madVR - high quality video renderer (GPU assisted)

[gebla]

:helpful:I am trying to set a 3dlut file to correct my display. I added my measurement values in the settings and pressed apply. Madvr asks if I want to downlaod and install yCMS now? I press yes download starts and madvr reports

Downloading yCMS failed.

How can I get yCMS installed? (windows 7 32bit)

OK I found out myself. I copied madvr into program files and installed it from there. However program file sis protected and therefore madvr could not install ycms. I changed location of madvr outside program files and it si working now

[huhn]

you could try argyllCMS when your meter is supported: http://www.avsforum.com/forum/139-display-calibration/1471169-madvr-argyllcms.html

not an answer to your question sorry.

[StinDaWg]

he said something like it is not high on his "list" but sample are always welcome.
If he shows up here I'll provide a sample. No point doing it now as it might get looked over.

[a8213711]

this picture can help
http://madshi.net/madVR/ChromaRes.png
watch it unscaled @ 100 % zoom. without PC mode you shouldn't be able to see the 4:4:4.

i guess something is blocking madVR from switching to 50 hz. try it without any other program in the background.

In Samsung TV's, PC mode is available only at 60hz and when you select the HDMI (or the input you are using) to be called PC or rename it PC. The picture above should clear things up.

I've used that ChromaRes.png :
with Display at 59.9Hz I see only 444,
with MPC closed I glimpse 422 but definetely see 444,
at 50Hz (using the WMC/madVR bug) is like MPC closed,
at 50Hz (directly) like MPC closed,
at 23.9Hz like MPC closed.
Doesn't this mean with all the above Hz my TV is in PC Mode? And if it's so, why the image become more saturated switching back from WMC which take me from 59.9Hz to 50Hz?


Replying to huhn: without any program in background wtv files are still at 59.9Hz.

[panetesan2k6]

Hi.

I want to test different upscaling methods and hardware combinations with 720p content up to 1080p and I'm in the need for some sharpness pattern in native 720p... I just can't find any, it's really odd.

My goal is to compare the upscaling results from MadVR with those from my TV. With some high bitrate files I prefer to set the desktop to 720p to lighten CPU workload, but I don't know how much quality I'm losing to this, with movies is very difficult to notice. With a sharpness patterns it would be easier.

Does anybody know where can I grab such a test video file?

[huhn]

Hi.

I want to test different upscaling methods and hardware combinations with 720p content up to 1080p and I'm in the need for some sharpness pattern in native 720p... I just can't find any, it's really odd.

My goal is to compare the upscaling results from MadVR with those from my TV. With some high bitrate files I prefer to set the desktop to 720p to lighten CPU workload, but I don't know how much quality I'm losing to this, with movies is very difficult to notice. With a sharpness patterns it would be easier.

Does anybody know where can I grab such a test video file?
why should setting your desktop to 720 lighten your CPU workload?

[Asmodian]

It doesn't do that for me. It flickers for like 2 seconds (just like when starting a game) and then it's perfectly stable and not flickery at all.

But yes, it doesn't do anything since madVR is adjusting the framerate it sends based on the monitor refresh rate it reads...

Also, on my machine with two different GPUs (gtx 670 and gtx 970) it results in abnormal 100% GPU usage (only with video playback and madVR as a renderer) resulting in very low performance (insufficient for upscaling properly with jinc 3 for example). That's purely a G-Sync/madVR bug though, maybe it doesn't even happen with all driver versions. But it's not worth trying to fix it for reasons explained above.

And the same thing happens to me; terrible rendering performance when G-sync is enabled, ~12ms to ~40ms with G-sync.

I cannot reproduce flickering either. :confused:

[panetesan2k6]

why should setting your desktop to 720 lighten your CPU workload?

(Maybe I'm wrong, but I understand the process as follows)

Because I'm not using the GPU to upscale 720p to 1080p, but the renderer (madVR custom pixels shaders) and if I'm not wrong, that relays on the CPU, hence the sloppy playback and frame dropping that I get when upscaling to 1080p via custom pixels shaders, specially in high bitrate files.

So, in other words:

720p to 1080p performed by GPU: Smooth.
720p to 1080p performed by CPS (ie Lanczos 3 taps): Sloppy.

