madVR - high quality video renderer (GPU assisted)

[madshi]

Originally Posted by omarank

All right, I will explain with an example. If I change the source black level from 0 to say +10 using some custom keyboard shortcut, madVR will change the video level from 16-235 to maybe 20-235. I was wondering if madVR could also display this information of current video levels "20-235" along with the increment "+10".

The source black level control is meant to allow you to tweak playback of broken encodes. For such encodes you will probably play around with the black level control until the image looks "right". So why would you need to know the combined video levels + black level adjustment? I don't really see the practical benefit right now. I expect the black level adjustment control to be more like trial and error.

Originally Posted by omarank

BTW, now I have understood how the new contrast control works. However, I don't clearly understand when should this contrast control be used?

Maybe if you have a source which was encoded in a weird way, using the source contrast control could be used to make it look better. For a properly encoded movie, none of the source color controls should be used. Their only purpose is to give you some room to play with badly encoded movies...

[vivan]

Originally Posted by madshi

Interesting, I didn't know that "use a separate device for presentation" had any specific effect on Optimus systems. I've no personal experience with Optimus. I imagine the "use a separate device for presentation" could run into trouble if the OS decided to switch which GPUs it uses behind madVR's back. Could that be happening here? Isn't there any way to force the OS to use one specific GPU? Maybe you can totally disable one of the GPUs?

No, in nVidia Optimus systems intel GPU is always active and outputting image to display. When nVidia GPU is active - it's copying the result into iGPU memory. http://forum.doom9.org/showpost.php?...ostcount=12641
Also I've posted log before and you said

Originally Posted by madshi

The freeze report and debug log don't show me anything that looks wrong to me, except that suddenly out of the blue after about 30 seconds one the Direct3D APIs stops working, which has worked just fine the past 30 seconds. I can't see any obvious reason for that. Of course there's always a chance that there's a bug in madVR somewhere that causes this, but looking at that log I rather think the problem is with the GPU driver.

Interestingly, the crash happens in the "same" API that fails in the freeze. Well, it's not really the same API. The freeze occurs in D3D9, while the crash happens in D3D11. But in both cases the problem occurs when trying to copy a rendered frame into the D3D backbuffer. I don't know what's going on there, but I do think it's a GPU driver problem. The crash happens in an NVidia dll, not directly in madVR.

I guess nothing have changed since then...

[madshi]

Originally Posted by ajp_anton

So how would you use DXVA scaling *always* for everything?

I don't know if that's really necessary or even useful. Chroma upsampling alone is relatively cheap. Using DXVA for only chroma upsampling would require me to follow up the DXVA processing with some added processing steps on my own, so I don't know if using DXVA for chroma upsampling would even bring any performance benefit.

[madshi]

Originally Posted by vivan

No, in nVidia Optimus systems intel GPU is always active and outputting image to display. When nVidia GPU is active - it's copying the result into iGPU memory.

Also I've posted log before and you said I guess nothing have changed since then...

Ah yes, I remember. But I don't remember having read that disabling the "use a separate device for presentation" option might help with Optimus systems. That was news to me. Or maybe I just forgot about it, don't know...

[Hwasin]

For optimus systems. The best configuration is usually to use Nvidia for decode and deinterlacing. Then the Intel GPU for rendering. If you never watch interlaced material though, it might not matter.

[leeperry]

Originally Posted by madshi

The source black level control is meant to allow you to tweak playback of broken encodes. For such encodes you will probably play around with the black level control until the image looks "right". So why would you need to know the combined video levels + black level adjustment? I don't really see the practical benefit right now. I expect the black level adjustment control to be more like trial and error.

I was the one who came up with the original request, so I guess I might as well just hop on

I didn't really see how you could just play around blindly with saturation, but OK some movies might look undersaturated and that'd be a quick fix...but setting up the levels conversion blindy? A real world scenario for my custom levels filename tagging is movies that went through several successive PC>TV conversions for instance, I currenty use this script in avisynth:

SmoothLevels(preset="tv2pc")
#SmoothLevels(input_low=30,input_high=218)
#SmoothLevels(input_low=42,input_high=204)
#SmoothLevels(input_low=52,input_high=192)

Those figures aren't random at all, they reflect dual/double/triple TV>PC conversion for those movies that still look washed out after a single TV>PC conversion(or two/three for that matter).

So the real world solution for me to "fix" those movies w/o relying on 8bit avisynth processing will be to add levels conversion PS scripts using those very values. Otherwise, they could also be used via those new mVR settings you were mentioning....playing around with custom levels blindly might be possible for black level but hard to calibrate for white IMO, so at least a visible indication of where you are currently standing might help

[DragonQ]

Originally Posted by Hwasin

For optimus systems. The best configuration is usually to use Nvidia for decode and deinterlacing. Then the Intel GPU for rendering. If you never watch interlaced material though, it might not matter.

