madVR - high quality video renderer (GPU assisted)

[SamuriHL]

Madshi, for now you could document in the op and the madvr read me what settings to use for various levels of hardware. Make the information widely and easily available. Then if someone tries the defaults and it's slow, they can quickly check the read me and determine what to change to make it work properly. The issue now is that information isn't obvious for a beginner and difficult to find. I just recently had to help someone tweak settings for an underpowered laptop and I wasn't even sure exactly what to tell them. After a bit of trial and error they got it. But it'd be easier if they could look in a doc and see what settings they should try. Hell you've got about 50 people in this thread will to do the write up... Just include it in the op and read me and then we just have to make it clear that info is out there.

Sent from my Xoom using Tapatalk 2

[DragonQ]

Originally Posted by pie1394

For HDTV programs in US, they are also broadcasted with MPEG-2 1280x720p60 format. It has higher vertical motion resolution (sensitive to human eyes) if compared to video stream encoded with 1920x1080i30 mode at the same bit rate.

Unfortunately 1080p24 HDTV video encoding format is only defined in ATSC system. So the FILM type DTV contents outside the North America (US, CA) need to send out via either of following 2 ways:

1. Make it telecined to i60. So the compressed MPEG-2 / H.264 video are actually 60 field images in 30 fps. The TV's post-processing engine detects if the decoded video is truly progressive and perform the IVTC process.
(Old day's many NTSC DVDs were also made in this way when the DVD player capable of 24fps progressive output was not popular...)

2. Just make the telecine flags (i.e. field first order, repeat field) in MPEG-2 / H.264 video header, and the video stream is still encoded with 24 full-frame image mode. The TV performs necessary post-processing depending on the display characteristics. Ex:

- 1080i CRT / ALiS PDP --> Telecine.
- Progressive PDP --> 48 / 72 / 96 fps panel driving
- Progressive LCD --> 120 Hz driving + black-image insertion with scanning LED backlight

The 2nd way is now widely used. The progressive full-frame often needs less bit rate to achieve the same video quality with the interlaced field-frame's. The DVD player / TV settop box is also cheaper to be made in order to achieive better FILM contents' video display quality.

I can't comment on 60 Hz regions but for 50 Hz regions, any progressive content is typically just broadcast as 1080i (films sped up from 24p to 25p beforehand). As long as the deinterlacer is half-decent, it'll notice a lack of movement between the fields and play it back using weave. No weird IVTC needed.

1080i/25 is also far more prevalent than 720p/50 in Europe.

[AndreaMG]

Originally Posted by madshi

I'd really like to get people to understand that madVR is not only about satisfying the highest quality demand people, anymore, but that it can be used very well with low-performance GPUs now, too (with lowered settings). Basically I want madVR to become the new default renderer for all but stone-age GPUs. But in order to make madVR attractive for owners of rather old/slow GPUs, I'd like to change the default settings to run smoothly even with old/slow GPUs. That doesn't have to stop anybody from changing the settings to get back all the lost quality.
Anyway, maybe I should delay this to v1.0 and then implement some kind of performance vs. quality slider or something like that. It just hurts me a little if I think of users trying madVR now and then dropping it again because the first impression is that it's too slow for their PC.

You surely have a point, but I think people who use HTCPC and decide to go beyond power dvd, beyond VLC, beyond basic players, and begin to choose filters, etc. have knowledge of what they are doing and a person who tries MadVR and plays a file and sees that it stutters most probably will interrogate himself and he will play with settings, I don't think that he will simply go away from MadVR... There is also a risk that some people with strong GPUS will find the MadVR with the "easy default settings" perfectly play everything without even realizing that they are reproducing interlaced videos at half rate

[aufkrawall]

I think madVR should not sacrifice any quality over performance by default. That's what makes it different from other renderers and what it's known for, offering best quality for enthusiasts.
Maybe later there should be some kind of wizards or autodetection for slower hardware. Doubled framerate is the only real advantage of interlacing left, so why drop it?
And coming spring/summer there will be faster GPUs, leading to even more consumer hardware being able to handle madVR with very most content with the current default settings.

[druneau]

Originally Posted by aufkrawall

I think madVR should not sacrifice any quality over performance by default. That's what makes it different from other renderers and what it's known for, offering best quality for enthusiasts.
Maybe later there should be some kind of wizards or autodetection for slower hardware. Doubled framerate is the only real advantage of interlacing left, so why drop it?
And coming spring/summer there will be faster GPUs, leading to even more consumer hardware being able to handle madVR with very most content with the current default settings.

Could be a forced educational tutorial explaining the algorithms lol.

With a quiz at the end. You fail, MadVR self destructs.

