madVR - high quality video renderer (GPU assisted)

[ajp_anton]

Originally Posted by AndreaMG

I use this for everything and works like a charm (give it a try):

http://micromighty.com/Products/Screen_Mode_Switch.aspx

But then it won't auto-detect the movie fps, it won't go back automatically when I close the media player, and it won't help me figure out what's causing the problem I had a few posts up. Plus it requires yet another external program instead of extending a feature that's already there.

[AndreaMG]

Originally Posted by ajp_anton

But then it won't auto-detect the movie fps, it won't go back automatically when I close the media player, and it won't help me figure out what's causing the problem I had a few posts up.

I'm not following youYou said you wanted madshi to add a feature to MANUALLY change the refresh rate and I simply suggested a program that does exactly that not only under MadVR but under everything, of course it does not detect the actual fps of what is going on screen, for that just use the built-in MPCHC autochange fullcreen monitor mode which in my case gives always the correct results even when MadVR fails

[DragonQ]

Originally Posted by madshi

It means that the DXVA2 APIs report failure for some reason. I don't know why. And it will be hard to fix without having access to such hardware. The same DXVA2 code in madVR appears to work fine for newer Intel GPUs, so I'm wondering if it's maybe a driver issue.

I'm sure I've used MadVR before on this machine (with software decoding and hardware deinterlacing) and have never seen this error before, plus EVR does deinterlacing fine. I'm using the latest drivers.

Anything I can do? Provide a sample, or let you access my laptop remotely perhaps?

[secvensor]

Problems occured:
1.At adjustment the active mark on Jink in image upscaling vanishes.
2.The new screenshot option through madvr does not keep influence of external active filters.

[ajp_anton]

Originally Posted by AndreaMG

I'm not following youYou said you wanted madshi to add a feature to MANUALLY change the refresh rate and I simply suggested a program that does exactly that not only under MadVR but under everything, of course it does not detect the actual fps of what is going on screen, for that just use the built-in MPCHC autochange fullcreen monitor mode which in my case gives always the correct results even when MadVR fails

I said "activating/deactivating the most appropriate refresh rate for the currently playing video". I don't want to need to know the framerate of the video, I was just asking for a way to manually change the refresh rate to the correct one chosen by madVR whenever I want.

[omarank]

Originally Posted by dansrfe

You're using a GTX 260? I thought only >= GTX 660 Ti could handle madVR max settings...?

Well, I think cards >= GTX 260, >= GTX 460, >= GTX 560 and >= GTX 660 should all be able to handle these settings, at least for commercially available content. My TV has native resolution of 1360x768, so madVR always has to do either upscaling or downscaling. I have tested these settings on a variety of content and haven't encountered any serious drop in performance while rendering. GTX 260 has VP2 hardware decoder. It gives up on high bitrate (>100 Mbps) H264 videos, some VC1 videos and of course those fancy 1080p 60 fps and 4k videos.

[pie1394]

Originally Posted by italospain

Can you explain the enhancements exactly.

thanks.

Just like what I have mentioned, the de-ring and de-block processes with per-pixel-adaptive algorithms. It often needs high computation power, but old-day CPUs were not powerful enough. That's why almost no one design the CPU software-based solutions for MPEG2 decoder.

About the noise reduction and image sharpening (like Darbee), I think most people here don't like it unless it does not kill the details. Fortunately the nVidia Pure Video does a good job about these parts, too.

To my eyes that I can see nVidia's Pure Video MPEG-2 decoder indeed does better jobs on handling interlaced MPEG2 video of most DVD and HDTV contents. For very high bit-rate MPEG-2 (>= 20Mbps) and most H.264 HDTV contents, however, I agree that IQ difference is not so much between software and Pure Video decoders.

Originally Posted by omarank

This is really strange. I am too using a GTX 260 in an old system having Pentium D Processor. I also use CUDA for decoding videos. My system is able to handle all kinds of commercially available content with the following settings: chroma upscaling - Jinc 4 taps AR, image upscaling - Jinc 3 taps AR and image downscaling - Catmull Rom AR with Linear Light. My system fails (drop frames) only while playing 1080p 60 fps videos, 4k videos etc. For these files also, it is the hardware decoder which is not able to cope up. Do you have Antialiasing and Anisotropic filtering settings in NVIDIA Control Panel not set to "use 3D Application setting"?

