madVR - high quality video renderer (GPU assisted)

[6233638]

Since my main PC died (lightning ...) I am trying to use my humble Atom netbook for media purposes, it does have HDMI. madVR is unusable in it I'm afraid. DXVA scaling results in out of order frames for a couple of seconds and eventually a BSOD. Bilinear is the only option, but it still drops frames. Only EVR Custom in MPC-HC works correctly, set in Bilinear mode too. The GPU is an Intel GMA 3600. I suspect it has rather crappy drivers too. No newer version either on either the manufacturer's site or Intel's.If you're stuggling for performance, there are some quality options available: don't use dithering use 10bit chroma buffer instead of 16bit use 10bit image buffer instead of 16bit
You might also want to try turning up the CPU & GPU queue sizes, but leave the backbuffer/present options at their defaults.

Be sure to avoid any of the display calibration options such as 3DLUTs, and try using CPU decoding instead of DXVA2. (if you're using DXVA)

If that still doesn't work, changing the flush settings might help.

[DragonQ]

If you can only have anti-ringing on either chroma or luma (for performance reasons), which is more useful?

[omarank]

Hi Madshi, will it be possible to show the current video levels when we change the source white and black levels, as pointed out by Leeperry earlier? I also wanted to know what does the new contrast control do exactly?

[nevcairiel]

If you can only have anti-ringing on either chroma or luma (for performance reasons), which is more useful?

Use a sharper Luma scaler with AR, and use a softer chroma scaler which doesn't necessarily need AR. IMHO anyway.

[madshi]

While DXVA2 is selected as the decoder in LAV Video and a file is playing, if MPC-HC's window is dragged to another screen then the video size resizes to what MPC-HC originally started with. This doesn't happen when software decoding is selected though.
This falls under minor cosmetical issues and as such doesn't even make my to do list for now... :D Feel free to report it again when madVR has reached v1.0.

Just noticed that It's pretty difficult to see progress in the FSE seekbar when you're a bit back from the monitor unless the screen is pretty dark. Is it possible to get the progress color darkened or something? Obviously isn't anything major, and I seem to remember you said that you wouldn't rethink the presentation until v 1.0, but just thought I'd mention it.
I don't really feel like redesigning the seekbar now.

Since that madVR is very mature already I think it might be usefull that a WIKI / manual will be created for madVR. I do not have the knowledge for this unfortunately.
Not sure about mature, at least it's not feature complete yet. I plan to create a documentation/help for v1.0, but not any sooner than that.

I have small idea - perhaps not very smart - is there any chance to add display support for YCbCr data without conversion to RGB on RGB output - this can provide capability to change RGB analog output to YPbPr output - very simple adapter (perhaps some buffer only, perhaps only some caps + sync combiner) should be enough to connect any graphic card to YPbPr input of TV (one one typical video modes need to be set anyway) - in normal conditions people need to buy/build ( VGA RGB to YPbPr converter (http://forums.benheck.com/viewtopic.php?f=20&t=36782) ) special converter from RGB to YPbPr. With such settings - i.e. RGB used as YPbPr output only sync combiner will be required.
Create an empty file with the name "YCbCr" in the madVR folder... ;)

Redone the test with 0.85.2:
Now (1), (3) and (4) are the same, (2) is different (in the same way with oversaturated green like in 0.85.1). So using native decoding without dxva scaling is still broken (at least for my setup).
Interesting, I'll try to reproduce that here. Could you please double check with the next build another time? Thanks.

With 0.85.2 GPU load with native DXVA decoding is about 0 - 30% HIGHER than copyback decoding (depending on scaling algorithm!)
GPU load is expected to be a bit higher with AMD GPUs because the GPU is used to convert the DXVA2 output to something madVR can handle. When using copyback, the CPU is used to do the conversion, so that saves a bit GPU performance, but CPU consumption should be higher. Furthermore, the copyback can be a bottleneck for fast framerates.

