madVR - high quality video renderer (GPU assisted)

[Ver Greeneyes]

Mathias: Sigmoidization was definitely pointing you in the wrong direction. It's dead. (Apologies.)
On the other hand, I'm pretty excited about this: http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=25935&p=113483#p113483

Interesting! So, in words:

1) Convert input (a) from sRGB to linear RGB

2a) Copy (a) to (b)
2b) Apply a gamma of 3 to (b)
2c) Resize (b) using whatever method
2d) Apply a gamma of 1/3 to (b)

3a) Copy (a) to (c)
3b) Negate each channel of (c)
3b) Apply a gamma of 3 to (c)
3c) Resize (c) using whatever method
3d) Apply a gamma of 1/3 to (c)
3e) Negate each channel of (c)

4a) Set output (d) to the mean of each channel of (b) and (c)
4b) Convert output (d) from linear RGB to sRGB

This then produces a result with similarly reduced ringing as if you had converted to LAB before resizing?

Edit: I guess if this thread is solely for madVR support, this might need its own topic.

[Stereodude]

Ok, in that case it should also be 24fps for decimated 720p60 content. Is there already an entry in the bug tracker for this? If not, could you please make one? Thx.
Okay, I made one. Sorry for the delay. #228 (http://bugs.madshi.net/view.php?id=228)

Also, I know you're officially not taking feature requests, but have you given any thought to adding a masking step in the processing pipeline. Something where a custom number of pixels on each edge could be masked/replaced with black? Something like the equivalent of crop + addborders in Avisynth. This would allow people with fixed pixel displays who are running with 1:1 pixel mapping to hide unwanted artifacts from the edges of content. Thanks!

[billqs]

I know that you can uncheck various parts of MadVR if you don't have enough horsepower in your CPU/GPU, but I'm building a 4k box and want to know if my i7 3820 CPU and GTX 650Ti GPU would be strong enough to scale to 4k using Jinc3 + AR filter (My Sony projector upscales well, but has too much ringing.) If this is not quite powerful enough to do this, I am open to suggestions a better GPU.

Thanks!

[dansrfe]

I know that you can uncheck various parts of MadVR if you don't have enough horsepower in your CPU/GPU, but I'm building a 4k box and want to know if my i7 3820 CPU and GTX 650Ti GPU would be strong enough to scale to 4k using Jinc3 + AR filter (My Sony projector upscales well, but has too much ringing.) If this is not quite powerful enough to do this, I am open to suggestions a better GPU.

Thanks!

Depends on what source you're scaling from. If you're scaling from full 1080p then you'll need a 8xx or Titan I believe for Jinc 3 AR.

[Asmodian]

My Titan can do 1080p30 -> 4K with NNEDI3 32 neuron luma doubling and Jinc3+AR for both chroma and image scaling with smooth motion on, but it is pretty busy at that point. What this says about a 650Ti doing Jinc3, I am not sure. :(

If you are willing to use something like bicubic75+AR for chroma and Jinc3+AR for image the 650Ti might do it but it would be close at best.

[huhn]

a r9 280 + should be able to do it my r9 270 gets 44 ms with image only and 50ms+ with chroma and image at jinc 3 ar on a 1080p23 source. spline 3 ar is a joke for it it's about 12 ms i'm not even sure if it is in the highest powerstate.

BTW you can easily test this bypressing numblock 9 and force a fake UHD output on mpc HC/BE.
for some reasons image doubling is faster with 16 neurons...

[Asmodian]

I did a madVR performance test on my Titan with Jinc3+AR/Jinc3+AR, smooth motion on, debanding low/low, ordered dithering. 4 frame rendering buffer, 1 present queue, windowed mode (new path), Windows 8.1. SLI Off.

1080p24 -> 3840x2160@60Hz, 17.5ms

While playing Nvidia Inspector says:
GPU 41-52%
MCU 10-14%
GPU CLK 549-679 MHz
RAM 1429MB

Assuming performance is directly proportional to shaders*MHz a stock GTX650Ti has 39% of the GPU performance of a Titan @ 679 MHz. It also has 28% of the memory bandwidth of my Titan, well above the 14% I used.