You are right about Jinc3 AR mode on GTX260+ partially. I have retested some contents and just realized the exact cases which GTX260+ is overloaded. For progressive contents, GTX260+ has the enough computation power.

The frame dropping only happens to interlaced contents regardless of video content's format and resolution (720x480 ~ 1920x1080 4:2:0 30fps). So it looks like the pixel-adaptive deinterlacing algorithm on GTX260+ is very computation power demanding.

ps: You can set the card to output 1920x1080 resolution to check this. My TV is 1024x1080 interlaced PDP. But I still need to set it to send out 1920x1080 signal to have best IQ. The Hitachi engineers did different tricks on handling different signal resolution + refreh rates to utilize the panel's best characteristics for this TV.

[madshi]

Originally Posted by ajp_anton

Could you add a keyboard shortcut for manually activating/deactivating the most appropriate refresh rate for the currently playing video?

Why? You're the first/only one requesting this.

Originally Posted by DragonQ

I'm sure I've used MadVR before on this machine (with software decoding and hardware deinterlacing) and have never seen this error before, plus EVR does deinterlacing fine. I'm using the latest drivers.

Anything I can do? Provide a sample, or let you access my laptop remotely perhaps?

Anything you can do? Yes, send me your laptop...

Ok, if you say older madVR builds might have worked, you could try different older madVR builds to find out which exact version worked and with which exact build it stopped working. Then maybe I can do something.

Originally Posted by secvensor

Problems occured:
1.At adjustment the active mark on Jink in image upscaling vanishes.

This has been reported already, but I can't reproduce it on my PC. I need a detailed step-by-step instruction to reproduce the problem.

Originally Posted by secvensor

2.The new screenshot option through madvr does not keep influence of external active filters.

I'm not sure what you mean?

Originally Posted by leeperry

I've been killing time until PotP would support PS script with mVR, comparing screenshots between EVR/mVR/HR with the 6240 MPC build using the test pattern attached to this post(mediafire refuses to accept it and they've killed all my previous uploads of it, lemme know where else I could upload it if need be).

I could never get the same colors in mVR when using YV12 so I decided to output RGB32HQ from ffdshow, I've also disabled all color conversions in mVR.

That's without processing:

-EVR -HR -mVR

They're all identical

And that's with the stock "nightvision script" set on post-scaling, and this is:

-EVR
-mVR with my display set to 0-255
-mVR with my display set to 16-235

I might be doing something wrong! But if that's not the case, improving on the existing PS script support sounds like a great idea but atm getting the exact same colors as with EVR would prove to be extremely useful if any possible please.

I've looked at the nightvision script and found it to be badly written. It actually uses the alpha channel of the renderer output to calculate its colors, and it seems the EVR alpha channel has different data than the madVR alpha channel. The alpha channel is really totally nonsense and should not be used by any sane custom shader script. The current nightvision script has this line:

float c = dot(tex2D(s0, tex), float4(0.2, 0.6, 0.1, 0.1));

Replace this line with the following to remove the incorrect alpha channel use:

float c = dot(tex2D(s0, tex), float4(0.25, 0.65, 0.1, 0.0));

[thedragon1]

hi madshi,
i'm new
i have a laptop (with Nvidia GF GT540M) and try to use madVR. When i played the video , madVR always ran on intel graphic card (HD 3000) and sometime it loaded full (100%) then crashed. Please tell me , how to use madVR for Nvidia card ?
Thanks.
p/s i tried to set high - performace Nvidia processor (Gobal setting) in Nvidia control panel but it's not work
win 8 pro x64

[wanezhiling]

Originally Posted by thedragon1

i have a laptop (with Nvidia GF GT540M) and try to use madVR. When i played the video , madVR always ran on intel graphic card (HD 3000)

http://forum.doom9.org/showpost.php?...ostcount=10533

[pirlouy]

Originally Posted by bjd

..and if you check the footer you will see a 1080p tv listed so I would assume the scaling is actually nearest neighbour (no scaling)

Sorry but I don't get your message. :/

[thedragon1]

Originally Posted by wanezhiling

http://forum.doom9.org/showpost.php?...ostcount=10533

it's work ^^ thanks for helping ^^

[ajp_anton]

Originally Posted by madshi

Why? You're the first/only one requesting this.