Madshi
this is my cpu (http://www.cpu-world.com/CPUs/K10/AMD-Athlon%20II%20X2%20245%20-%20ADX245OCK23GQ%20%28ADX245OCGQBOX%29.html)
and now that i have complained about it, it appears to be acting like when you take a car to the mechanic, it just won't make that same sound :D or in this case, a slide-show. i will report back later if it starts doing it again.
FWIW, if your CPU doesn't support SSE4.1 then madVR is not doing copyback internally, so the behaviour is different to what CUDA/DXVA-copyback in LAV Video Decoder do. In theory native DXVA should perform better in your situation, but with a small hit on chroma image quality.

Does that imply that Jinc might be less demanding on nvidia boards someday in the future?
Maybe, or maybe not, I don't know.

When Madvr crashed earlier today , I noticed this:

However , the correct value should be this :

Please consider revisiting the code that determines said value.
The crash analyzer asks the OS default, not the current user default. Anyway, that's not really a very important information...

What if you watch a movie in native resolution? Then it will use the selected chroma scaling method (by default bilinear), right?.
Correct.

If true, that's why I think an option to select DXVA chroma scaling for such situations would be a good idea since it looks better than bilinear and doesn't require (m)any ressources.
It probably doesn't require many resources on Intel GPUs. But the situation is different with AMD and NVidia, I think.

Why? Isn't chroma upscaled all the way up to full resolution (and converted to RGB) by DXVA? If the uses chooses DXVA, why change it when the luma scaling factor happens to be 1?
Imagine a user has DXVA activated for luma upscaling, and Catmull-Rom for luma downscaling. When the scaling factor is 0.9, madVR will use Catmull-Rom for luma, and the selected chroma algorithm for chroma. When the scaling factor is 1.1, madVR will use DXVA scaling, which covers both chroma and luma. When the scaling factor is exactly 1.0, madVR does not scale luma at all, so neither the "luma upscaling" nor the "luma downscaling" settings have any meaning.

Madshi, I've just seen a new bug introduced in 0.85.2

If you are downscaling with DXVA2 and switch to anything else, it results in a green image. I have to switch to a non-DXVA2 scaling algorithm a second time to restore the correct colours.
Should be fixed in the next build.

Madshi, how do I submit a freeze report? I'm getting freezes using the NVS 5200M in my laptop with MadVR (no issues using the Intel HD 4000 with MadVR, or either card with EVR).
Does just the image freeze? Or the media player user interface, too? In the first case a madVR debug log may help. In the latter case press Ctrl+Alt+Shift+Break to get a freeze report saved to your desktop.

DragonQ, try to untick "use a separate device for presentation" in madVR settings. This happens (image freezes after some time, not madVR itself) on all Optimus systems, when using dGPU.
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?

bro, r u goin to release new update for madFlac?
What for? It works perfectly (for me at least), so why releasing a new update? The only thing missing would be x64 support, but since madVR is x86 only, too, that doesn't really matter so much.

should i tick delay playback start after seeking, too?
That's your decision. It will make seeks slower, but might avoid temporary stuttering after seeks. Just try it out and choose which way you like it better.

@madshi

http://forum.doom9.org/showthread.php?p=1604267#post1604267

Thank You in Advance madshi
Please do not bump your posts. This is highly unwanted in this thread.

I'd like to ask for a feature:
Make MadVR power aware - have a combo box for selecting scaler used when laptop is on batteries.
Values can be either :
* DXVA scaler
* Bilinear (D3D)
* Same as connected to AC power.

Or just duplicate the existing propery pages for selecting the scaler(s) in power savings mode.

As for defaults, I'd rather have DXVA used for power savings mode, but maybe other people have different ideas.
Yeah. Some users have also asked for separate scaling settings for SD and HD content, or for different video framerates. I do have a solution for all this in mind, but it will require some deeper changes in the settings dialog, so it's not very high on my priority list right now. There are some more pressing things I need to work on first.

Regarding my sound break-up issue with 85.2.

I tried running it on a different machine (i5, W7, MPC-HC+LAV+madVR+Reclock) and this time there was sound but no picture. (It always amazed me that this setup would run happily run on a machine without a discrete GPU, but it did for most material without crashing). I then tried downloading the 85.2 zip file from your 'previous versions' page rather than the header on page 1 of this thread. It worked perfectly so not sure if there is a problem with one of your zip files or links?
That is quite weird. Maybe the zip you download was damaged somehow? But there are checksums in a zip which should make the unzip tool complain if there's any damage. So I don't really understand what happened to you.

