide an HDR output toggle in the driver settings and it would work correctly even without passing through any metadata from the player? (Though if you had a windowed player and enabled that mode the rest of your desktop would probably scorch your eyeballs.)