Right now, for debugging the problem I'm having. Also you could more easily switch back and forth to see the difference (some screens allow non-60Hz input while still displaying at 60).

But for something actually useful - changing the refresh rate takes time and sometimes messes with background programs. When watching random short videos, I'd rather just stick to 60Hz, even in fullscreen.
IMO it fits nicely with the other on/off keyboard shortcuts already present.

[leeperry]

Originally Posted by madshi

I've looked at the nightvision script and found it to be badly written. It actually uses the alpha channel of the renderer output to calculate its colors, and it seems the EVR alpha channel has different data than the madVR alpha channel.

oops, will try

But as you can imagine the only script that REALLY matters to me at this point is the one that does gamut mapping from AVS, I will embed it within PotP automatic profiles based on frame rate and resolution, save PotP's configuration into an .ini file, make it read-only and roll it when rolling displays = masterplan to success

Its colors when using EVR/VMR9 match pretty much exactly ddcc() even though they both use completely different code AFAIK. There are only very slight differences in the <16 and >235 regions so if you don't crush BTB and WTW anymore as you explained earlier, they might very well end up perfectly identical I guess.

The other PS scripts that would matter to me would be mirroring (but nobody bothered writing one apparently) and levels conversion.....and that's me that my custom levels/gamut/matrix filename tag feature request will be implemented at some point

PS: so, when using the following gamut mapping script:

sampler s0 : register(s0);
float4 p0 : register(c0);

static float4x4 r2r =
{
0.879235699565134,0.10128068131873,0.0194836191161357,0,
0.0448339014270118,0.915215520228331,0.0399505783446596,0,
0.0154998582994896,0.0971141863242463,0.887385955376264,0,
0, 0, 0, 0
};

float4 main(float2 tex : TEXCOORD0) : COLOR
{
    float4 c0 = tex2D(s0, tex);
    c0 = pow(c0, 1/0.45);
    c0 = mul(r2r, c0);
    c0 = saturate(c0);
    c0 = pow(c0, 0.45);

    return c0;
}

With the 6240 build of MPC, feeding RGB32HQ(dithering disabled in ffdshow) and this script set for post-resizing:

1) untouched:

EVR CP:

mVR(dithering and all color conversions disabled):

= identical

2) aforementioned script processed:

EVR CP:

mVR(dithering and all color conversions disabled):

= not identical

[madshi]

Originally Posted by leeperry

The other PS scripts that would matter would be mirroring (but nobody bothered writing one apparently)

Mirroring is extremely easy:

sampler s0 : register(s0);

float4 main(float2 tex : TEXCOORD0) : COLOR
{
  return tex2D(s0, float2(1 - tex.x, tex.y));
}

[DragonQ]

Originally Posted by madshi

Anything you can do? Yes, send me your laptop...

Ok, if you say older madVR builds might have worked, you could try different older madVR builds to find out which exact version worked and with which exact build it stopped working. Then maybe I can do something.

Hmmm...I can't find a previous build that works, even back to 0.61.

That leaves two possibilities: a problem with the latest driver, or MadVR deinterlacing has never worked on Arrandale.

[madshi]

I think it probably never worked. Don't know why...

[leeperry]

Originally Posted by madshi

Mirroring is extremely easy:

sampler s0 : register(s0);

float4 main(float2 tex : TEXCOORD0) : COLOR
{
  return tex2D(s0, float2(1 - tex.x, tex.y));
}

oh my oh my, very cool stuff!

I think someone should create a separate thread with easy access to PS scripts in the OP, much like the one they have on HCFR.

[bjd]

Sorry but I don't get your message. :/

The point I was making if you are watching 1080p source on a 1080p screen and have MadVR set to use Jinc, it will default to nearest neighbour as there is no scaling to perform so rendering times are a lot lower.

On my GT430, Jinc4 upscaling (576p/720p24 -> 1080p24) renders in around 35-40ms which is too slow really.

[pirlouy]

It won't "default nearest neighbour", it will just do no image scaling (but still chroma scaling). But even if I take 576p or 720p, there's no noticeable difference in rendering time.
Indeed, your GT430 is not able to handle Jinc (but it's a very low market GPU), the same goes for Intel HD3000. My point was not to tell all GPU can handle Jinc, but no need to get latest expensive GPU to handle current cases (except if you're a perfectionist which adds several filters).