Since my main PC died (lightning ...) I am trying to use my humble Atom netbook for media purposes, it does have HDMI. madVR is unusable in it I'm afraid. DXVA scaling results in out of order frames for a couple of seconds and eventually a BSOD.
Ouch. Sounds like a GPU driver bug to me. Maybe with 6233638's suggestions you can make Bilinear work with madVR?

If you can only have anti-ringing on either chroma or luma (for performance reasons), which is more useful?
It might depend on the source material, but I'd say generally luma scaling is more important than chroma scaling.

[madshi]

Hi Madshi, will it be possible to show the current video levels when we change the source white and black levels, as pointed out by Leeperry earlier?
What do you mean exactly? I'm not sure...

I also wanted to know what does the new contrast control do exactly?
It's supposed to stretch the grayscale away from the middle or nearer to the middle, while keeping peak black and peak white identical. You can read more here:

http://tutes.tonebytone.com/ImageMagick/Enhance/SigmoidalContrast/index.php

However, madVR always sets the midpoint at 0.5.

[6233638]

If you can only have anti-ringing on either chroma or luma (for performance reasons), which is more useful?It really depends on the type of content you're watching, and what scaling algorithms you're using.

Some of the Luma algorithms look absolutely fine without the anti-ringing filter. There's less ringing with Mitchell-Netravali or SoftCubic, even compared to Jinc, Lanczos or Spline with the anti-ringing filter enabled.

They aren't nearly as sharp though.


With my preference for sharper chroma, there are definitely cases where anti-ringing would be more noticeable with that. (it's usually a smaller performance hit as well)

If you watch a lot of animated content with large flat areas of colour, I'd probably go with chroma, and a smooth luma resampler.

If not, or if you're using one of the sharper luma resamplers, I'd say it's probably more beneficial with luma.


For example: Bicubic 75 Chroma, Lanczos 3 AR Luma (http://www.abload.de/img/b75l3ary9rpx.png)
Bicubic 75 AR Chroma, Lanczos 3 Luma (http://www.abload.de/img/b75arl3lcqck.png)
SoftCubic 100 Chroma, Lanczos 3 AR Luma (http://www.abload.de/img/s100l3ar1eqq4.png)
Bicubic 75 AR Chroma, Mitchell-Netravali Luma (http://www.abload.de/img/b75armnq8oqn.png)1 shows the most ringing, because the ringing here is primarily chroma. (note: this pattern actually has some ringing encoded in it - Bicubic 75 is not usually that bad for ringing)
3 shows why SoftCubic is not a great idea for Chroma. It's too soft, and causes the colour to bleed out the edges, resulting in soft edges and dark bands between the colours.
4 shows the least ringing, but also has the lowest rendering times of all four.

Lanczos 3 (http://www.abload.de/img/l3qhpf6.jpg)
Lanczos 3 AR (http://www.abload.de/img/l3arxspss.jpg)
Mitchell-Netravali (http://www.abload.de/img/mnaipp5.jpg)
SoftCubic 80 (http://www.abload.de/img/s804oqii.jpg)Here, the results are focused on Luma scaling. You can see that if you are using Lanczos 3 (or Jinc/Spline) the anti-ringing filter makes a huge difference.
But you will also see that if you switch to Mitchell-Netravali without the anti-ringing filter, there's still less ringing than Lanczos 3 AR, and it runs a lot faster.
And if you were to use SoftCubic 80, there's almost no ringing at all, and it helps reduce/mask the aliasing and compression.

But as you start using scaling algorithms that have less ringing, you see that the image isn't nearly as sharp. I actually find this to be beneficial with most commercial DVDs, which tend to already have their high frequency detail filtered out (so you're not actually losing any detail) and have ringing/compression artefacts encoded into the video.
If you're watching 720p content, it might be too soft though.

For what it's worth, I only recommend SoftCubic 70/80 for Luma. If you don't want it that soft, Mitchell-Netravali is usually a better choice than SoftCubic 50/60.


