madVR - high quality video renderer (GPU assisted)

[Soukyuu]

Just played around with mpv and was wondering why all the videos were so laggy on pans even though no frames were dropped... Then remembered mpv does not have smooth motion. Scary, how fast one gets accustomed to some features and forgets they are even there.

[huhn]

Hello everyone, a friend of mine told me he is sure that our video card output can be 10 bit because hdmi 1.3 . To do so, i have to select ybcr 4:2:2 in the catalyst control panel. I'm not sure he is right , but i'm interested in it because my Sony panel are 10 bit native. He has a radiance too setted in that way , supported by Lumagen creator.
Another friend of mine has hacked an nvidia card to transform it in a Quadro card with 10 bit output. So is it true? Our video card are really 10 bit output?

Thanks

you don't need a quadro or some thing special to output 10 bit every amd/nvidia card can do this with directx 10/11.
but madvr can't do this so this only harms the picture or at least doesn't help at all.
to output 10 bit on Directx 10/11 you need to disable aero or it automatic gets disabled.

rgb 8 bit (if fully supported) should result in a way better picture then 10 bit 4:2:2 YCbCr with the current stand of MadVR.

a quadro can output 10 bit opengl that's it and it can do this with dvi.

[annovif]

I have found that In november 2013 , Madshi wrote this :

"In any case, I think if 10bit is supported at all, it's likely to be with DX11, but not with DX9, and madVR currently uses DX9 for presentation. So supporting 10bit is not an easy thing to do with madVR. I might have to move presentation to DX11. And that means: Not anytime soon.

I would be interested in hearing whether NVidia and AMD really do support 10bit output properly. Here's a way you could test it, if you're interested: You could render a FullHD grayscale. Left most pixel black, right most pixel white (or the other way round). Calculate every pixel in between to be a 10bit intermediate step between black and white. Don't use dithering, just straight 10bit rounded results. Then display the same thing with 10bit and 8bit. If the 10bit result looks noticably smoother, 10bit output seems to be working. "

Now Madshi support dx11 for dithering , has changed something for 10 bit output too? Someone tried Madshi requests? Thank you

[James Freeman]

to output 10 bit on Directx 10/11 you need to disable aero or it automatic gets disabled.

I'm 99.9% sure only Quadro can output 10-bit.

[annovif]

Incredible !!!!!! Amd 10 bit output with a registry hack .

http://forums.guru3d.com/showthread.php?t=384562
http://forums.anandtech.com/showthread.php?t=2358239

Incredible!!! AMD Firepro Modded Drivers (13.152.4-WHQL) for Radeon Cards 11/22/2013

http://www.overclock.net/t/1403233/amd-firepro-modded-drivers-13-152-4-whql-for-radeon-cards-11-22-2013

If there is Madshi madvr support , my real vw200 10 bit panel are just warming up the engine.

[nevcairiel]

I'm 99.9% sure only Quadro can output 10-bit.

Quadro can do 10-bit in OpenGL, but any (NVIDIA) consumer card can do 10-bit through Direct3D full-screen, its just that nothing ever uses that.
Don't even need a registry hack or modded quadro drivers.

See here for official confirmation:
http://nvidia.custhelp.com/app/answers/detail/a_id/3011/~/10-bit-per-color-support-on-nvidia-geforce-gpus

madshi said that he might support this in the future, but he doesn't provide ETAs for his features, so all we can do is wait.

[James Freeman]

Can we test this with "Enhanced Video Renderer" that comes with MPC-HC (it supports 10-bit output)?
I use DisplayPort + Dell U2410 (supports 10-bit input with DP).

Although I'm still pretty sure no other Nvidia card except Quadro can do true 10-bit output.

[leeperry]

Nvidia GTX 750ti OC yet? I'm thinking about replacing my current vid card with that one.
BTW, it's interesting to see that even though its TDP is advertized as 60W, most boards are capped to 38.5W (http://cryptomining-blog.com/1014-how-to-increase-the-geforce-gtx-750-ti-power-target-limit/) and you have to flash a modified BIOS in order to try uncapping it. It's measurable too. (http://cryptomining-blog.com/1141-a-bit-more-on-the-actual-power-usage-of-geforce-gtx-750-ti/) Makes you wonder why some manufacturers add a PCI-E power connector, maybe to be taken seriously.