How? By setting LAV to CUDA and MPC-HC to use HD 4000?

[ajp_anton]

Originally Posted by madshi

I don't know if that's really necessary or even useful. Chroma upsampling alone is relatively cheap. Using DXVA for only chroma upsampling would require me to follow up the DXVA processing with some added processing steps on my own, so I don't know if using DXVA for chroma upsampling would even bring any performance benefit.

Well, even if chroma itself is cheap, there's also the conversion to RGB. But yeah, i forgot DXVA can potentially mess that up.

[HoP]

@madshi
i have some TS files like this↓
sample - 36 Sec.trimmed with Avidemux
when seeking ,MPC-HC crash.[madVR + Lav DXVA2 native]
If I use LAV Decoder in "software mode" the video plays fine.
EVR CP + Lav DXVA2 native,no problem.

[Hwasin]

Originally Posted by DragonQ

How? By setting LAV to CUDA and MPC-HC to use HD 4000?

Exactly

[DragonQ]

Originally Posted by Hwasin

Exactly

Hmm. Probably not worth it though - the only advantage really is probably being able to have anti-ringing on both chroma and luma, in exchange for probably a significant decrease in battery life.

[dansrfe]

madshi,

Is there a way to use MPC-HC, madVR, and LAV Filters on unix, OSX specifically? Possibly a way to port it? I don't think there's anything that can be done but thought I'd ask since a friends of mine who uses osx likes the setup with madVR, MPC-HC, and LAV Filters that I showed her.

I was wondering if it might work well in a Windows 7 virtual machine as a possibility. Would madVR, MPC-HC, and LAV Filter be able to perform well/appropriately in a virtual machine? The macbook pro is from 2010 mid-level specs. i5-520M, 4GB DDR3, GeForce GT 330M w/ 256MB VRAM.

Thanks!

[ajp_anton]

You could always dual-boot with W7, but it depends on how much time you're willing to spend on rebooting all the time.

[dansrfe]

Originally Posted by ajp_anton

You could always dual-boot with W7, but it depends on how much time you're willing to spend on rebooting all the time.

Yeah, that's the thing... Maybe if put dual boot on her laptop she might end up spending more time on Windows and eventually switch over lol. Problem solved then.

[6233638]

Some VM software like parallels offers 3D support, but I suspect it is not going to work well with madVR. The easiest solution would definitely be dual-booting, and I would recommend Windows 8 due to its low price, and much quicker boot times.

[konakona]

Hello everyone-my first post here!
Id like to thank you for all the work you did to make madvr possible.

I've got a few questions about it,tho:
First of all,my pc is winxp 32bit,amd phenom II x2 3,1ghz,and radeon HD4870(latest drivers),and i got 4gb of ddr3 ram.

Deinterlacing-activating deinterlacing on any type of content (from dvd to 1080i) makes my pc to drop half of the frames. Why is that so? Is my pc lacking resources to deinterlace PAL dvds?
Also,i noticed that deinterlaced video is brighter and got somehow more vivid colors,why is that so?

I'd also like to ask what exacly are delayed frames,they usually happen a lot when im watching hi10p anime (720p),but it rarely drops any frames (at end of playback you can have like 1000 or more delayed frames),if that helps,im running dell 2209wa on 76,416hz using powerstrip.
Is it possible that 720p hi10p video needs more resources than full hd material that doesnt drop or delay any frames during the playback?

Im using LAV decoders and splitter(im streaming audio through s/pdif) ,and Haali media splitter for mkv files (i tried to increase its buffer size to 192000,but didnt help in anything i think).

Id be glad if you could answer me,also keep up good work,this is awesome renderer!

edit:im using latest mpc-hc

[DragonQ]

What are your settings for deinterlacing in the Catalyst Control Centre?

[konakona]

Originally Posted by DragonQ

What are your settings for deinterlacing in the Catalyst Control Centre?

ive tried both by setting it manually and by leaving control to application,when deinterlacing is on in the madvr settings i drop so many frames its unwatchable

[madshi]

Originally Posted by leeperry

I didn't really see how you could just play around blindly with saturation, but OK some movies might look undersaturated and that'd be a quick fix...but setting up the levels conversion blindy? A real world scenario for my custom levels filename tagging is movies that went through several successive PC>TV conversions for instance, I currenty use this script in avisynth:

SmoothLevels(preset="tv2pc")
#SmoothLevels(input_low=30,input_high=218)
#SmoothLevels(input_low=42,input_high=204)
#SmoothLevels(input_low=52,input_high=192)

Those figures aren't random at all, they reflect dual/double/triple TV>PC conversion for those movies that still look washed out after a single TV>PC conversion(or two/three for that matter).