This works out to at worst a 25.9ms rendering time on a GTX650Ti at 90% GPU usage; fine for 30fps or lower but not 60fps (16.6ms).

If I use Bicubic75+AR for chroma scaling and Jinc3 for image it changes to a 15.6 ms rendering time with at most 44% GPU usage. Scaling to a 19.5ms rendering time on a GTX650Ti @ 928MHz and 90% usage. Still not enough for 60 fps but plenty of headroom for 30 fps (33.3ms).

These are worst case estimates though, I wouldn't be surprised if the GTX650Ti had the GPU power for Jinc3 Image and Bicubic75 chroma for 1080p60. Sadly the memory bandwidth is probably insufficient.

Memory usage is a major issue though; if I turn the buffers down to 4 rending and 1 present I still hit 1429MB of GPU memory used and 4/1 is the minimum.

Edit: I was able to get memory usage down to 958MB with Jinc3+AR chroma and Jinc3+AR image by turning on most of the trade quality for performance options (10 bit instead of 16 bit) and turning off debanding and smooth motion.

[billqs]

Thanks, guys! You've given me some great information! I'm not looking to game in 4k at this point, just watch movies, so I could probably handle staying at 30fps for the time being. I'm much more familiar with using ATI/AMD cards for my HTPCs but I had read where the Kepler cards had beta drivers allowing them to expose 4k all the way up to 60hz over the same HDMI 1.4 card. My projector doesn't take display port, is there a good 4k option over HDMI for the R9 series from AMD?

[Asmodian]

Wow, I had no idea. I happened to be running the newest Nvidia betas and just unplugged my DP 1.2 cable and plugged in an HDMI. It switched to 3840x2160@30 Hz and I have the options for 29, 25, 24, and 23 Hz in Windows.

These are great options for madVR but other then that we are going off topic (as happens all too often here). :(

[nikola]

Wow, I had no idea. I happened to be running the newest Nvidia betas and just unplugged my DP 1.2 cable and plugged in an HDMI. It switched to 3840x2160@30 Hz and I have the options for 29, 25, 24, and 23 Hz in Windows.

These are great options for madVR but other then that we are going off topic (as happens all too often here). :(

On the topic of getting the best visual quality:

Be aware that nVidia is actually using chroma downsampling (4:2:2 if I recall correctly) to squeeze 4K into the bandwidth constraints of HDMI 1.4. Which defeats the whole purpose of madVR's RGB processing.

[huhn]

it's using 4:2:0 for 4k 60 fps but laeve it at this.

[Ceremony]

I've got an issue with madvr on my new laptop (i5-4210, HD4400 or GTX850M, 8GB RAM):

When watching, I get frame drops after a minute or so. The backbuffer queue (all other queues are full!) is suddenly empty and causing dropped frames till i restart the player for another minute of flawless playback. I tried every option in madVR there is. Using either the integrated HD4400 GPU or the dedicated GTX850M doesnt seem to make a difference either. Exclusive fullscreen mode doesnt change anything either.

Overall resource usage is very low while watching and with frames taking merely up to 10ms to render, i doubt raw power is the issue... HELP!

[Asmodian]

It sounds like something is going into a low power mode? Check all your advanced power management options or try a "High Power" preset for testing. Try renaming the video player's exe? I don't think this is really a madVR issue, maybe open a new thread so we can discuss possibilities without clogging up this thread with general (video playback) PC support. Maybe link to it from here? :)

[Ceremony]

It sounds like something is going into a low power mode? Check all your advanced power management options or try a "High Power" preset for testing. Try renaming the video player's exe? I don't think this is really a madVR issue, maybe open a new thread so we can discuss possibilities without clogging up this thread with general (video playback) PC support. Maybe link to it from here? :)

mhh yeah i will, once i tried that power thingy - though it also happens when im on AC, so i somewhat doubt it. also, render queue also drops. ignore my previous statement^^

[JarrettH]

I did a madVR performance test on my Titan with Jinc3+AR/Jinc3+AR, smooth motion on, debanding low/low, ordered dithering. 4 frame rendering buffer, 1 present queue, windowed mode (new path), Windows 8.1. SLI Off.