[HoP]

@madshi
play with madVR v0.85.4



play with EVR CP



why??

file info:

Format : MPEG-TS
Format profile : No PAT/PMT
File size : 107 MiB
Duration : 3mn 40s
Overall bit rate mode : Variable
Overall bit rate : 4 060 Kbps

Video
ID : 67 (0x43)
Format : MPEG Video
Format version : Version 2
Format profile : Main@Main
Format settings, BVOP : Yes
Format settings, Matrix : Custom
Format settings, GOP : M=3, N=12
Duration : 3mn 40s
Bit rate mode : Variable
Bit rate : 3 601 Kbps
Maximum bit rate : 10 000 Kbps
Width : 720 pixels
Height : 576 pixels
Display aspect ratio : 4:3
Frame rate : 25.000 fps
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Interlaced
Scan order : Top Field First
Compression mode : Lossy
Bits/(Pixel*Frame) : 0.347
Stream size : 94.4 MiB (89%)

Audio
ID : 46 (0x2E)
Format : MPEG Audio
Format version : Version 1
Format profile : Layer 2
Duration : 3mn 40s
Bit rate mode : Constant
Bit rate : 256 Kbps
Channel(s) : 2 channels
Sampling rate : 48.0 KHz
Compression mode : Lossy
Delay relative to video : -498ms
Stream size : 6.72 MiB (6%)

[pie1394]

Originally Posted by aufkrawall

Maybe later there should be some kind of wizards or autodetection for slower hardware. Doubled framerate is the only real advantage of interlacing left, so why drop it?

I agree with you about this.

If madshi wants madVR to be widely used in public, however, it cannot be just designed for enthusiast's hardware or newly coming hardware. The most commonly used hardware's life cycle need to be considered, too. It is something like chicken-egg quiz --- who is the first...

The simpler configuration method by default is more important than the flexibility for most people who do not have the knowledge to figure out what is the best setup for his equipments. It is engineer and enthusaist's job to sort out these combinations.

Tell you guys a story about my mother who is 66 years old. She has the PC using experience near 10 years (Win98 / WinXP / Win7). But I found she still cannot remember any function which needs more than 2 mouse-click steps or by hot-key. Last year when I replaced her broken old PC with a new one + Win7 OS, she complained to me that she didn't know how to use it... The evil Microsoft had changed the operation flow and UI look. It also happens to office 2007/2010, which drives me crazy, too.

Instead of teaching her everything from scratch, I just setup the UI arragement to layouts which she is familiar with...

Fortunately her toy's (was iPad2, now iPad3) most function usage is very straight forward and consistent. So she seldom calls me for help now ...

[leeperry]

Originally Posted by madshi

Sure.

Having more thought about it, it's be great if you could make it "PC > TV > 2X > 3X > 2X > TV >" so we could quickly roll between 2X and 3X because that's usually where the big dilemma lies IME....I've got that washed out 720p video where I can see that the ffdshow histogram starts around 30 but I kinda like the oversaturated colors of 3X anyway....OTOH, it's on a SMPTE-C CRT with a REC709 video and I don't have gamut rolling enabled (yet)

Anyway, it'd be great if we could roll them from bright to dark then gradually brighter instead of going bright > dark > bright. I also really enjoy how the hotkey starts from the current position instead of starting all over again from "PC"

I haven't gotten through the gamut PS script PotP automatic profile stuff yet, but I have to say that I'm as thoroughly impressed as ever by mVR's PQ when using Jinc3 AR for chroma/luma

And I believe it's only a matter of knowing about the ITU issue because I now can fairly easily horizontally stretch the video in order to look natural....you really learn new stuff everyday, huh!

Also, I really enjoy how it will finally be possible to force EBU for 25p HD and SMPTE-C for 23.976/24p HD via PotP automatic profiles because as much as I realize that this is out of specs, I still find the colors more true to life....and so do Kazuya, the french CEO of the ISF, Joe Kane and many other ppl....what matters in the end is that we can all set up the picture exactly the way we want it

I still think that you may wanna setup a dumbed down config panel for newbies and a more advanced panel for power users so the former wouldn't be scared by the zillion settings and the latter would find all the options they could ever crave. I can only imagine how a complete newbie could be scared by the current config panel, but this isn't quite a good reason to dumb it down to death IMHO.

[SyrupBuccaneer]

Originally Posted by madshi

Anyway, according to the log, PotPlayer does not use madVR's screenshot functionality at all. I suppose that PotPlayer grabs the video frames directly from the decoder. You'll have to ask the PotPlayer devs to use madVR's screenshot functionality instead. It's available through the official Microsoft interface (IBasicVideo::GetCurrentImage).

Can you suggest a way of contacting them? I've been looking around for a way to them but all I've found is the Korean Daum media center page. Do they have an English team at all?