I will wait for the giant die version of Maxwell, if they ever release it.
Yep, the more I think about it the more I agree with you....we need a twice faster version of the 750Ti, I guess they take their time to release it as they currently own the performance-per-watt market and they wanna milk the cow without cannibalizing Kepler.

all the videos were so laggy on pans even though no frames were dropped
I recently saw a 29.97fps movie (http://www.imdb.com/title/tt2166616/), what a relief to see smooth pans :eek:

They can whine all they want for 48fps but if movies were 29.97 this would already be a major breakthrough.

[DragonQ]

FWIW I'd have gotten an nVidia 750Ti if buying new. Seems to be easily the best card right now for non-hardcore gamers, who'd require a more heavy-duty GPU. The only reason I got a an R9 270 instead is because I got a great deal on a barely-used one.

[n3w813]

I ordered an EVGA 750Ti Superclock and will be here sometime this week. I hope madVR will be able to handle the following on 720p24 sources on a 1920x1080p display...

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: ED1, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)
3DLUT: Enabled

Currently, with a 650Ti on a 720p24 source, madVR can handle...

Image Doubling: NNEDI luma doubling 16 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: Ordered, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)
3DLUT: Enabled

[leeperry]

Sounds good, more feedback so I can avoid impulse purchases but FWIR it's barely as fast my 7850, efficiency drops once you overclock it & it gets hot too...even more so if you increase voltage and o/c it to 1.4Ghz.

BTW, is there a way we could set the number of queues in profiles please? Ideally I would like to have a profile for watching still pictures, I would use a renamed PotP .exe and would like to disable refresh rate rolling(as PotP's picture splitter forces 50Hz) & lower all queues to the minimum in order to avoid useless black frames. Watching pictures with mVR would become far more enjoyable :)

[hozes]

for 1080p24 bluray movies,

does chroma upscaling ( not chroma image doubling) nnedi3 32 netrons usefull or bc75ar is enough...most of the time i cannot see any difference, but i want not most of the time better, i want all time the best settings..( i cannot neutron64 for chroma upscaling, it causes dropped frames)

[6233638]

Sounds good, more feedback so I can avoid impulse purchases but FWIR it's barely as fast my 7850, efficiency drops once you overclock it & it gets hot too...even more so if you increase voltage and o/c it to 1.4Ghz.Well the point is that the 750 is efficient, not that it's fast.

While performance might be the same, it should be about 60% more efficient than a reference 7850 - but I would say that it's not worth replacing a card for efficiency reasons, only for performance ones.

Unless you need more performance now, I would wait for the higher-end Nvidia 8xx cards later this year, which should have similar efficiency but higher performance.

[leeperry]

Well, the only thing I'm afraid of the crazy summer heat on that o/c 7850 but a stock 750Ti will make feel frustrated(especially after madshi's latest update for AMD boards, brilliant! I can do 64X chroma + 256X luma on 576*340) and once overclocked they do get hot. My german's rusty but faster Maxwell's would appear to have been taped out (http://www.3dcenter.org/news/nvidias-maxwell-chips-gm204-gm206-legen-tape-out-fuer-die-28nm-fertigung-hin) and I gotta say that the MSI Twin Frozr 4 cooler looks very impressive(dual direct-touch heatpipe, giant fans, barely audible under load)...too bad it only comes with a 2y warranty. A $200ish faster TF4 750Ti sounds full of win :p

for 1080p24 bluray movies, does chroma upscaling ( not chroma image doubling) nnedi3 32 netrons usefull or bc75ar is enough...most of the time i cannot see any difference, but i want not most of the time better, i want all time the best settings..( i cannot neutron64 for chroma upscaling, it causes dropped frames)
To some ppl, enough is never enough(and sometimes it's even placebo ^^)...it all depends on your display, your expectations, whether you are willing to substantially increase your electricity bill for NNEDI, but rest assured that OCD'ed ppl aren't necessarily happier(they keep chasing a few percents here and there in the desperate hope that they willl add up in the end) so if BC75AR does it for ya then more time & money for less frivolous activities, winning :cool:

[artins90]

With my 7950 I can handle luma image doubling nnedi3 with 32 neurons but only for 24 fps content, any video recorded at 30 fps or higher is too much for my GPU so I have to disable it manually when needed to avoid dropped frames. I would really appreciate a setting to automatically disable image doubling when the fps is higher than a set value.
I hate the fact that I'm almost able to run at 30 fps, the render time needed for 30fps is 0.33ms while my upscaled 720p videos take 0.36-0.38 ms to render I'm so close :mad:

[hu1kamania]

I ordered an EVGA 750Ti Superclock and will be here sometime this week. I hope madVR will be able to handle the following on 720p24 sources on a 1920x1080p display...