Hmmmm... How about changing the "source levels" control then to not only allow fullrange and limited range, but also "limited range (double expanded)" and triple expanded. Does quad expanded (52-192) really exist!?!?

BTW, those numbers confuse me slightly. The formula to expand is "newValue = oldValue * 219 / 255 + 16". So if we do the math, we should end up with this:

fullrange: 0 - 255
limited range: 16 - 235
double expanded: 29.74 - 217.82
triple expanded: 41.54 - 203.07
quad expanded: 51.68 - 190.40

Originally Posted by konakona

Deinterlacing-activating deinterlacing on any type of content (from dvd to 1080i) makes my pc to drop half of the frames. Why is that so? Is my pc lacking resources to deinterlace PAL dvds?

The 4870 should be powerful enough. Can you please make a screenshot of the madVR debug OSD when this happens (Ctrl+J)? Or you can also write down all the numbers, if you prefer that over a screenshot.

Make sure you don't force anti-aliasing, anisotropic filtering and that kind of 3D stuff in the GPU drivers. That might help.

Originally Posted by konakona

Also,i noticed that deinterlaced video is brighter and got somehow more vivid colors,why is that so?

That shouldn't happen. Are you using v0.85.2? Or still v0.85.1? Try with v0.85.2. Or try with the next build, once it's out (not yet).

Originally Posted by konakona

I'd also like to ask what exacly are delayed frames,they usually happen a lot when im watching hi10p anime (720p),but it rarely drops any frames (at end of playback you can have like 1000 or more delayed frames)

Delay frames are frames that were displayed slightly later than planned. It's not as bad as a dropped frame, but it's not really good, either. I suppose this happens mostly in windowed mode? What about fullscreen exclusive mode? Do you get delayed frames there, too? Do you also get the deinterlacing dropped frames there?

Originally Posted by konakona

if that helps,im running dell 2209wa on 76,416hz using powerstrip.

Is there a reason you're using this exact refresh rate? Theoretically, for PAL content 75.00Hz would be better. Ideal would be 50.00Hz.

Originally Posted by konakona

Is it possible that 720p hi10p video needs more resources than full hd material that doesnt drop or delay any frames during the playback?

That depends on the resolution of your monitor. If madVR needs to scale 720p up, but doesn't need to scale full hd material at all, then madVR has more work to do with 720p content. If madVR needs to scale up both then it shouldn't matter so much.

Originally Posted by HoP

i have some TS files like this↓
sample - 36 Sec.trimmed with Avidemux
when seeking ,MPC-HC crash.[madVR + Lav DXVA2 native]
If I use LAV Decoder in "software mode" the video plays fine.
EVR CP + Lav DXVA2 native,no problem.

I've tried that sample on my win7 x64 PC with NVidia 9400 GPU. Using the latest LAV version for native DXVA2 decoding works fine here with that sample. I've seeked like 10 times, no issues at all. Are you sure you're using the latest LAV version? Which OS and which GPU are you using?

[konakona]

http://img248.imageshack.us/img248/3100/debugosd.jpg

here is the OSD after 5 seconds of playback with deinterlacing on. without it there would be like 0/0 dropped/delayed frames,you can imagine the stuttering!

Originally Posted by konakona
Is it possible that 720p hi10p video needs more resources than full hd material that doesnt drop or delay any frames during the playback?
That depends on the resolution of your monitor. If madVR needs to scale 720p up, but doesn't need to scale full hd material at all, then madVR has more work to do with 720p content. If madVR needs to scale up both then it shouldn't matter so much.

to be exact,i was comparing 720p hi10p with non hi10p 1080p- the hi10p 1080p videos are completly unwatchable-i lose more than half of frames (but it might be that my pc is just too weak for hi10p 1080p) upscaling/downscaling would happen in both cases as im on 16:10 1680x1050 resolution (im using 4taps spline for both chroma and luma up/downsampling,tried to switch to other ones,no real difference in dropped/delayed frames)

ive been trying to force 72hz on this screen for fluid playback of hd 23,976fps materials,but it will take me some time to figure right values in powerstrip :3 (this panel is capable of such refresh rates,it doesnt drop frames)

edit: i just upgraded from v0.85.1 to v0.85.2,deinterlacing still looks bright and shiny