[iamhacked]

Originally Posted by madshi

Try Bilinear for everything. Your GPU is really slow, so you'll have to compromise on quality. Also try activating "use half framerate DXVA processing", if your content is interlaced and you can't get it smooth any other way.

P.S: What exactly do you mean with "static glitches"?

I turned on all the settings for compromise quality. The glitches look like static noise flashing, I think you guys call it presentation glitches. It might be because of the flush settings I changed for performance (from someone in this thread). But I don't always get glitches. I'm still getting it even with bilinear for everything. I'm using it with LAV filters and KMPlayer

[6233638]

Originally Posted by aufkrawall

I think madVR should not sacrifice any quality over performance by default. That's what makes it different from other renderers and what it's known for, offering best quality for enthusiasts.
Maybe later there should be some kind of wizards or autodetection for slower hardware. Doubled framerate is the only real advantage of interlacing left, so why drop it?
And coming spring/summer there will be faster GPUs, leading to even more consumer hardware being able to handle madVR with very most content with the current default settings.

I'm not sure that we need to make such a big fuss over it right now to be honest. I think it is best to have madVR set up in such a way that most systems can handle it by default, while still offering a visual improvement over other video renderers. It's not like the quality options are going to disappear, or that your current settings will be changed.

By the time it hits 1.0, I expect Madshi will have implemented a look-up table to pick good defaults based on the card detected in your system, or possibly even a quick benchmark on first launch to see what it's capable of.

We will hopefully also see the ability to use different scaling algorithms for different types of content (e.g. different algorithms for SD upscaling, HD upscaling/downscaling, and 4K downscaling) which could also alleviate some of the problems by using less demanding algorithms on HD content than SD content for example.

Originally Posted by iamhacked

I turned on all the settings for compromise quality. The glitches look like static noise flashing, I think you guys call it presentation glitches. It might be because of the flush settings I changed for performance (from someone in this thread). But I don't always get glitches. I'm still getting it even with bilinear for everything. I'm using it with LAV filters and KMPlayer

Presentation glitches look like stutters in the video similar to dropped/delayed frames, not static noise flashing. I actually wonder if it could be problems with your video card. (is there anything else you run that stresses the card?)

What happens if you use LAV filters to decode this video using another renderer?

[markanini]

Originally Posted by 6233638

By the time it hits 1.0, I expect Madshi will have implemented a look-up table to pick good defaults based on the card detected in your system, or possibly even a quick benchmark on first launch to see what it's capable of.

We will hopefully also see the ability to use different scaling algorithms for different types of content (e.g. different algorithms for SD upscaling, HD upscaling/downscaling, and 4K downscaling) which could also alleviate some of the problems by using less demanding algorithms on HD content than SD content for example.

This would be great.

[iamhacked]

Originally Posted by 6233638

Presentation glitches look like stutters in the video similar to dropped/delayed frames, not static noise flashing. I actually wonder if it could be problems with your video card. (is there anything else you run that stresses the card?)

What happens if you use LAV filters to decode this video using another renderer?

It's fine with other renderers, I don't run anything else graphic-related

[Niyawa]

Hey people, this could have already been answered but I did a search and didn't find anything, so I have to ask. Why madVR overrides MPC-HC OSD interface, and is there a way to turn it off so I can customize it?

[iSunrise]

Originally Posted by 6233638

We will hopefully also see the ability to use different scaling algorithms for different types of content (e.g. different algorithms for SD upscaling, HD upscaling/downscaling, and 4K downscaling) which could also alleviate some of the problems by using less demanding algorithms on HD content than SD content for example.

Yes, I am also waiting for that to happen. SD content is not going to disappear over night anytime soon and for some sources, never. Together with customizable SD/HD resolution threshold settings (if enabled, otherwise use the auto-detect logic), these would indeed be great additions to madVR.

[vivan]

Originally Posted by Niyawa

Why madVR overrides MPC-HC OSD interface, and is there a way to turn it off so I can customize it?

In fullscreen mode? It's because exclusive mode is active (which you can rutn off in madVR settings)...

[aufkrawall]

Originally Posted by madshi

Oooh, that's interesting... Without full/half floating point processing, basically EVR converts to 8bit RGB, I believe. madVR refuses to do that because it's too low quality. So it seems EVR isn't really faster than madVR in this case, it just uses lower (too low) quality by default than madVR.

You could try the test build from yesterday (see a few posts up), which now uses float16 instead of float32. Maybe that will be enough to make madVR run fluidly on your E-350? What happens if you switch EVR to half floating point processing? Is that running smoothly?

Yep, that madVR builds works and it works also with EVR half floating point processing (maybe a little worse).
With madVR I don't get any dropped frames after lowering queues in windowed mode.
But unfortunately, in FSE I get dropped frames, present queue sometimes gets emtpy.