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: ED1, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)
3DLUT: Enabled

I think you're going to be fine. With 720p25 I'm doing:
Image Doubling: NNEDI luma doubling 64 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Lanczos 4 AR
Deband: Low/High
Dithering: ED2, colored noise, change every frame
SmoothMotion: Disabled
3DLUT: None
Shader: Luma Sharpen post resize via mpc-hc

This is with +240 to the GPU, with a regular EVGA GTX 750 Ti. I'm even able to get 256 neurons with lower quality, 576x432 clips. 480p is just out of reach unfortunately, and I'm stuck with 128 luma neurons + 32 chroma doubling. With 29.97fps clips I'm stuck at 32 neurons, regardless of resolution. All in all I think it's a great card, and amazingly efficient.

[seiyafan]

Is Low/High the most popular setting for debanding?

[Mangix]

PSA: setting the flush settings to don't wait for everything results in fewer frame drops with NNEDI3 enabled.

edit: I just did a few more tests and found exclusive mode to be underwhelming. The upload and render queues were always low whereas windowed mode(both overlay and non) had queues that were relatively full(5-6 and maybe 2-3). This does result in occasional frame drops. No such issues with windowed modes. I thought exclusive mode was supposed to be better.

[Procrastinating]

AMD have released new drivers (http://support.amd.com/en-us/kb-articles/Pages/latest-catalyst-windows-beta.aspx), I wonder if there has been any improvements?

[QBhd]

With my 7950 I can handle luma image doubling nnedi3 with 32 neurons but only for 24 fps content, any video recorded at 30 fps or higher is too much for my GPU so I have to disable it manually when needed to avoid dropped frames. I would really appreciate a setting to automatically disable image doubling when the fps is higher than a set value.
I hate the fact that I'm almost able to run at 30 fps, the render time needed for 30fps is 0.33ms while my upscaled 720p videos take 0.36-0.38 ms to render I'm so close :mad:

That's what profiles are for as an example here is mine (I push my GPU to 90%+ for each and every one of these profiles):

if (srcHeight < 481) and (deintFps < 31) "SD 30fps"
elseif (srcHeight < 541) and (deintFps < 26) "1/2 HD"
elseif (srcHeight < 541) and (deintFps < 31) "1/2 HD 30fps"
elseif (srcHeight < 721) and (deintFps < 26) "720p24"
elseif (srcHeight < 721) and (deintFps < 31) "720p30"
elseif (srcHeight < 721) and (deintFps > 31) "720p60"
else "1080p24"

I can add new profiles if the content I watch requires it. Each profile only contains chroma upscaling, image doubling and dithering

QB

[Qaq]

I hate the fact that I'm almost able to run at 30 fps, the render time needed for 30fps is 0.33ms while my upscaled 720p videos take 0.36-0.38 ms to render I'm so close :mad:You can try some simplier dithering and chroma upsampling options and see if it helps.

[G_M_C]

madshi, what are your plans to fully implement the NNEDI improvements for AMD?

I use NNEDI only for chroma upscaling, and you mentioned that you hadn't focussed on implementing the improvements for that function yet. I was wondering when the improvements come to 'NNEDI for chroma upscaling'.

[ryrynz]

I was wondering when the improvements come to 'NNEDI for chroma upscaling'.

Going to take a guess here and assume it'll be in the next version, I think madshi was trying to determine which improvement to use but has decided on both with an option to select what suits best. Probably unlikely that a chroma build will come about as the luma tested the modes sufficiently?

I also don't think we're that far from a new release either, he's been spending a bit of time on it recently.

[wolfman2791]

Technical question for you guys... is it better to downsample 2880x1620 resolution, which can only handle Jinc 3 for my graphics card OR leave it at 1920x1080 with NNEDI? You're the most knowledgeable crowd by far so i thought i'd ask.

[blu3wh0]

I finally got around to testing the new interop builds for AMD, and I'm very happy to say that all my issues with luma doubling are now gone with my 7950's on PCI Express 2.0 x8. Thank you madshi!