720p with no upscaling (720p desktop fed to TV that TV upscales to 1080 panel): Smooth.

Yes, I can get smooth playback with some options, but I would like to test the sharpness with each one of them. ;)

[nevcairiel]

Pixel Shaders run on the GPU.

And while reducing your resolution will result in less work for the renderer, it'll also cripple your performance, as the screen has to do the upscaling, in questionable quality at best..
You are better off using cheaper upscalers in madVR than letting your TV do it.

[romulous]

madshi: A small feature request. I know why you have done things this way, but please give an option to disable 'on top' for the madVR settings dialog. With it sitting on top of all other windows (and it has no minimise button either), it makes it more difficult to help other people with problems with madVR (for example, you open the options dialog to find the name of an option, and then want to type that name into an email - you can't until you then drag that window out of the way somewhere, and you have to do this each time you open it as it always opens in the middle of the screen).

Thanks!

[bcec]

What's the recommended upscaler to use for upscaling 1080p to 4k UHD? Nedi? Jinc?

[nevcairiel]

What's the recommended upscaler to use for upscaling 1080p to 4k UHD? Nedi? Jinc?

NNEDI3 is ideal for perfect resolution doubling (from 1080p to 2160p), it does however require the most performance.

Otherwise, it greatly depends on your taste, some prefer sharpness, some prefer less aliasing/ringing.

Jinc is a good alternative if NNEDI is too much for your GPU, as are Lanczos or Spline

[Anteys]

Nvidia's 347.25 WHQL (http://www.geforce.com/drivers) driver released today and on my system it improved average rendering times running the same settings in MadVR by 2-3ms, which means I can leave the card at stock clocks instead of having to OC it to avoid dropped frames in some cases (mainly high fps TV recordings).

[bcec]

NNEDI3 is ideal for perfect resolution doubling (from 1080p to 2160p), it does however require the most performance.

Otherwise, it greatly depends on your taste, some prefer sharpness, some prefer less aliasing/ringing.

Jinc is a good alternative if NNEDI is too much for your GPU, as are Lanczos or Spline

Cheers man. I will give NNEDI3 a try when my new gtx 970 arrives. Any particular settings within NNEDI3 I should pay attention to?

[nevcairiel]

As much neurons as your GPU can run at smoothly, that's all.

[cyberbeing]

Nvidia's 347.25 WHQL (http://www.geforce.com/drivers) driver released today and on my system it improved average rendering times running the same settings in MadVR by 2-3ms, which means I can leave the card at stock clocks instead of having to OC it to avoid dropped frames in some cases (mainly high fps TV recordings).
On my system with a GTX 770, this driver is bugged with madVR "Error Diffusion - Option 1" dithering. As far as I can tell, this issue does not occur with any of the other dithering modes.

Thin vertical dotted line columns are displayed over the video, evenly spaced exactly 16px apart. (http://i.imgbox.com/BOi3anKj.png) Same issue with 347.12 which was bundled with the Cuda 7 RC, which was the first public driver to have this problem. The previous 347.09 WHQL geforce driver and prior have no issues.

Can anyone else reproduce this?

[rack04]

On my system with a GTX 770, this driver is bugged with madVR "Error Diffusion - Option 1" dithering. As far as I can tell, this issue does not occur with any of the other dithering modes.

Thin vertical dotted line columns are displayed over the video, evenly spaced exactly 16px apart. (http://i.imgbox.com/BOi3anKj.png) Same issue with 347.12 which was bundled with the Cuda 7 RC, which was the first public driver to have this problem. The previous 347.09 WHQL geforce driver and prior have no issues.

Can anyone else reproduce this?

I have the exact same problem.

[cyberbeing]

Ok, thanks rack04. Looking through you post history, I assume that was with your GeForce GTX 650 Ti. If so, that probably confirms that at least all Kepler GPUs are likely affected.

@madshi
Would you like me to create a madVR bug tracker entry for this, even though it seems to be an NVIDIA driver bug?

[nevcairiel]

I'm not seeing it on my GTX 980 or my GTX 750 Ti.

[panetesan2k6]

Pixel Shaders run on the GPU.