1080p24 -> 3840x2160@60Hz, 17.5ms

While playing Nvidia Inspector says:
GPU 41-52%
MCU 10-14%
GPU CLK 549-679 MHz
RAM 1429MB

Assuming performance is directly proportional to shaders*MHz a stock GTX650Ti has 39% of the GPU performance of a Titan @ 679 MHz. It also has 28% of the memory bandwidth of my Titan, well above the 14% I used.

This works out to at worst a 25.9ms rendering time on a GTX650Ti at 90% GPU usage; fine for 30fps or lower but not 60fps (16.6ms).

If I use Bicubic75+AR for chroma scaling and Jinc3 for image it changes to a 15.6 ms rendering time with at most 44% GPU usage. Scaling to a 19.5ms rendering time on a GTX650Ti @ 928MHz and 90% usage. Still not enough for 60 fps but plenty of headroom for 30 fps (33.3ms).

These are worst case estimates though, I wouldn't be surprised if the GTX650Ti had the GPU power for Jinc3 Image and Bicubic75 chroma for 1080p60. Sadly the memory bandwidth is probably insufficient.

Memory usage is a major issue though; if I turn the buffers down to 4 rending and 1 present I still hit 1429MB of GPU memory used and 4/1 is the minimum.

Edit: I was able to get memory usage down to 958MB with Jinc3+AR chroma and Jinc3+AR image by turning on most of the trade quality for performance options (10 bit instead of 16 bit) and turning off debanding and smooth motion.

Speaking of rendering times, how do I know what to aim for? My refresh rate is 60 hz. Where did 16.6 ms come from? :thanks:

[Stereodude]

Speaking of rendering times, how do I know what to aim for? My refresh rate is 60 hz. Where did 16.6 ms come from? :thanks:
1/60= ? :devil:

[Asmodian]

Speaking of rendering times, how do I know what to aim for? My refresh rate is 60 hz. Where did 16.6 ms come from? :thanks:

In addition to what Stereodude said; it isn't based off of your screen's refresh rate but the frame rate of the content you want to watch. Also madVR reports the rendering times you need to be below as "movie frame interval" in its OSD.

[seiyafan]

How do you show real rendering time? After the last NNEDI3 improvement my rendering times were all in the 1-2 ms.

[Anime Viewer]

How do you show real rendering time? After the last NNEDI3 improvement my rendering times were all in the 1-2 ms.

My guess is that something has disabled NNEDI3 in madVR/MPC (or whatever you may be using in place of MPC...like Potplayer, etc). Have you changed anything in madVR settings recently (like setting image upscaling to DXVA2)?

[madshi]

It's already disabled, and GPU queue and pre-presented is at 8 -- should I lower it?
That probably won't help. Playing games works alright on your laptop, using the NVidia GPU? I've no idea what's going on there, to be honest. Probably a driver issue of some sort... :(

Blu-Ray is encoded and stored on the disc as Ycbcr 4:2:0, then madVR converts YCbCr to RGB and does all the processing and output in RGB, correct?
Well, madVR first upscales chroma to get from 4:2:0 to 4:4:4, and then converts to RGB, but other than that: Yes.

I have read on AVSForums that most TVs convert incoming RGB signal back to YCbCr for processing unless we choose a special option to tell the TV to skip the conversion and output pure RGB.
On a Panasonic this option called "1080 Pixel Direct".
Quite true. However, usually this converting back to YCbCr is done in more than 8bit and with some better TVs also in 4:4:4. If done that way it's not too much of a problem. But if the display processes in 4:2:2 or 4:2:0, then obviously you're losing quality. If you have a "pure" mode, that is a good mode to use. However, some displays are fixed to 60Hz in that mode. So there's no simple recommendation that would be valid for every TV.