Depending on your preference for sharpness, and the type of content you're watching though, the best compromise might be to use Mitchell-Netravali for Chroma, which has minimal ringing without needing the AR filter, and is reasonably sharp (there is some loss of brightness/saturation, but it's not too bad) which could let you use a sharper luma resampler with the anti-ringing filter enabled.

As good as the anti-ringing filter is, I'm still not sure that I like using Lanczos 3 AR for luma scaling though. It's just too sharp for the content I watch, and there is still some residual ringing left in the image. Jinc 3 AR scaling has similar levels of residual ringing (perhaps a little more) but has other benefits such as considerably lower aliasing, which I feel makes that trade-off worthwhile. Jinc is very demanding on hardware though.

[omarank]

What do you mean exactly? I'm not sure...

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".

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

[DragonQ]

Use a sharper Luma scaler with AR, and use a softer chroma scaler which doesn't necessarily need AR. IMHO anyway.
Cool, that's what I had initially - Bicubic + Jinc3 AR. :)

Does just the image freeze? Or the media player user interface, too? In the first case a madVR debug log may help. In the latter case press Ctrl+Alt+Shift+Break to get a freeze report saved to your desktop.
Um...I think it's just the image but I can't remember right now.

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?
I doubt it's to do with switching GPUs - I set the nVidia Control Panel to use the nVidia GPU for MPC-HC (after renaming it to "MPC-HC player.exe") and GPU-Z clearly shows that the nVidia GPU is being used and not the Intel IGP.

[ajp_anton]

Imagine a user has DXVA activated for luma upscaling, and Catmull-Rom for luma downscaling. When the scaling factor is 0.9, madVR will use Catmull-Rom for luma, and the selected chroma algorithm for chroma. When the scaling factor is 1.1, madVR will use DXVA scaling, which covers both chroma and luma. When the scaling factor is exactly 1.0, madVR does not scale luma at all, so neither the "luma upscaling" nor the "luma downscaling" settings have any meaning.So how would you use DXVA scaling *always* for everything?
You should add a separate checkbox in the chroma upscale page to "use DXVA when possible" (meaning every time there is NOT a non-DXVA luma scaler active) (assuming you can't use DXVA for chroma only).

[madshi]

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.

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]

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?p=1563696&postcount=12641
Also I've posted log before and you said 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]

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]

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]

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 :p

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]

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]

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 (http://www.files.com/shared/50c2808190aac/sample.zip) - 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]

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

Exactly

[DragonQ]

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]

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. :D

[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]

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]

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

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.

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).

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?

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.

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.


i have some TS files like this↓
sample (http://www.files.com/shared/50c2808190aac/sample.zip) - 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

[leeperry]

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
After writing this message, I thought you would indeed be able to offer yet another magical solution to my everyday extrelemy annoying problems :scared:

Oh, well LaTo provided me with those figures so don't blame me :D

well, I've got some VHS transcodes that went through a hell lot of failtastic conversions and one of them needs 2 and the other 3...so indeed quad might be overkill.

still, I would really welcome the opportunity to tag filenames in order to enforce poor levels, aspect ratio and matrix once and for all.....but you said that you'd think about it, and I'm cool with that.

:thanks: once more for providing such exemplary dedication for such a stunning looking VR!

[madshi]

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

here is the OSD after 5 seconds of playback with deinterlacing on.
It seems that your GPU can't keep up. The "render" queue is the first queue (read from top to bottom) which is near empty, so the rendering speed is too slow. Which scaling settings are you using?

BTW, since this seems to be a movie, not a video, you could try forcing madVR into film mode (e.g. by pressing Ctrl+Alt+Shift+T twice). That would convert the movie into 576p25. DXVA Deinterlacing results in 576p50. So when using DXVA deinterlacing the GPU has to work twice as hard.

(Please note that film mode currently only supports 8bit sources.)