And while reducing your resolution will result in less work for the renderer, it'll also cripple your performance, as the screen has to do the upscaling, in questionable quality at best..
You are better off using cheaper upscalers in madVR than letting your TV do it.

Thanks for your answer, very clarifying ;)

Anyways, as I couldn't get any 720p video test file, I used the LAV filters capability to display PNG images, so I cropped some 1080p sharpness patterns down to 720p 1:1 and then upscaled them to see how much "damage" you get with each upscaler. The MPC-HC, MadVR and LAV treat the png file as it was a video, so it was easy to toggle upscalers with the keyboard and watch the differences. My TV does some bilinear upscaling, I think, but I prefer the results from the renderer :D

If anybody wants to tinker with the files I cropped, they are attached in this post.

[James Freeman]

Can anyone else reproduce this?

Yes.
GTX660

[panetesan2k6]

Wow... I just updated the Catalyst drivers on my laptop and they let me choose 8/10/12 bpc for my TV without any issues... How is this possible? I always thought 10bpc panels (not to mention 12bpc) were the high-end products, and my tv is just a Series 6 Samsung from 2009, wich I assume is 8bit...

[mbordas]

Is there any interest in adding "NNEDI3 image doubling on/off" to the OSD? It would be a similar idea to "Deinterlacing on/off". It would make testing a little easier.

[huhn]

this is just the signal send to the TV. has nothing to do with the panel or what ever. madVR send 8 bit to the GPU driver so nothing to gain there. set it to 8 bit and hope it is the untouched picture madVR send to the GPU.

[snowpiercer]

Is there any interest in adding "NNEDI3 image doubling on/off" to the OSD? It would be a similar idea to "Deinterlacing on/off". It would make testing a little easier.

You could create two profiles (only one with nnedi3) and assign two different hotkeys to these profiles, then you could easily switch between nnedi3 on/off.

[rack04]

Ok, thanks rack04. Looking through you post history, I assume that was with your GeForce GTX 650 Ti. If so, that probably confirms that at least all Kepler GPUs are likely affected.

@madshi
Would you like me to create a madVR bug tracker entry for this, even though it seems to be an NVIDIA driver bug?

Sorry for not elaborating. I'm experiencing this with a GTX 680. Selecting a dithering option other than "Error Diffusion - Option 1" solves the problem. I rolled back to 347.09.

[mbordas]

You could create two profiles (only one with nnedi3) and assign two different hotkeys to these profiles, then you could easily switch between nnedi3 on/off.

right, but what I mean is, I can't tell if image doubling is actually in affect ie, if I set it to "only if scaling factor is 2x", I can test it with different sources but am never sure if it's really activated or not. Unless I see it starting dropping frames that is :)

[huhn]

if you can't see the difference why should you use it?

[friver]

After updating my Nvidia drivers to the most recent version 347.25 from 22.1.2015, using MadVR dithering "Error Diffusion - option 1" shows these artifacts:
http://i.imgur.com/WfBbsmq.jpg (take a look at vertical dotted lines)

For now, I've returned back to 347.09 driver, and everything works just fine.

Is it Nvidia or MadVR bug?

----
GeForce GTX 650 Ti BOOST

[bejita7]

After updating my Nvidia drivers to the most recent version 347.25 from 22.1.2015, using MadVR dithering "Error Diffusion - option 1" shows these artifacts:
http://i.imgur.com/WfBbsmq.jpg (take a look at vertical dotted lines)

For now, I've returned back to 347.09 driver, and everything works just fine.

Is it Nvidia or MadVR bug?

----
GeForce GTX 650 Ti BOOST

Same thing here, GTX 760, changed to Option 2 for a while.