In my Optimus enabled laptop with GT650m graphics, I can use NNEDI3 upscaling (with no frame drops for SD videos) by adding a registry key to force NVIDIA gpu for OpenCL processing; but I am not able to use any of the error diffusion algorithms due to frames getting dropped when they are enabled. I was wondering if this was due to intel igpu being used for directcompute processing. If this is a possibility, can a registry key be added similarly to force the NVIDIA gpu for directcompute processing?
That's technically not possible, unfortunately. I can choose the GPU for OpenCL processing, but I can't choose the GPU for Direct3D. Have you tried renaming your media player? Reportedly sometimes that helps to activate the NVidia GPU instead of the Intel one.

i was under the impression that onboard graphics is more than sufficient for video playback, but the little time spent reading up on madvr so far has proven me wrong. i know plenty about audio related matters but not so much video, hence this is doubly confusing for me

where do you think the law of diminishing returns kicks in with regards to settings used and GPU power?
A good current onboard graphics is powerful enough for madVR to beat any other renderer out there in quality, and most external video processors and Blu-Ray players etc, too. However, madVR also contains some very power hungry algorithms which are able to squeeze out another few percent of added quality. I can't tell you whether your eyes would see/notice the difference or not, or whether the improvement is worth the added price of a dedicated GPU for you. I just try to give you the best quality I can provide for any GPU power you can offer. Now it's your job to decide whether you're satisfied with very good (if so, just get a good onboard GPU), or whether you want to pay more dollars for a few percent more quality.

EDIT : I've changed the settings with 1080p60 and 1080p59. It seems to work because the refresh rate is changed during playback but it's either 60Hz or 59.9Hz.
Actually that is correct! 1080p59 is a weird name for "1080p with 59.940Hz". Basically 1080p59 is "60Hz / 1.001".

USA NTSC and ATSC use this weird 1.001 factor, which is why for video playback it's usually better to use 24/1.001 instead of 24, and 60/1.001 instead of 60.

as far as i know repeated or dropped frame forced by vsync are not shown in the OSD.
They should be, I think?

Not even in LAV, but in ffdshow... and I find it interesting that madVR specifies that the fullrange information comes not from upstream, but specifically from ffdshow, too.
Cool, eh? :p

Also, I know you're officially not taking feature requests, but have you given any thought to adding a masking step in the processing pipeline. Something where a custom number of pixels on each edge could be masked/replaced with black? Something like the equivalent of crop + addborders in Avisynth. This would allow people with fixed pixel displays who are running with 1:1 pixel mapping to hide unwanted artifacts from the edges of content. Thanks!
Some media players already have a feature like this, e.g. J.River MC. However, I do plan to add masking features in a future version. It's mainly intended for front projection, though (CIH etc).

Be aware that nVidia is actually using chroma downsampling (4:2:2 if I recall correctly) to squeeze 4K into the bandwidth constraints of HDMI 1.4. Which defeats the whole purpose of madVR's RGB processing.
it's using 4:2:0 for 4k 60 fps but laeve it at this.
^ Yes. AFAIK NVidia uses 4:2:0 for 4Kp60, because that's the only way you can send 4Kp60 with the HDMI 1.4 bandwidth. Framesrates of 30p and lower should not be affected, IIRC.

[madshi]

Mathias: Sigmoidization was definitely pointing you in the wrong direction. It's dead. (Apologies.)
On the other hand, I'm pretty excited about this: http://www.imagemagick.org/discourse-server/viewtopic.php?f=22&t=25935&p=113483#p113483
Hmmmm... Sounds interesting. I'm not familiar with the ImageMagick scripting language, though. Could you do me the favour of describing in a few words how the algorithm works exactly? E.g. I don't understand the exact meaning of "-evaluate-sequence mean -colorspace gray -auto-level" and of "-delete 0 -compose over -composite". Thanks! :)

[DarkSpace]

Not even in LAV, but in ffdshow... and I find it interesting that madVR specifies that the fullrange information comes not from upstream, but specifically from ffdshow, too.
Cool, eh? :p
Indeed it is! And although I don't use ffdshow myself, I wonder if it makes sense to select profile rules based on the upstream filter, or at least based on whether or not the upstream filter is ffdshow, since madVR can obviously detect that :p

[dansrfe]

madshi, I've discovered that my iGPU -> 750M problem is resolved when I select "System Default", "VMR 7 (Renderless)", or "Overlay Video Renderer" with "High-performance NVIDIA processor" selected for a renamed mpc-hc.

Strangely, I had only tried EVR and VMR 9 (Renderless) other than madVR and both of the others also gave the same black screen result as madVR. Should I open a bug report for this on the tracker and/or would you need me to provide more information? Thanks.