Here are my results:
720p 24fps
Test 1- No frames dropped all the way up to NNEDI luma doubling at 32 neurons, chroma upscaling at 32 neurons, SM on, and ED option 2. However, for a few minutes I can do luma doubling at 64 neurons before I start dropping frames.
Test 2- No frames dropped all the way up to NNEDI luma doubling at 32 neurons, chroma upscaling at 32 neurons, SM on, and ED option 2. At 64 neurons for luma doubling, too much subtitle rendering will also cause frames to be dropped, but otherwise it seems to for a short period. Haven't performed longer testing on this version.

I'll need to do some longer testing to get a better feeling for this. Although overall, I believe Test 1 performs better for me with the added benefit of less memory usage. Again, thank you!

[noee]

AMD builds OP1 and OP2, more results (after a complete driver/media chain refresh):

with SD film 720x352 8bit, SM/FRC On, Deband Low, OrdDith, NNEDI3 luma dbl 32n

OP1: 520MB GPU RAM, ~88% GPU load, no drops
OP2: 591MB GPU RAM, ~92% GPU load, one drop every 10 sec

Couldn't run SM at all with 87.9 and any NNEDI. So, looks like OP1 is better here afterall.

[Neet009]

I ordered an EVGA 750Ti Superclock and will be here sometime this week. I hope madVR will be able to handle the following on 720p24 sources on a 1920x1080p display...

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: ED1, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)
3DLUT: Enabled


test with my 750Ti

GPU 1150MHz/Memory 5400MHz
720p24 on 1920x1080 display

madVR setting:
Chroma Upscaling: Jinc3+AR
Image Doubling: NNEDI3 luma doubling 32 neurons
Image Upscaling: Lanczos3+AR
Image Downscaling: Catmull-Rom AR LL

Deband: Low (fade in/out high)
Dithering: ED1, colored noise, change every frame
SmoothMotion: on
3DLUT: Enabled

rendering time:35.8 ms (no dropped frames)

GPU 1285MHz/Memory 5600MHz (OC)

rendering time:32.5~33 ms

I can set Chroma Upscaling to NNEDI3 16 with stable playback,no dropped frames if I turned SmoothMotion off or change ed1 to od. (rendering time:36~37 ms)

[madshi]

madshi, have you looked into OpenCL 2.0? Shared Memory looks like it might improve things.
I don't want or need shared memory. All the processing should run on the GPU RAM. And that's exactly what is *not* happening with AMD's OpenCL <-> D3D9 interop at the moment, which is why it's so slow.

My GTX 690 gave a very poor performance when using image doubling. i can not use chroma upcaling, luma and chroma image doubling even at 16 neurons for 720p.it drops frames. Am i doing something wrong?
Sounds weird. The 690 should have enough power. However, you left out a key information: Are you talking about 720p24 or 720p60? 60p requires 2.5x as much GPU power as 24p. If you're talking about movies which are encoded as 60p, try activating forced film mode with v0.87.10, that should convert those 60p back to 24p, which means your GPU should get 2.5x faster, so to say. You will probably have to force deinterlacing on by pressing Ctrl+Alt+Shift+D twice. Or add "deint=film" to the video file name.

Madshi , for people that use jriver, can you add lumasharpen in madvr?
Not in v0.87.x. Maybe in the next build, but I don't know that yet.

I found out why my 270x cannot do 32 neurons. :stupid:
The video I've been testing is encoded at 30 fps. I have no issues with 23.976 fps videos.
Is it a movie or is it native interlaced content? If it's a movie, try forced film mode. That should allow you to do 32 neurons.

Does MadVR support capture devices such as webcams or capture cards (from game consoles)? I tried to use MPC to show the device and the webcam sometimes works. When I turn on smooth motion, the image freezes (and fails) to function. Does smooth motion not work with capture devices?
There is nothing specific that would allow madVR to (or stop madVR from) supporting capture cards. What madVR needs is a proper clock and proper timestamps, especially for smooth motion FRC. If those are provided, madVR should work fine. Otherwise not.

To make installing or updating to stable releases of madVR (and MPC-HC and LAV Filters) a bit more comfortable, I've released an open-source utility (http://forum.doom9.org/showthread.php?t=170459) which you might want to check out.
Nice! :)

Is it possible to make madVR read matroska tags (http://matroska.org/technical/specs/tagging/index.html) in addition to "tags" with madVR settings in the file name?
Yes, it is possible. But I don't accept feature requests at the moment.