[Patrik G]

what an amazing video renderer this is! :)

i just discovered it 2 weeks ago just when i had bought the new Pioneer BDP LX58 blu ray player.
that blu ray player was a waste of money and i sold it right away when i compared the picture quality with madvr.

with a calibrated 9200 patches 3D LUT the picture quality is amazing.
the LX58 player doesnt even come close in details and color sharpness in the picture.

so you know that you have the best picture quality money can buy with this solution :)

these are the settings that i use

Player: MPC-HC

Graphics card: GTX 780
GPU Usage: 50%

Madvr settings:
3D LUT calibrated with 9200 patches

Scaling Algorithms:
Chroma upscaling: NNEDI3 128 neurons
Image Doubling: Always use NNEDI3 to double Luma resolution (128 neurons)
Image upscaling: DXVA2
Image downscaling: DXVA2

Rendering:
Smooth Motion: Always Enabled
Dithering: Error Diffusion - Option 2
Trade Quality for Performance: none


i just have one wired issue
when i change chapter in MPC-HC sometimes it seems that the video gets out of sync and starts to stutter and speed up the video like x2 fast forward
is there a solution to fix it?

one photo from my KRP-500A with MPC-HC and madvr with 9200 patches 3D LUT calibration.
http://privat.bahnhof.se/wb192876/Frozen10.jpg

[ikakun]

Scaling Algorithms:
Chroma upscaling: NNEDI3 128 neurons
Image Doubling: Always use NNEDI3 to double Luma resolution (128 neurons)
Image upscaling: DXVA2
Image downscaling: DXVA2

Rendering:
Smooth Motion: Always Enabled
Dithering: Error Diffusion - Option 2
Trade Quality for Performance: none


A few suggestions:

Scaling Algorithms:
Chroma upscaling: NNEDI3 16 or Jinc3AR would suffice. More than that is a waste of GPU power without noticeable effect on picture/image quality.
Image Doubling: As high as your GPU can handle.
Image upscaling: Use other scaling method than DXVA2, like Jinc3AR for example.
Image downscaling: Most people here, AFAIK, recommends Catmull-Rom AR LL and/or lanczos(don't know what taps they use though or whether if AR & LL is needed on it).

Rendering:
Smooth Motion: "only if there would be motion judder without it"


i just have one wired issue
when i change chapter in MPC-HC sometimes it seems that the video gets out of sync and starts to stutter and speed up the video like x2 fast forward
is there a solution to fix it?


What is your general settings?
You can try enabling "delay playback until render queue is full", "separate device for presentation" and/or "enable full screen exclusive mode" and see if that helps.

[Asmodian]

On my system with a GTX 770, this driver is bugged with madVR "Error Diffusion - Option 1" dithering. As far as I can tell, this issue does not occur with any of the other dithering modes.

Thin vertical dotted line columns are displayed over the video, evenly spaced exactly 16px apart. (http://i.imgbox.com/BOi3anKj.png) Same issue with 347.12 which was bundled with the Cuda 7 RC, which was the first public driver to have this problem. The previous 347.09 WHQL geforce driver and prior have no issues.

Can anyone else reproduce this?

I have been unable to reproduce this on my GTX Titans with 347.25. ED1 looks fine. Is your image exaggerated or enhanced in any way? I don't see anything like that even using 2-bit output.

[Asmodian]

Graphics card: GTX 780
GPU Usage: 50%

Madvr settings:
3D LUT calibrated with 9200 patches

Scaling Algorithms:
Chroma upscaling: NNEDI3 128 neurons
Image Doubling: Always use NNEDI3 to double Luma resolution (128 neurons)
Image upscaling: DXVA2
Image downscaling: DXVA2

Rendering:
Smooth Motion: Always Enabled
Dithering: Error Diffusion - Option 2
Trade Quality for Performance: none


madVR is not doing what you think it is. DXVA2 upscaling overrides all other upscaling settings, it does not allow image doubling or chroma scaling. DXVA2 is better for low end GPUs that cannot run madVR's high quality scaling options fast enough. It basically hands the video to the GPU's hardware scaler and the actual algorithm used is dependent on the GPU.

However, if luma is not being scaled then DXVA2 image scaling doesn't activate and the chroma scaling option is used. This means that you are getting NNEDI3 128 chroma scaling when watching 1080p but your luma doubling settings will never do anything and you don't get NNEDI3 chroma scaling when watching any other resolution source.

I am pretty sure a GTX 780 cannot do 128 neuron NNEDI3 for both chroma scaling and luma doubling when watching 720p on a 1080p screen. Definitely not at only 50% GPU usage. My Titan takes >55 ms to render each frame using your settings (except using bilinear instead of DXVA2) playing 720p24 at 1080p120 with a GPU load >80%, resulting in lots of dropped frames. I was also using bilinear downscaling. bilinear was chosen for speed not quality.