[Asmodian]

Have you tried "enable windowed overlay" in the rendering options of madVR? It is interesting that only the old renderers work for you.

[dansrfe]

Have you tried "enable windowed overlay" in the rendering options of madVR? It is interesting that only the old renderers work for you.

When I select overlay mode in madVR I get "Overly mode not supported on this GPU".

Furthermore, when madVR or any other unsupported renderer (EVR/VMR9) is selected then no activity is shown for the 750M in GPU-Z. While the other mentioned renderers utilize the 750M.

[lansing]

don't know if it's a bug or something that I missed, I'm using version 0.87.10. I wanted to play my video at its native 24fps, but turning deinterlacing off in madvr's setting doesn't do anything, the video still plays at 60fps.

[Flux]

madVR 0.87.4 + MPCHC (Windows 7 SP1 64-bit)

Chroma upscaling: NNEDI3 (32 neurons)
Image upscaling: Jinc (3 taps) + anti-ringing filter
Smooth motion: enabled
Trade quality for performance: all unticked except "use random dithering instead of OpenCL error diffusion"
Exclusive mode: enabled

AMD R9 280 3GB GPU (940 MHz / 1250 MHz) with Catalyst 13.12 drivers.

I'm playing 1080p24 10-bit videos and rendering times hover around ~ 22 ms with GPU usage ~ 50% (~ 70ms with error diffusion and ~90% GPU usage).

It is normal that I can't use error diffusion with this GPU? There is tons of dropped frames. Rendering times are correct for this GPU?

[sneaker_ger]

Upgrade to the latest version of madVR.

[Flux]

I thought it was latest. Now I'm using 0.87.10 version.

0.87.4

I'm playing 1080p24 10-bit videos and rendering times hover around ~ 22 ms with GPU usage ~ 50% (~ 70ms with error diffusion and ~90% GPU usage).

0.87.10

I'm playing 1080p24 10-bit videos and rendering times hover around ~ 19 ms with GPU usage ~ 43% (~ 20ms with error diffusion and ~47% GPU usage).

Well, I guess that noticeable difference with error diffusion :)

Thanks for correcting this simple mistake :thanks:

[James Freeman]

I am currently experimenting with my Panasonic ST60 plasma and MadVR, contrast ratio is superb (above 10,000:1), and colors are perfect (sub 2 dE with built in CMS).

First, with this particular model the madLevelsTweaker does nothing, but on my other displays it has effect; It appears the the GPU outputs 0-255 no matter what (like PC mode) when it detects this TV,
In RGB or YCbCr444 modes (nVidia card). I also use the "Pixel Direct" to get 4:4:4 Chroma.

To get a band-less picture I have to use the RGB Limited setting on the TV, and output 16-235 from madVR, this clips the OS shades but it does't matter for me.
When using the RGB Full setting & 0-255 with madVR, the TV creates minor banding like its compressing the incoming 0-255 to 16-235 then stretches back to 0-255.
Its' definitely not MadVR or Nvidia that creates the banding, it appears that the TV's processor only accepts 16-235 steps for bandless picture.

All in all, I preserve BTB and WTW using the RGB Limited on the TV & 16-235 with MadVR, and without any banding.
In effect madVR use one less dithering step because the image does not need to be stretch to 0-255, it stays at the native 16-235.

Bottom line is: 0-255 is not always best; it depends on the display device.
I use the Greyscale Ramp from the AVS709HD disc, and engage the "Reduce Banding Artifacts" in madVR for maximum smoothness, then I select the setting on the TV that looks smoothest/bandless.
I suggest you do the same, to know what's the native range the TV processor expects at the input; it may accept 0-255 and show all shades perfectly but create banding as a side effect.

One question remains; Can madTPG use 16-235, or is it locked to 0-255?

[madshi]

Indeed it is! And although I don't use ffdshow myself, I wonder if it makes sense to select profile rules based on the upstream filter, or at least based on whether or not the upstream filter is ffdshow, since madVR can obviously detect that :p
It would be possible to add this functionality, but I don't want to do anything just because in theory it sounds useful. Nobody had a practical situation yet where he actually needed it in real life, at least no one reported such a case yet.

madshi, I've discovered that my iGPU -> 750M problem is resolved when I select "System Default", "VMR 7 (Renderless)", or "Overlay Video Renderer" with "High-performance NVIDIA processor" selected for a renamed mpc-hc.