Can you please explain how Source Levels work?
For example; when I cut the Black Source Level to -5, does madVR actually clip the blacks or rolls off gradually (in 16-bit) like the brightness slider?
madVR doesn't clip, it does the math in 32bit floating point and stores the results in 16bit, but Direct3D will clip at 0-255 if you output PC levels.

Yes i am using error diffusion type 2. now as you say i am gonna try ordered dithering. thanks!
Yeah, that makes sense because error diffusion runs quite slowly on current generation NVidia GPUs - except for Maxwell, which seems to run error diffusion quite a bit faster than Kepler.

I don't understand why the image upscaling has any effect, when using nnedi3. If I'm using nnedi3 to upscale 720p with a display resolution of 1080p, why is the image upscaling effecting performance?
You're probably talking about nnedi3 luma doubling? If so, chroma still needs to be upscaled, too, and for that the image upscaling algorithm is used.

madVR handles the input levels automatically, you don't need options or anything for them. You can override it on a per-file basis, but its only useful if madVR detects the wrong input level. The input level is defined by the file, not any setup in your display chain.

For the level settings, there is basically 3 combinations of settings:

(1) TV accepts PC/Full levels: Set madVR to PC/Full, set GPU to PC/Full - This is usually considered the better option, if your TV can do it.

(2) TV accepts only TV/Limited levels:
(a) Set GPU to PC/Full and set madVR to TV/Limited
(b) Set GPU to TV/Limited and set madVR to PC/Full

In the second case, both a and b have small drawbacks.
With (a), you get crushed black/white on the desktop, but everything is perfect in madVR.
With (b), you get proper colors everywhere, except that the GPU is doing the PC range -> TV range conversion, which may not be the highest quality and end up in banding.

NEVER set both madVR to TV/Limited AND GPU to TV/Limited, you get double range compression.

Because of these drawbacks with either a or b, its usually recommended to use (1) if your TV accepts it. Its only "downside" is that BTB/WTW are lost, but its arguable if those are even worth keeping. If you insist on WTW/BTB, the 2a setting is probably the next best for video quality, if you can live with crushed black/white on the Desktop.

However you set it up, the test patterns should only show 16-235 after calibration.
Quoted for truth.

But how a Blu-Ray player does this (Chroma upscaling) without complicated dithering, and without introducing error?
If the Blu-Ray player outputs "untouched" YCbCr 4:2:2, at least the luma channel doesn't need dithering. The chroma channel would benefit from dithering (because it's impossible to output it untouched) and probably no Blu-Ray player applies dithering, but the loss in quality for the chroma channel is probably rather small.

I wonder what madshi is cooking...
I'm rarely cooking. Delivery services ftw! :p

Just played around with mpv and was wondering why all the videos were so laggy on pans even though no frames were dropped... Then remembered mpv does not have smooth motion. Scary, how fast one gets accustomed to some features and forgets they are even there.
:D

Now Madshi support dx11 for dithering , has changed something for 10 bit output too?
No.

AMD have released new drivers (http://support.amd.com/en-us/kb-articles/Pages/latest-catalyst-windows-beta.aspx), I wonder if there has been any improvements?
Haven't tested it, but I doubt we'll get a fix so fast.

madshi, what are your plans to fully implement the NNEDI improvements for AMD?
There should already have been some improvements in the test build, I think. Nothing more planned.

[madshi]

madVR v0.87.10 released

http://madshi.net/madVR.zip

* added some optimizations to reduce AMD OpenCL interop cost
* added new windowed presentation path ("present several frames in advance")
* added support for decimating 50p/60p movies to 25p/24p
* added profile strings "filePath/Name/Ext", with wild char ("?", "*") support
* fixed: #181: profile auto switching sometimes invalidated file name tags
* fixed: #192: black flashing with Smooth Motion + NNEDI3 chroma doubling
* fixed: #193: image corruption when up&down scale is needed at the same time
* fixed: crash on Vista when trying to activate error diffusion
* fixed: Intel OpenCL CPU driver sometimes crashed
* OpenCL should now automatically prefer NVidia GPUs on Optimus laptops
* refresh rate hack is now only installed on Windows 8 (and newer)
* "Pause" OSD message no longer blocked
* file "madshi.net/madVR/version.txt" lists the current version number
* file "madshi.net/madVR/sha1.txt" lists SHA1 hash of the current "madVR.zip"
The interop test build 1 is now active by default. The new option (see settings dialog) activates test build 2.

madVR should now be able to decimate 720p50/60 movies to 720p25/24. You need to manually activate deinterlacing for that and force film mode. Or add "deint=film" to the file name.