I suggest you turn off luma doubling and switch to Jinc3+AR for image upscaling and Catmull-Rom+AR+LL for image downscaling. If you only watch 1080p/bluray then luma doubling will never activate anyway. Or you could setup profiles and test what your system can really do with different resolution sources.

You can use ctrl-J to display madVR's OSD, this will tell you how long it takes to render a frame, how many frame drops, etc. you have, along with a lot more information. If the rendering time does not change when switching from bilinear to Jinc you know that option is not being used. ;)

Edit: I should add that the suggestions I gave will not change 1080p viewed at 1080 in anyway. When the source resolution is the same as the viewing resolution only the chroma upscaling setting has an effect.

[Canadane]

After updating my Nvidia drivers to the most recent version 347.25 from 22.1.2015, using MadVR dithering "Error Diffusion - option 1" shows these artifacts:
http://i.imgur.com/WfBbsmq.jpg (take a look at vertical dotted lines)

For now, I've returned back to 347.09 driver, and everything works just fine.

Is it Nvidia or MadVR bug?

----
GeForce GTX 650 Ti BOOST

I just recovered my password to post the same issue.
Using Option 2 for the time being has been my workaround rather than downgrading the drivers.

[cyberbeing]

I have been unable to reproduce this on my GTX Titans with 347.25. ED1 looks fine. Is your image exaggerated or enhanced in any way? I don't see anything like that even using 2-bit output.
Interesting that the Titan doesn't have this issue. I wonder if that is unique to the GK110 Kepler silicon. Does anyone have a GTX 780 or GTX 780 Ti to confirm they are also unaffected?

So far, it seems that all GK104, GK106, GK107 GPUs in the Kepler range are affected.

That image was not exaggerated or enhanced in any way. This is just how ED1 appears in the 347.12 & 347.25 driver for the GPUs which experience this driver bug. You can see another example, in the madVR bug report someone created: http://bugs.madshi.net/view.php?id=250

[Asmodian]

That image was not exaggerated or enhanced in any way. This is just how ED1 appears in the 347.12 & 347.25 driver for the GPUs which experience this driver bug. You can see another example, in the madVR bug report someone created: http://bugs.madshi.net/view.php?id=250

Wow! I wouldn't miss that. A strange bug to only affect ED1.

[bcec]

NNEDI3 is ideal for perfect resolution doubling (from 1080p to 2160p), it does however require the most performance.

Otherwise, it greatly depends on your taste, some prefer sharpness, some prefer less aliasing/ringing.

Jinc is a good alternative if NNEDI is too much for your GPU, as are Lanczos or Spline

Wow, so my GTX 970 can handle 1080p -> 4K using NNEDI3 32 neurons fine. No frame drops, smooth playback at 24Hz.

BUT, in about 10 minutes playing the video, the temperature of my graphics card gets to the point where it is going to almost melt. Fans start running like crazy. Same thing happens with 16 neurons.

Is this normal?

[TheElix]

Rendering:
Smooth Motion: Always EnabledWhat other guys suggested are correct. I also suggest you putting this option here: "only if there would be motion judder without it". But most important, use correct refresh rate on your plasma to watch Blu-Ray movies (that would be 23 Hz in your control panel). There's an option to switch automatically to the correct resolution each time you start a video (well, in cases the correct source framerate is given in MediaInfo in the video file, anyway). Go to Display -> Video Modes -> And put this in the box:
1080p23, 1050p50, 1080p59

You may also try setting a custom resolution for your plasma 1920x1080x47.952 Hz or 1920x1080x71.928 Hz if you want to try SVP. Although, IIRC, Kuro has a great frame interpolation algorithm of its own.

[ryrynz]

What other guys suggested are correct. I also suggest you putting this option here: "only if there would be motion judder without it". But most important, use correct refresh rate on your plasma to watch Blu-Ray movies (that would be 23 Hz in your control panel). There's an option to switch automatically to the correct resolution each time you start a video (well, in cases the correct source framerate is given in MediaInfo in the video file, anyway). Go to Display -> Video Modes -> And put this in the box:
1080p23, 1050p50, 1080p59

You may also try setting a custom resolution for your plasma 1920x1080x47.952 Hz or 1920x1080x71.928 Hz if you want to try SVP. Although, IIRC, Kuro has a great frame interpolation algorithm of its own.