Btw: I figured out why FSE wasn't working at all: It seems it didn't like 16 frames presented in advance, with lower values like 4 or 8 this doesn't happen. Then there are just those frame drops.

[Niyawa]

Originally Posted by vivan

In fullscreen mode? It's because exclusive mode is active (which you can rutn off in madVR settings)...

Nope, I don't use fullscreen mode. It's simple the OSD style that madVR overwrites from MPC-HC, I don't know why it does that and neither how to turn it off (if it's possible).

[Kado]

Originally Posted by Niyawa

Nope, I don't use fullscreen mode. It's simple the OSD style that madVR overwrites from MPC-HC, I don't know why it does that and neither how to turn it off (if it's possible).

I only get the custom osd if in fullscreen exclusive mode, you can disable it in mad vr settings, rendering, general settings, uncheck "enable automatic fullscreen exclusive mode".

Feliz Natal! (Merry Christmas!)

[Edit]Unless you're talking about this (custom buttons), just erase "toolbar.bmp" in mpc folder.

[/Edit]

[Edit2]OK, just saw your picture because before it wasn't approved yet. I think ost implementation is renderer dependent, haali renderer does not have osd for instance, and that was the style used by madashi.[/Edit2]

[madshi]

Originally Posted by HoP

play with madVR v0.85.4

play with EVR CP

why??

file info:

EVR exposes different interfaces to the media player compared to VMR9. Try VMR9, I think it will behave similar to madVR. madVR currently exposes the same interfaces that VMR9 exposes. I plan to add support for EVR interfaces in a future version, but probably not too soon.

Originally Posted by leeperry

Having more thought about it, it's be great if you could make it "PC > TV > 2X > 3X > 2X > TV >" so we could quickly roll between 2X and 3X because that's usually where the big dilemma lies IME....

I'm not sure if I like that. All other toggles work differently, and I somehow have to allow switching back to auto detected levels, too.

Originally Posted by SyrupBuccaneer

Can you suggest a way of contacting them? I've been looking around for a way to them but all I've found is the Korean Daum media center page. Do they have an English team at all?

I think there's a forum, but I don't know where and whether there's support for English users. @leeperry? (Question is about how to contact PotPlayer developers.)

Originally Posted by iamhacked

I turned on all the settings for compromise quality. The glitches look like static noise flashing, I think you guys call it presentation glitches. It might be because of the flush settings I changed for performance (from someone in this thread). But I don't always get glitches. I'm still getting it even with bilinear for everything. I'm using it with LAV filters and KMPlayer

So you mean you're getting some video frames which look corrupted or overly noisy? This doesn't sound like a usual problem, no other user has reported a problem like this. This may be a problem with your GPU drivers, or maybe the GPU is overheating, or something like that. I can't really say. Do you get the same problem in windowed and fullscreen exclusive mode? You could try going back to madVR default settings, just to double check.

Originally Posted by Niyawa

Hey people, this could have already been answered but I did a search and didn't find anything, so I have to ask. Why madVR overrides MPC-HC OSD interface, and is there a way to turn it off so I can customize it?

madVR does not override anything. madVR offers various different OSD interfaces which the media player can use. MPC-HC could show the same OSD with madVR which is shows for VMR/EVR. The MPC-HC devs would have to add support for that, too, by using one of the madVR OSD interfaces. Currently MPC-HC uses a very simple madVR interface to show text messages, which are then shown by madVR's own style/design. Actually the code MPC-HC uses for that was added to MPC-HC by me. Making MPC-HC show the same OSD it shows for EVR/VMR9 would be possible, but would cost a bit of extra work. Furthermore it's a matter of taste, too. Personally, I like the madVR style much more than the MPC-HC style. If you want to get the MPC-HC style, you'd have to ask the MPC-HC devs to add support for that for madVR. But I'm not sure if they want to spend the development time for that, considering that the current solution works well, and the only difference would be a different "look". Also I think some users prefer the current solution.

Originally Posted by aufkrawall

Yep, that madVR builds works and it works also with EVR half floating point processing (maybe a little worse).
With madVR I don't get any dropped frames after lowering queues in windowed mode.
But unfortunately, in FSE I get dropped frames, present queue sometimes gets emtpy.

Btw: I figured out why FSE wasn't working at all: It seems it didn't like 16 frames presented in advance, with lower values like 4 or 8 this doesn't happen. Then there are just those frame drops.

Ah, that's interesting. I'm glad to hear that madVR seems to run just as fast as EVR with your ultra-slow GPU, when switching EVR to half-way decent quality settings. That does confirm that madVR can work with very low-power GPU hardware, too.