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)
Ideally, try 72.000Hz for 24.000Hz Blu-Rays, and 71.928Hz for 23.976Hz Blu-Rays, DVDs and broadcasts. And 50Hz for PAL, if your display supports that. Otherwise 75Hz.

edit: i just upgraded from v0.85.1 to v0.85.2,deinterlacing still looks bright and shiny
Please make sure you have all funny GPU "video" algorithms turned off in your GPU driver control panel. The only thing you should leave on is "use application settings" (where available) and "automatic deinterlacing" and "pulldown detection". Everything else should be turned off. Does that help?

well, I've got some VHS transcodes that went through a hell lot of failtastic conversions and one of them needs 2 and the other 3...so indeed quad might be overkill.
Ok, I'm trying to figure out how often such double-expanded movies occur. I guess double-expanded (30-218) occurs once in a while? Quad-expanded (52-190) probably never? And triple-expended (42-203) very rarely? If so, I might prefer only adding double-expanded to the madVR "source levels" option. I don't want to add a triple-expanded option if it's like 1 in a million video files. Thoughts?

[konakona]

after installing 0.85.2 looks like my hi10p playback improved much,no frames dropped whatsoever,ill playing with deinterlacing in a moment

edit:forcing deinterlacing to film mode works perfectly,thanks for advice!

ill try to force 75hz to get smoother playback.

[leeperry]

Ok, I'm trying to figure out how often such double-expanded movies occur. I guess double-expanded (30-218) occurs once in a while? Quad-expanded (52-190) probably never? And triple-expended (42-203) very rarely? If so, I might prefer only adding double-expanded to the madVR "source levels" option. I don't want to add a triple-expanded option if it's like 1 in a million video files. Thoughts?
How often? Much like HD movies encoded to SD with 709 coeffs, it would depend on the technical knowledge of the encoder. I have seen a bunch of 720p movies(not BD rips) that needed two conversions fairly often. I also have VHS rips that either need or three, and now that I think of it I also have some 720p movies that do need three.

Anyway, it would only be meant to fix really broken encodes so maybe you could just add a "troubleshooting" last resort hotkey that would go from 1X to 3X and leave the current general purpose "source levels" hotkey for 1X TV/PC only? Otherwise don't sweat it, I'll just use a PS script....once PotP will support them :o

[madshi]

When you say "needed two conversions", do you mean 30-218, or do you mean 42-203? I'm not sure.

[Qaq]

Please make sure you have all funny GPU "video" algorithms turned off in your GPU driver control panel.
BTW, for ATI 12.* drivers it seems useless. You can turn all the crap off but its still there. As result I no longer see the difference between Spline 3 and 4. And I can't use 11 drivers with my 7750. Maybe HDCP encription is only the solution against buggy drivers.

[HoP]

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?

info:
LAVFilters-0.53.2
MPC-HC.1.6.5.6293.x86.Lite
Win7 x64 SP1
GPU: intel i5-430M.GPU-Z info (http://imgair.net/i/1-1354962821.gif)
i know its a weak gpu.i just want to test that :D

madvr:
chroma: SoftCubic 70
upscale: DXVA2
downscale: DXVA2

PS: i can play 720p videos (like youtube mp4 files) without any problem.
i have problem with TS files.

[pie1394]

info:
LAVFilters-0.53.2
MPC-HC.1.6.5.6293.x86.Lite
Win7 x64 SP1
GPU: intel i5-430M.GPU-Z info (http://imgair.net/i/1-1354962821.gif)
i know its a weak gpu.i just want to test that :D

madvr:
chroma: SoftCubic 70
upscale: DXVA2
downscale: DXVA2

PS: i can play 720p videos (like youtube mp4 files) without any problem.
i have problem with TS files.

Maybe you can try lavfilter 0.54.1 first. There are some issues on detecting old Intel HD graphics DXVA decoding capability with older versions.

Actually I don't think you can get lavfilter's DXVA decoding work with MPEG-2 contents on Arrandale and SandyBridge based Core i processor's GPU. (i.e. if not HD2500/HD4000)

In your case, you need to use "Microsoft DTV-DVD video decoder" to handle MPEG-2 contents to get the maximum available DXVA HW acceleration on such GPU.

[nevcairiel]

Actually I don't think you can get lavfilter's DXVA decoding work with MPEG-2 contents on Arrandale and SandyBridge based Core i processor's GPU. (i.e. if not HD2500/HD4000)

MPEG-2 DXVA works just fine on my Sandy Bridge HD3000 with LAV.
Should even work on older CPUs, too.