Strangely, I had only tried EVR and VMR 9 (Renderless) other than madVR and both of the others also gave the same black screen result as madVR. Should I open a bug report for this on the tracker and/or would you need me to provide more information? Thanks.
VRM7, System Default and Overlay Video Renderer all don't use Direct3D9 for rendering, that's why they work and why madVR and EVR don't work. D3D9 is broken on your Windows installation. This is not a bug in madVR, but a problem with your Windows installation. As I mentioned before, try installing D3D9 with the D3D9 web installer. If that doesn't help, you may have to reinstall the whole OS.

don't know if it's a bug or something that I missed, I'm using version 0.87.10. I wanted to play my video at its native 24fps, but turning deinterlacing off in madvr's setting doesn't do anything, the video still plays at 60fps.
What is the native format of that video (resolution, framerate, interlaced/progressive etc)?

I am currently experimenting with my Panasonic ST60 plasma and MadVR, contrast ratio is superb (above 10,000:1), and colors are perfect (sub 2 dE with built in CMS).

First, with this particular model the madLevelsTweaker does nothing, but on my other displays it has effect; It appears the the GPU outputs 0-255 no matter what (like PC mode) when it detects this TV,
In RGB or YCbCr444 modes (nVidia card). I also use the "Pixel Direct" to get 4:4:4 Chroma.

To get a band-less picture I have to use the RGB Limited setting on the TV, and output 16-235 from madVR, this clips the OS shades but it does't matter for me.
When using the RGB Full setting & 0-255 with madVR, the TV creates minor banding like its compressing the incoming 0-255 to 16-235 then stretches back to 0-255.
Its' definitely not MadVR or Nvidia that creates the banding, it appears that the TV's processor only accepts 16-235 steps for bandless picture.

All in all, I preserve BTB and WTW using the RGB Limited on the TV & 16-235 with MadVR, and without any banding.
In effect madVR use one less dithering step because the image does not need to be stretch to 0-255, it stays at the native 16-235.

Bottom line is: 0-255 is not always best; it depends on the display device.
I use the Greyscale Ramp from the AVS709HD disc, and engage the "Reduce Banding Artifacts" in madVR for maximum smoothness, then I select the setting on the TV that looks smoothest/bandless.
I suggest you do the same, to know what's the native range the TV processor expects at the input; it may accept 0-255 and show all shades perfectly but create banding as a side effect.
That's very disappointing to hear. It seems that Panasonic has implemented 0-255 support in a very bad way, by stretching it to 16-235 in 8bit. Ouch. Anyway, you're right: Every display behaves differently, so there can only be recommendations what to try first. But in the end everybody should test which setup works best for his particular hardware/display.

One question remains; Can madTPG use 16-235, or is it locked to 0-255?
madTPG uses the same settings you've selected for madVR, which is no wonder since madTPG internally uses madVR for rendering the test patterns.

[James Freeman]

That's very disappointing to hear. It seems that Panasonic has implemented 0-255 support in a very bad way, by stretching it to 16-235 in 8bit.

Found the problem.

When using the "Direct Pixel" for 4:4:4, it disables the internal high-bitdepth processing.
When "direct pixel" is Disabled the Panel chroma is 4:2:2 but the grey ramp is silky smooth 16-235 or 0-255.
Well I guess my previous post meant to say: "Know thy-TV".

I should add that on a plasma the Contrast control is also the Panel Luminance (how bright is the TV), so I can't turn the Contrast up to clip over 235 like in an LCD.
So I prefer to use 0-255 in madVR (now that it's smooth), to clip above 235 so that I will not see unwanted artifacts, but chroma resolution is compromised.


@Madshi,

What do you think:
No Chroma sub-sampling (4:4:4) at 16-235 & BTB+WTW VS 4:2:2 and 0-255 no BTB/WTW.
The first option is what the TV expects and what any player would output, no Luma dithering, 4:4:4.
The second is a PC configuration, has no BTB/WTW, but also dithers the Luma to 0-255 and is 4:2:2 on my TV.