[turbojet]

Thanks for adding 720p ivtc but it doesn't work here without adding deint=film to the file name. I have deinterlacing and film mode activated.

As for my message earlier about deinterlacing 50i/60i at 50p/60p being the final thing needed for deinterlacing+film mode to work for all, while it's getting closer another thing I notice with 2:2 30i content is it changes to a 24hz refresh rate and leaves sm disabled.

[DragonQ]

At first it seemed the "Use alternative interop hack" option didn't do anything but after restarting MPC-HC I can use NNEDI3 16 neurons, as with the test 2 build. The interop stat is ~5 ms with the rendering stat at ~21.5 ms for 720x576i/25 (with Smooth Motion to 60 Hz). Surely <20 ms is required for 50 fps without framedrops, and I have no frame drops. Deinterlacing increases from ~1 ms to >4 ms when increasing to 32 neurons. Basically the stats make no sense lol. :scared:

[bozokaydin]

Sounds weird. The 690 should have enough power. However, you left out a key information: Are you talking about 720p24 or 720p60? 60p requires 2.5x as much GPU power as 24p. If you're talking about movies which are encoded as 60p, try activating forced film mode with v0.87.10, that should convert those 60p back to 24p, which means your GPU should get 2.5x faster, so to say. You will probably have to force deinterlacing on by pressing Ctrl+Alt+Shift+D twice. Or add "deint=film" to the video file name.



For a 720p24 this is the maximum settings i can use:

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: Ordered, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)

anything above this my gtx690 cannot handle. for example if i enable dithering to error difusion it starts to drop frames. or if i activate chroma doubling 16 neurons it cannot handle.

[seiyafan]

edit: I just did a few more tests and found exclusive mode to be underwhelming. The upload and render queues were always low whereas windowed mode(both overlay and non) had queues that were relatively full(5-6 and maybe 2-3). This does result in occasional frame drops. No such issues with windowed modes. I thought exclusive mode was supposed to be better.

Same here, exclusive mode leaves the queue empty and frame dropped.

[madshi]

Thanks for adding 720p ivtc but it doesn't work here without adding deint=film to the file name. I have deinterlacing and film mode activated.
As written in the release notes, you have to manually activate deinterlacing, meaning you have to press the deinterlace keyboard shortcut. Or add a file name tag. Maybe I'll add options to optionally automatically activate film mode for 60p/50p content in a future version, but first I'd like to get some feedback on whether it works reliably or not.

As for my message earlier about deinterlacing 50i/60i at 50p/60p being the final thing needed for deinterlacing+film mode to work for all, while it's getting closer another thing I notice with 2:2 30i content is it changes to a 24hz refresh rate and leaves sm disabled.
Currently forcing film mode for 30i=60i content always results in 24Hz refresh rate. That's the way it is for now.

The interop stat is ~5 ms with the rendering stat at ~21.5 ms for 720x576i/25 (with Smooth Motion to 60 Hz). Surely <20 ms is required for 50 fps without framedrops, and I have no frame drops. Deinterlacing increases from ~1 ms to >4 ms when increasing to 32 neurons. Basically the stats make no sense lol. :scared:
Don't take the render times as the ultimate gospel. They're a good indication, that's all. There's a reason why I disabled the render times in the interop test builds.

For a 720p24 this is the maximum settings i can use:

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: Ordered, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)

anything above this my gtx690 cannot handle. for example if i enable dithering to error difusion it starts to drop frames. or if i activate chroma doubling 16 neurons it cannot handle.
That sounds surprisingly slow for a 690. You sure you don't have anti-aliasing or anisotropic filtering forced on or something like that? Not sure what else could cause it to be that slow. Anyway, not much I can do about it, in any case...

[seiyafan]

Is MadVR deinterlace same as LAV video decoder software deinterlace in terms of performance?

Another question is the Hardware deinterlace in LAV video decoder doesn't seem to be working, with it enabled I don't see any difference, has anyone else notice it?