[madshi]

Are you sure you're using the latest LAV version?
LAVFilters-0.53.2
You're not using the latest LAV version. Please update and try again.

[pie1394]

MPEG-2 DXVA works just fine on my Sandy Bridge HD3000 with LAV.
Should even work on older CPUs, too.

Really? :eek:

On my Toshiba Z830 (Core i5-2467M), DXVA checker does not show DXVA2_MPEG2_VLD capability with the latest HD graphics 9.x driver. Looks like a driver-broken issue?

Something wrong in my memory... It does show DXVA2_MPEG2_VLD... :scared:

[leeperry]

When you say "needed two conversions", do you mean 30-218, or do you mean 42-203? I'm not sure.
well, I really mean:
double expanded: 29.74 - 217.82
triple expanded: 41.54 - 203.07

There are quite a lot of WMV 720p files around that require 2X IME, and the most broken ones actually require 3X....I dunno how they manage to mess up so badly, some funky presets in the official WMV encoder or in Premiere might take the blame somehow or maybe some HD cameras record in 0-255(mandatory for xvyCC compliance) so ppl go overboard with levels conversions afterwards, but they often come up in rather poorly deinterlaced 50i@25p form too..

I believe we can forget about 4X, I kept it as a worst case scenario but never found an actual use for it.

[HoP]

Maybe you can try lavfilter 0.54.1 first. There are some issues on detecting old Intel HD graphics DXVA decoding capability with older versions.


You're not using the latest LAV version. Please update and try again.

yes,now it works perfectly :p
thanks guys :thanks:

[ser_renely]

Since that madVR is very mature already I think it might be usefull that a WIKI / manual will be created for madVR. I do not have the knowledge for this unfortunately.

Would love to see that. I have so many questions...I just don't ask them bc I am trying to catch up to speed on everything first.

[NicolasRobidoux]

Mathias:

In case you feel like trying something new that you have already, try upsampling with Mitchell-Netravali with sigmoidization set to about 6.5.

I've not taken the time to find the exact value for Mitchell-Netravali, but for the fairly similar EWA Robidoux there is a justification for sigmoidal contrast = 6.5 +- an irrelevant nudge. (As usual, this is for sRGB content, and I'm not totally sure things carry over well with other "non cube-based" colorspaces, but I'm putting this out there just in case.)

Worth the addition computation time? Not sure.

Again, this is a shot "in the dark".

[turbojet]

I have a problem with 85.x versions, when opening mpc-be\hc on extended display the first frame is displayed and then pauses. Tried resetting madvr and mpc to defaults with same results. 84.7 and EVR plays like it should with same madvr, mpc settings. Tested on 2 computers, nvidia and intel gpus with a few different videos. OSD probably doesn't help but uploaded it to http://www.imagebam.com/image/d30537224825014

Is there a way to set color matrix by file name? I tried bt.601 and bt.709 with no luck. It would be really nice if the bitstream doesn't inform it would set by these in the file name:
bt.709: *hd*/*bd*/*blu-ray*/*bluray*
bt.601: *dvd*/*vhs*
What resolution does it guess 709 instead of 601? 852px width detects as 601 but I would think 721px and up would be 709.

One last thing for madshi, I know you've talked about implementing a sharpen filter like finesharp as a pixel shader but after watching finesharp with bicubic 75/100 and lanczos3 over the past 6 months or so and spline144 over the past month. Spline144 results in the nicer and more natural picture, but not quite as sharp. If Spline144 to 256 difference is anything like 100 to 144, 256 would sharper than finesharp with a substantially lower demand on gpu and maybe most important a lot less work to implement, spline coefficients are already calculated. Finesharp at times makes faces look 'waxy' like all other sharpeners I've tried.

Also haven't seen it reported or if it was known but 85.2 no longer has the green line problem with dxva resizing with HD3000.

[madshi]

Mathias:

In case you feel like trying something new that you have already, try upsampling with Mitchell-Netravali with sigmoidization set to about 6.5.