The question is: What will benefit the picture quality: 4:4:4 & no Luma Dithering OR clipped WTW/BTB?

[omarank]

That's technically not possible, unfortunately. I can choose the GPU for OpenCL processing, but I can't choose the GPU for Direct3D. Have you tried renaming your media player? Reportedly sometimes that helps to activate the NVidia GPU instead of the Intel one.


Yes, I have renamed the media player as "mpc-hc1".

The very first implementation of error diffusion algorithm that you made was based on OpenCL, but later the implementation was made using directcompute processing. I think it was due to so many issues that arose during that time with OpenCL processing. Now that Open CL processing works properly on most systems, would you consider offering an optional support for error diffusion algorithms with OpenCL processing?

[madshi]

The second is a PC configuration, has the benefits of 4:4:4 Chroma, but also another dithering stage in madVR.
There are no multiple dithering stages in madVR. There is always exactly one, at the end of all processing.

(The only exception is when a high-bitdepth source must be fed into DXVA for deinterlacing/scaling. In that case madVR has to dither the source down to 8bit for DXVA.)

The very first implementation of error diffusion algorithm that you made was based on OpenCL, but later the implementation was made using directcompute processing. I think it was due to so many issues that arose during that time with OpenCL processing. Now that Open CL processing works properly on most systems, would you consider offering an optional support for error diffusion algorithms with OpenCL processing?
It works properly, but is slower (for error diffusion) than DirectCompute, much slower on AMD hardware. Sorry, no plans to do error diffusion via OpenCL.

[innocenat]

That probably won't help. Playing games works alright on your laptop, using the NVidia GPU? I've no idea what's going on there, to be honest. Probably a driver issue of some sort... :(

Yeah playing game work fine. [I know it work by checking GPU-Z graph that the games use nvidia gpu] DGDecodeNV doesn't work either, but I don't know whose fault it is.

[huhn]

as far as i know repeated or dropped frame forced by vsync are not shown in the OSD.
They should be, I think?


i tested it for repeated frames ( 23p source at 25HZ)and they are shown.

[huhn]

Found the problem.

When using the "Direct Pixel" for 4:4:4, it disables the internal high-bitdepth processing.
When "direct pixel" is Disabled the Panel chroma is 4:2:2 but the grey ramp is silky smooth 16-235 or 0-255.
Well I guess my previous post meant to say: "Know thy-TV"

read this : http://www.hdtvtest.co.uk/news/panasonic-tx-p42st60-201303312779.htm
Full 4:4:4 reproduction (PC) Yes, with [1080p Pixel Direct] enabled, but [HDMI Content Type] must be set to “Photo” to avoid severe sharpening artefacts
worth a try or not:-)
a plasma with 4:4:4 support. i'm getting very very weak right now.

[James Freeman]

read this : http://www.hdtvtest.co.uk/news/panasonic-tx-p42st60-201303312779.htm
worth a try or not:-)
a plasma with 4:4:4 support. i'm getting very very weak right now.

Thanks huhn.

I have tried my PS3 and everything is smooth when viewing the greyscale ramp image on a USB stick with RGB Full mode.
Apparently this banding problem is created when using an Nvidia gpu HDMI in RGB mode & my TV.
Switching to YCbCr444 makes everything smooth again, even with "Direct Pixel" 4:4:4.
Seems like the ST60 doesn't like the RGB Full NVidia has to offer, but with the PS3 it's just fine.

The "HDMI Content Type" is an HDMI 1.4 feature that automatically switches/sets TV functions like Sharpness, Vivid, Color etc...
Read: HDMI Content Type (http://www.hdmi.org/manufacturer/hdmi_1_4/content_type.aspx)
In the Nvidia control panel its in "adjust desktop color settings" -> "content type reported to the display".
I have disabled this feature straight away on my TV and everything is just right.
Sharpness should be on "0" with this model anyway.