[turbojet]

Okay I thought changelog notes were to manually activate in settings. I'll start renaming files.

What I mean by 50i/59i/60i that at 50p/59p/60p has no duplicates and 25i/29i/30i does have duplicates. Is there a better or more preferred way to represent this?

[GCRaistlin]

Is MadVR deinterlace same as LAV video decoder software deinterlace in terms of performance?

Another question is the Hardware deinterlace in LAV video decoder doesn't seem to be working, with it enabled I don't see any difference, has anyone else notice it?

I do see the difference while playing interlaced video with this LAV Video option enabled and disabled. I have noticed as well that LAV Video detects interlaced video better than madVR. But I'm not sure if LAV Video's deinterlace feature doesn't interfere with madVR's IVTC.

[QBhd]

madVR v0.87.10 released

The interop test build 1 is now active by default. The new option (see settings dialog) activates test build 2.

Is the setting you are talking about?

http://i.imgur.com/6xq7ngb.png

QB

[nevcairiel]

Is the setting you are talking about?

That setting was there before, so thats not it. ;)

[sneaker_ger]

It's at the bottom of the scaling algorithms->image doubling dialog.

[QBhd]

That setting was there before, so thats not it. ;)

I thought as much

Well D'uh... I even renamed my settings.bin to see if I could find that sucker... it's so out of the way... thanks guys

QB

[6233638]

* fixed: #181: profile auto switching sometimes invalidated file name tagsUnfortunately this doesn't seem to be working for me.

MPC-HC reports that video deinterlacing is being used, but I have to toggle film/video deinterlacing for the combing artifacts to disappear.

In JRiver MC19, deinterlacing isn't being activated at all.



P.S. I don't know what changed, but it used to be that a refresh rate change would clear the LUT being used by f.lux - in this build it doesn't change at all!
I actually though it was a shortcoming of f.lux when running in "safe mode" rather than something that madVR was causing.

[Asmodian]

For a 720p24 this is the maximum settings i can use:

Image Doubling: NNEDI luma doubling 32 neurons
Image Downscaling: Catmull-Rom AR LL
Chroma Upscaling: Jinc3 AR
Deband: Low/High
Dithering: Ordered, colored noise, change every frame
SmoothMotion: Always (display is running at 60p)

anything above this my gtx690 cannot handle. for example if i enable dithering to error difusion it starts to drop frames. or if i activate chroma doubling 16 neurons it cannot handle.

I assume you have SLI on due to the 690 being a dual GPU card. Can you turn SLI off and test performance? Also try the newest madVR, I had a large improvement with SLI on (SLI Titans) with the AMD interop optimization test builds but identical performance with SLI off. SLI on is still a bit slower than SLI off.

[Anime Viewer]

madVR v0.87.10 released

http://madshi.net/madVR.zip

[code]
* added support for decimating 50p/60p movies to 25p/24p
* added profile strings "filePath/Name/Ext", with wild char ("?", "*") support
* OpenCL should now automatically prefer NVidia GPUs on Optimus laptops
madVR should now be able to decimate 720p50/60 movies to 720p25/24. You need to manually activate deinterlacing for that and force film mode. Or add "deint=film" to the file name.

Before installing I deleted the forceVendor key, installed, opened a file, and it did successfully use the Nvidia GPU. Just to make sure I went and deleted the Nvidia hive under the OpenCL hive (to put it in the state it was in prior to first using the forceVendor string), and it successfully re-added it again after I launched a video.

As far as the refresh rate downshifting / dinterlacing combo being forced on: couldn't that potentially cause the system to do more work in case like:
Wouldn't a monitor with a 60hz and 59hz refresh force it back up to 59/60 when it determines said monitor can't support displaying video at 24hz?
Might quality of a progressive video degrade if interlacing was forced on it?
I can probably determine the answers after testing (haven't had time to test the refresh/interlace forcing yet), but conceptually those potential issue come to mind...

[ThurstonX]

This may be of interest to some, old news to others, but I figured I'd share it. Before that, I will say I ran the restore default settings script, then unzipped v0.87.10, as I wanted to start from a clean slate. Prior to that I ran the latest AMD uninstall utility (really reset everything; had to redo the Scaling settings, which was a first) and installed the latest RC Catalyst package.