I've not taken the time to find the exact value for Mitchell-Netravali, but for the fairly similar EWA Robidoux there is a justification for sigmoidal contrast = 6.5 +- an irrelevant nudge. (As usual, this is for sRGB content, and I'm not totally sure things carry over well with other "non cube-based" colorspaces, but I'm putting this out there just in case.)

Worth the addition computation time? Not sure.

Again, this is a shot "in the dark".
Thanks for the suggestion! However, after all the experiments I did with sigmoidization (and I spent quite a lot of time on that), my opinion at the moment is that sigmoidization helps with some images and makes things worse with other images. I don't think this depends much on which exact resampling and window filters are used, so I have my doubts that results with Mitchell-Netravali would be any different to my earlier experiments with Jinc and Lanczos. Because of that I'm a bit weary of trying further "shots in the dark" with sigmoidization at the moment, because I simply have *so* many other things to do, most of which will bring immediate improvements.

BTW, for ATI 12.* drivers it seems useless. You can turn all the crap off but its still there. As result I no longer see the difference between Spline 3 and 4. And I can't use 11 drivers with my 7750. Maybe HDCP encription is only the solution against buggy drivers.
You mean you got problems with wrong colors with deinterlacing turned on? Or what do you mean? And what does HDCP have to do with anything?

I have a problem with 85.x versions, when opening mpc-be\hc on extended display the first frame is displayed and then pauses. Tried resetting madvr and mpc to defaults with same results. 84.7 and EVR plays like it should with same madvr, mpc settings. Tested on 2 computers, nvidia and intel gpus with a few different videos. OSD probably doesn't help but uploaded it to http://www.imagebam.com/image/d30537224825014
A debug log might help. What does "extended display" mean exactly? You mean some special mode of dual monitor setup? Does the problem only occur in that one mode? In all other modes playback works just fine?

Is there a way to set color matrix by file name?
Not at the moment, but it's on my to do list.

What resolution does it guess 709 instead of 601? 852px width detects as 601 but I would think 721px and up would be 709.
Anamorphic PAL is 1024x576 after aspect ratio correction. So width must be higher than 1024, or height bigger than 576 to quality for "HD".

One last thing for madshi, I know you've talked about implementing a sharpen filter like finesharp as a pixel shader but after watching finesharp with bicubic 75/100 and lanczos3 over the past 6 months or so and spline144 over the past month. Spline144 results in the nicer and more natural picture, but not quite as sharp. If Spline144 to 256 difference is anything like 100 to 144, 256 would sharper than finesharp with a substantially lower demand on gpu and maybe most important a lot less work to implement, spline coefficients are already calculated. Finesharp at times makes faces look 'waxy' like all other sharpeners I've tried.
To my eyes in the very few test images I saw, spline144 looked oversharpened. But I have to preserve final judgement because I simply didn't have enough time to evaluate it properly. Anyway, now is not the time to talk about FineSharp or alternatives. Busy with other stuff atm...

[ryrynz]

To my eyes in the very few test images I saw, spline144 looked oversharpened.

Correct, it's pretty bad. We're all curious to see how it looks with anti-ringing applied though.

[mandarinka]

Anamorphic PAL is 1024x576 after aspect ratio correction. So width must be higher than 1024, or height bigger than 576 to quality for "HD".

This (http://ps-auxw.de/cgi-bin/ar-calc.pl?source_type=pal&source_ar=widescreen&transform_type_1=none&transform_x_1=&transform_y_1=&transform_type_2=none&transform_x_2=&transform_y_2=&transform_type_3=none&transform_x_3=&transform_y_3=&transform_type_4=none&transform_x_4=&transform_y_4=&transform_type_5=none&transform_x_5=&transform_y_5=&transform_type_6=none&transform_x_6=&transform_y_6=&transform_type_7=none&transform_x_7=&transform_y_7=&transform_type_8=none&transform_x_8=&transform_y_8=&keep=true&calculate=Calculate) says 1050 × 576 (ITU scaling).

[madshi]

This says 1050 × 576 (ITU scaling).
Well, do the math yourself:

16/9 = ?
1024/576 = ?
1050/576 = ?
576 * 16 / 9 = ?

It's not a good idea to blindly trust some website if you can do the proper math yourself... ;)