Yes, a good Plasma is great fun. ;)


EDIT:
I found this great article which explains all I needed to know about RGB, YCbCr and Range.
http://www.audioholics.com/home-theater-calibration/hdmi-black-levels-xvycc-rgb

IMO:
The ideal output to a TV is RGB Full from GPU and Limited 16-255 from madVR and set the TV to RGB Limited, anything else expands/compresses the video content.
For example when YCbCr is selected in Nvidia CP the OS is compressed to 16-235 and along with it madVR output which already expanded the content from 16-235 to 0-255, in effect a double compressing/expansion, WTW/BTB are clipped.
Same goes for RGB limited at the GPU (Default before madLevelsTweaker) which compresses the OS to 16-235 along with anything on screen including madVR output.

The ideal option when using a HTPC is Full RGB from the GPU and Limited from madVR (or any other software player), then select Limited RGB on the TV.
This preserves the BTB and WTW and does not undergo any expansion or compression.
Brightness and Contrast on the TV will have the same effect as though a stand alone player is connected to a TV with its native YCbCr.
When you calibrate your display with these settings they will also be effective for any other YCbCr devices like Set-Top Box or Blu-Ray Player.

In effect YCbCr's range is 16-235, BTB 0-15, WTW 236-255.
So I decide to stick with RGB Limited on the TV and madVR, and RGB Full at the GPU output, to simulate a stand alone player only x1000 better because of madVR.

[lansing]

What is the native format of that video (resolution, framerate, interlaced/progressive etc)?


I've tried on 24fps, 29.97fps, at 720p, 1080p videos, all had the same problem.

[JarrettH]

Is anyone using Intel DXVA2 scaling? I find it looks too sharp/punchy, however the performance is great. Is it known what scaling Intel is using for this? Even Bicubic 75 AR provides an easier image on the eyes.

[huhn]

should be ~lanczos 3 ar in 8 bit. atleast in older intel driver

[JarrettH]

Oh, how do you know for sure? Just curious, that's all. I might as well not switch then. :thanks:

[michkrol]

Is anyone using Intel DXVA2 scaling? I find it looks too sharp/punchy, however the performance is great. Is it known what scaling Intel is using for this? Even Bicubic 75 AR provides an easier image on the eyes.

Go to Intel HD Graphics Control Panel (or similar-named) and disable (set to application or manual) all image/video enhancement.
Since I don't use iGPU anymore, here's an instruction from Google: https://communities.intel.com/message/240901#240901

[Asmodian]

Oh, how do you know for sure? Just curious, that's all. I might as well not switch then. :thanks:

This was determined by analyzing the images. It isn't the same Lanczos3+AR that madVR uses but it looks fairly similar to it.

I hope michkrol's suggestion works for you. There are a lot of other things the drivers might do to the video to make it look too sharp/punchy.

[JarrettH]

I already had them at application controlled. Do you mean I should try lowering the sharpness in the control panel? :cool:

[Asmodian]

Haha, yeah - turning off sharpness would be a good start. ;)

Turn off everything you can find in Intel's enhancements.

[dansrfe]

madshi, is it possible for you to add DX11 support again? When I use the web installer, it gives me an error saying that a newer version of DirectX has been detected on the system and it cannot proceed. I think there may be a link with DX9 and allowing NVIDIA to take over rendering.

Please let me know if there's anything else I can try or if you require more info. Thanks.

[Anime Viewer]

madshi, is it possible for you to add DX11 support again? When I use the web installer, it gives me an error saying that a newer version of DirectX has been detected on the system and it cannot proceed. I think there may be a link with DX9 and allowing NVIDIA to take over rendering.

Were did you get the DirectX installer from?
Try this one if you haven't already:
http://www.microsoft.com/en-us/download/details.aspx?id=35

[dansrfe]

Were did you get the DirectX installer from?
Try this one if you haven't already:
http://www.microsoft.com/en-us/download/details.aspx?id=35

That's the one I tried.

[iopq]

Hey madshi, not sure if you're aware of the problem of waking up from standby. I am using Windows 7 64 bit, MPC BE 32 bit (obviously), MadVR 0.87.9

When I resume a video after standby it doesn't play. Gives a Direct3D error or something.

I found a post of a user on XP complaining about the same issue: http://forum.inmatrix.com/index.php?showtopic=13646