I was curious to retry NNEDI3 Image Doubling on my passive Sapphire Radeon HD 6670, which is my main HTPC card. One thing I've never messed with was the Force Film setting. I was suspicious that the VOBs in some ripped DVDs were showing as 23.976 fps in MediaInfo, but that madVR's OSD was saying they were 29.970 fps. I forget which DVD I was testing, but forcing Film Mode on it made it playable using NNEDI3 doubling @ 32 neurons (Bicubic and Lanczos for Chroma and Image; no Jinc 3). Without forcing Film on it, I got dropped frames. Other DVDs played fine with those settings without disabling Automatic Source Type Detection (and forcing film mode had no bad effects). I think I saw some improvement with NNEDI3 vs. Jinc 3 AR for Chroma and Image upscaling, but it was pretty subtle at first blush. Rendering times certainly increased with NNEDI3, but at 38-39 ms, I had no dropped frames. In one instance, forcing film actually caused the queues to fill better than not forcing it. In another, forcing film stopped an insane number of repeated frames (didn't note a difference in playback quality, but I was fixated on the OSD at the time; there were no drops in this instance).

All that leads me to the conclusion that I really need to test a video file before settling on NNEDI3 or straight up Jinc 3 AR, which I can always use. The VOB info in MediaInfo also has me confused. Is it a reliable method for determining whether or not to force film mode (e.g., force film if MediaInfo reports 23.976)? With AVIs or MKVs that I've encoded and know are 23.976, I'm pretty sure I can always use NNEDI3 32 neurons for image doubling. It's DVDs that are a bit frustrating. Movies seem pretty reliable, but concert DVDs (try Rush in Rio... bah! ridiculous thing) are kind of tough.

I was happy to be able to use NNEDI3 for some things where at first it seemed too much for that old card. At least testing is pretty easy. I'm not ready to delve into profiles just yet, but maybe they will help down the line.

Thanks in advance for any insights.

[QBhd]

All that leads me to the conclusion that I really need to test a video file before settling on NNEDI3 ...

That's what profiles are for... to do the work for you

QB

[ThurstonX]

That's what profiles are for... to do the work for you

QB
Obviously I need to take some time to learn how to set them up. Now to find the time.

[huhn]

before i go and create a bunch of test files.

is this ONLY working with 720p60/50 source (the 720p TV and BD spec)so not with 1080p60?

i tried a 1080p30i source with constant 4:2:2:2 content this was not working at all. but this should work if i resize this source to 720p30i and treat it as progressive and now double the frame rate so the cadence is 8:4:4:4 right?

is 720p25 in 720p60 cadence 8:4:4:4:4 supported or worth testing?

[Anime Viewer]

before i go and create a bunch of test files.

is this ONLY working with 720p60/50 source (the 720p TV and BD spec)so not with 1080p60?

i tried a 1080p30i source with constant 4:2:2:2 content this was not working at all. but this should work if i resize this source to 720p30i and treat it as progressive and now double the frame rate so the cadence is 8:4:4:4 right?

is 720p25 in 720p60 cadence 8:4:4:4:4 supported or worth testing?

Based on madshi's post (quoted below) it should function on 50p and 60p videos regardless of resolution (ex: 1080, 720, 480, etc). If its not natively 50p or 60p then it shouldn't touch the file (ex: 30, 27, etc).

madVR v0.87.10 released

http://madshi.net/madVR.zip

[code]
* added support for decimating 50p/60p movies to 25p/24p

madVR should now be able to decimate 720p50/60 movies to 720p25/24. You need to manually activate deinterlacing for that and force film mode. Or add "deint=film" to the file name.

[huhn]

Based on madshi's post (quoted below) it should function on 50p and 60p videos regardless of resolution (ex: 1080, 720, 480, etc). If its not natively 50p or 60p then it shouldn't touch the file (ex: 30, 27, etc).

the point is that he used 720p50/50 not 50/60 in his example and he used a couple of frame size detection already for colorspaces. no native source in a BD or TV is 1080p60 with 1080p24 in it. the BD has definitely not a spec for this of cause i don't know all TV broadcasting specs of all countrys.

the next thing is that normal 720p24 -> 720p60 got a cadence of 6:4 not 8:4:4:4.

and 1080p30i 4:2:2:2 should work regardless. this is a normal case the hole IVTC thing is made for (30i ->24p). this is know to be a problem so this is more like a question if this should work now too cause it is technical the same problem.