Indeed. ikakun already suggested the Smooth Motion setting above.

Plasma's IME (I haven't tested them all) look terrible with lower refresh rates. I would recommend setting the panel to the highest available (generally 60Hz)

[ikakun]

Wow, so my GTX 970 can handle 1080p -> 4K using NNEDI3 32 neurons fine. No frame drops, smooth playback at 24Hz.

BUT, in about 10 minutes playing the video, the temperature of my graphics card gets to the point where it is going to almost melt. Fans start running like crazy. Same thing happens with 16 neurons.

Is this normal?

It's normal because doing an image doubling using NNEDI3 32 for 1080p -> 4K requires a huge amount of GPU horsepower normally required in playing games and other GPU intensive applications.

[khanmein]

Wow, so my GTX 970 can handle 1080p -> 4K using NNEDI3 32 neurons fine. No frame drops, smooth playback at 24Hz.

BUT, in about 10 minutes playing the video, the temperature of my graphics card gets to the point where it is going to almost melt. Fans start running like crazy. Same thing happens with 16 neurons.

Is this normal?

what kinda video u playing? no such issue happened on GTX 970.

[Patrik G]

You can try enabling "delay playback until render queue is full", "separate device for presentation" and/or "enable full screen exclusive mode" and see if that helps.

thanks
i will check it out

[Stereodude]

Plasma's IME (I haven't tested them all) look terrible with lower refresh rates. I would recommend setting the panel to the highest available (generally 60Hz)
My Samsung F8500 takes a 1080p24 and displays it at 96Hz. It has less flicker than 1080p60 displayed 60Hz. It won't accept 48Hz or 72Hz (per the EDID).

[Shiandow]

Hi all, not sure how many of you've been following the developments over at MPDN but I designed a somewhat interesting chroma scaling algorithm called SuperChromaRes. Eventually this might be ported to MadVR, but since it seems like it might take a while I have decided to release the experimental shaders that I used in the intial development (see here (http://forum.doom9.org/showthread.php?p=1682901#post1682901)), these can be used with MadVR, but to quote that post:

Be warned that support for these experimental shaders will be minimal, I will not be backporting all the improvements made in the renderscript, nor will I explain all the options, they also have some of the same issues as ChromaNEDI but will generally work well for HD sources.

That said, I have used them myself and they generally worked quite well with MadVR, just beware that they might degrade quality in some cases.

[romulous]

Thanks Shiandow :) I've been hoping that they may reach madVR one day - at least for the material I view (YMMV), MPDN has eclipsed madVR's quality on my system by quite a noticeable margin.

[panetesan2k6]

But most important, use correct refresh rate on your plasma to watch Blu-Ray movies (that would be 23 Hz in your control panel). There's an option to switch automatically to the correct resolution each time you start a video (well, in cases the correct source framerate is given in MediaInfo in the video file, anyway). Go to Display -> Video Modes -> And put this in the box:
1080p23, 1050p50, 1080p59

Careful with that, because tvs like Samsung jump out of PC mode (so they go back to subsampling chroma) when fed with other than 60hz signals. And if the TV is subsampling chroma I suppose we are missing the point of using MadVR in the first place...

[huhn]

Careful with that, because tvs like Samsung jump out of PC mode (so they go back to subsampling chroma) when fed with other than 60hz signals. And if the TV is subsampling chroma I suppose we are missing the point of using MadVR in the first place...

a point not the whole.

[bcec]

what kinda video u playing? no such issue happened on GTX 970.

Are you also doing 1080p to 4k upscaling? I was using a 1080p blu-ray remux. But happens on all mkvs when I use NNEDI3 to upscale.

I think it may be something inherent in the design of these 9 series. They get hot quickly from what I am reading. I am returning my card, and will wait for 950. Supposedly that will use only 75 watts so shouldn't get that hot.

[Asmodian]

The 950 will not be fast enough to do NNEDI3 1080p to 4K at all so it should not get hot but why not simply turn off NNEDI3 doubling? :confused: