madVR - high quality video renderer (GPU assisted)

[scollaco]

I'm having issues getting 3D to work. I'm a noob so maybe I missed something.

1) I have windows 10 and the latest 6th generation NUC with Intel Iris Pro 540 Graphics. (NUC6i5syk).
2) I installed madVR v.0.90.3 and the latest LAV filters fromt he nightly builds (0.67.0-65).
This is what I did:
1) I set the display control panel to have 3D "on".
2) In madVR I turned off "exclusive mode". I enabled "use Direct3D11 for presentation (is this correct?) and also check marked "present a frame for every VSync" (Is this a needed step too?)
3) I didn't touch any settings in madVr's stereo 3D area besides the default that si checked "enable stereo 3d playback"

When playing a 3d mkv made using makemkv, I see that LAV filters is using the "msdk mvc" decoder.

on playing a movie, the projector correctly switches resolution to 1920x2205 for framepacked 3D. However the image that shows up is frozen and the movie doesn't play any further. At that point I cannot exit out of my MPC-BE player and need to kill it using the task manager. I'm stuck because it looks like it all correctly switches to 3d and I have a proper 3d image on screen, but it is frozen and refuses to play the movie.

I also tried setting my scaling algorithms to the most basic but still same issue. Any advice what I'm doing wrong or what step I'm missing?

Are you using x86 or x64 version of MPC-BE? If you're using x64, try x86 instead, just to double check.

Thanks for the reply madshi.

I'm think I'm already using the x86 version of MPC-BE. As well madVR & LAV as well. LAV and mpc-be install in C:\Program Files (x86)\mpc-be.exe

Any other advice?

[madshi]

Thanks for the reply madshi.

I'm think I'm already using the x86 version of MPC-BE. As well madVR & LAV as well. LAV and mpc-be install in C:\Program Files (x86)\mpc-be.exe

Any other advice?
Ok, in that case please create a debug log, zip it and upload it somewhere. Don't attach it to this forum, thanks.

[huhn]

That would be great :thanks:

i switched the nvidia GPU with my AMD R9 270 and in general there are no problems with madVR.

windows doesn't wanted to change to FHD resolution and was sending a letterbox but i got that fixed.

1080p60
Radeon Software Version 15.11 (crimson)
windows 10 10547
madVR 0.90.3
cx700 hdmi

if you have an iGPU try it to make sure the issue isn't windows.

[BetA13]

THX madshi for the test builds.. i hope will come around soon to test em and report back..
as soon as i have time i will test them....

greetz

edit:

testing it now. With teh 32bit player like you wanted, didnt try 64bit yet

i used the zzz - builds2.rar and started with number 5.

5. (5 - videotext and vsync) = works
4. ..............(OSD).............= FAIL
3. OpenCL.........................= FAIL

Dindt bother testing 2 and 1, since they would also fail??
Or should i test them also?

Greetings :)

[MSL_DK]

I get an error in the upper left corner that says something like: Could not read display mode. It only happens if I disable FSE. AMD control panel crashes and windows behave very strangely and I will have to restart.

Windows 10 x64
MPC-HC x86 latest
madVR x86 latest
AMD R9 270 with latest driver
Display 1920x1080 60Hz

i switched the nvidia GPU with my AMD R9 270 and in general there are no problems with madVR.

windows doesn't wanted to change to FHD resolution and was sending a letterbox but i got that fixed.

1080p60
Radeon Software Version 15.11 (crimson)
windows 10 10547
madVR 0.90.3
cx700 hdmi

if you have an iGPU try it to make sure the issue isn't windows.

I have found the problem. The problem occurs at Rendering> windowed mode> how many video frames ....> 16

If I set it to 14, the problem does not occur.

[MSL_DK]

i switched the nvidia GPU with my AMD R9 270 and in general there are no problems with madVR.

windows doesn't wanted to change to FHD resolution and was sending a letterbox but i got that fixed.

1080p60
Radeon Software Version 15.11 (crimson)
windows 10 10547
madVR 0.90.3
cx700 hdmi

if you have an iGPU try it to make sure the issue isn't windows.

Thank you for looking at it.

While we're at it, can I get you to test properties> the native display bitdepth is> 10 bit (or higher) and Rendering> General settings> use Direct3D 11 for presentation> present a frame for every VSync

and tell me the result of your monitor behavior?

My screen seems to switch modes and render and present queue queue is not filled out. (1-3 / 14 and 1-3 / 14)

:thanks:

[omarank]

This is a GPU driver problem with AMD cards when using 24Hz and 60Hz refresh rates. With 23p, 50p and 59p the queues get filled properly. Some user reported here that with the newer drivers, the problem is fixed with either of 24p and 60p (don’t remember which one, perhaps 24p).

[Backflash]

Are you using x86 or x64 version of your media player and madVR? If you're using x64, try using x86 instead to see if you have the same display driver crash problem with that. If you have the same crashes in x86, then please look at this post:

http://forum.doom9.org/showthread.php?p=1754526#post1754526

I tested both MPC x64 and x86 nightly 1.7.10.86

Builds 5-2 - works fine i couldn't replicate the crash, relaunched different videos 7-10 times on both players.
Build 1 caught a crash on 4th video playback start on x64 player, driver recovered this time, works fine on x86 after 10 relaunches.
Oh and yes, it happens only on video start - timer doesn't tick and in 2-3 second screen goes black when driver can't recover.

Thing is, this crash is seemingly random, sometimes it doesn't happen at all during the day sometimes it happens once or twice. Very hard to tell when it actually will happen.

[madshi]

I have found the problem. The problem occurs at Rendering> windowed mode> how many video frames ....> 16

If I set it to 14, the problem does not occur.
Ah yes, some users had problem with 16, I'm not sure why. Could be running out of RAM, or a driver problem, or something else.

testing it now. With teh 32bit player like you wanted, didnt try 64bit yet

i used the zzz - builds2.rar and started with number 5.

5. (5 - videotext and vsync) = works
4. ..............(OSD).............= FAIL
3. OpenCL.........................= FAIL

Dindt bother testing 2 and 1, since they would also fail??
Or should i test them also?
Thanks. No need to test builds 1 and 2. How reliable is the "fail/works" for you? Are you 100% sure 5 is fine and 4 has a problem? Or is there a bit of randomness involved? Just asking, to be safe.

I tested both MPC x64 and x86 nightly 1.7.10.86

Builds 5-2 - works fine i couldn't replicate the crash, relaunched different videos 7-10 times on both players.
Build 1 caught a crash on 4th video playback start on x64 player, driver recovered this time, works fine on x86 after 10 relaunches.
Oh and yes, it happens only on video start - timer doesn't tick and in 2-3 second screen goes black when driver can't recover.

Thing is, this crash is seemingly random, sometimes it doesn't happen at all during the day sometimes it happens once or twice. Very hard to tell when it actually will happen.
Thanks. Reading your description it seems to be quite random, which makes me fear that maybe builds 2-5 might also crash if you try often enough. Just to be safe, can you try with just build 2, maybe 20-30 times, do be extra safe that build 2 is working fine? I'm asking because if builds 1 crashes, it's possible but "weird" to have build 2 work fine because the difference between builds 1 and 2 is very small. I would have expected build 1 and 2 to give identical results. So considering that it's somewhat random, I'd appreciate if you could double check, just to be extra safe. Thanks.

[Backflash]

Thanks. Reading your description it seems to be quite random, which makes me fear that maybe builds 2-5 might also crash if you try often enough

Ran build 2 on both x64/x86, 70 times on different videos, from sd to 720p 10bit - couldn't crash it.

[madshi]

Ran build 2 on both x64/x86, 70 times on different videos, from sd to 720p 10bit - couldn't crash it.
That's quite interesting. Practically this means that probably in x64 the GPU driver crashes when madVR tries to create multiple DirectCompute shaders at the same time in different threads. This looks like a GPU driver bug because in theory D3D11 should be thread safe.

So I'll have to redesign the DirectCompute stuff to be done in one thread, only. Very sad, but not a big problem.

Edit: Which GPU and OS are you using?

[Backflash]

Edit: Which GPU and OS are you using?

R9 390(i7 2600/8Gb) Windows 10 Pro 1511(10586.63) x64, Crimson 16.1

[BetA13]

Thanks. No need to test builds 1 and 2. How reliable is the "fail/works" for you? Are you 100% sure 5 is fine and 4 has a problem? Or is there a bit of randomness involved? Just asking, to be safe.




hy there,

yes, "5" is working everytime i think. i didnt had any odd behavior with it. testet most settings with it and watched about 3 movies..
resizing works, fullscreen switching works, no drops, no stutter when i move teh mouse, video is in sync and normal speed, shaders work, just like with 0.89.19 :)

Sidenote:
only thing i noticed wich is different (i think, i didnt see that change before) is that in windowed mode, videos are played with 48fps and in Fullscreen (exclusive) mode its 60fps.
Wasnt it 60fps in windowed and fullscreen before? Of course only if you have 60hz ??
Or was this changed? Anyway, im totaly fine with that because it eats less power from my gpu in 48fps (windowed) in fullscreen i prefer teh 60fps like it is...

"4" and downwards allways produce the error i have been struggling with in the start.

so, im 99% shure that 5 is the winner for me...

greetings
:thanks:

[daglax]

I dont get it. If i start any .mkv-movie (SBS-format) the madvr on screen display never shows 4:2:0 (3D), it's always 4:2:0.

I installed the latest LAV Filters and activated the filters in MPC's external filter section, but still no sign of 3D. Why is LAV not decoding in 3D?

[sneaker_ger]

At the moment the new 3D features of LAV and madVR are for MVC-in-MKV only. There is no special handling of SBS files.

[daglax]

Lol that makes sense. Didn't find any information about that the last couple of pages or maybe i didn't understand.
Thank you for clarification.

Is SBS- Playback planned eventually?

[scollaco]

Does someone have an example code of how they are using profiles for including 3d now? I'm not very code savvy and would appreciate the help.

Also what settings are people using for 3d as my scaling algorithms etc...seem to be too much to play 3D

[huhn]

Is SBS- Playback planned eventually?

madVR can already do that.

Thank you for looking at it.

While we're at it, can I get you to test properties> the native display bitdepth is> 10 bit (or higher) and Rendering> General settings> use Direct3D 11 for presentation> present a frame for every VSync

and tell me the result of your monitor behavior?

My screen seems to switch modes and render and present queue queue is not filled out. (1-3 / 14 and 1-3 / 14)

:thanks:

i don't have the card at the Tv anymore.

but don't waste your time on 10 bit and this screen.

[derpycat]

At the moment the new 3D features of LAV and madVR are for MVC-in-MKV only. There is no special handling of SBS files.

Sorry I'm a noob. Does this mean we can't play 3d Bluray yet?

[SamuriHL]

Sorry I'm a noob. Does this mean we can't play 3d Bluray yet?

Correct. Right now it only works with MKV's made with MakeMKV. At some point, once it matures a bit and some bugs are worked out, they'll expand support for blu-ray itself, albeit with some limitations. We'll get there, but, right now at this time, we're not there yet.

[derpycat]

Thanks, I guess it's easy enough to make my own MKVs. Was wondering why I couldn't get 3d working :p

[derpycat]

Edit: silly question

[madshi]

yes, "5" is working everytime i think. i didnt had any odd behavior with it. testet most settings with it and watched about 3 movies..
resizing works, fullscreen switching works, no drops, no stutter when i move teh mouse, video is in sync and normal speed, shaders work, just like with 0.89.19 :)

Sidenote:
only thing i noticed wich is different (i think, i didnt see that change before) is that in windowed mode, videos are played with 48fps and in Fullscreen (exclusive) mode its 60fps.
Wasnt it 60fps in windowed and fullscreen before? Of course only if you have 60hz ??
Or was this changed? Anyway, im totaly fine with that because it eats less power from my gpu in 48fps (windowed) in fullscreen i prefer teh 60fps like it is...
Can you please try this build:

http://madshi.net/madVRBetA13.rar

Does it work? If not, please create a debug log for me with this build (don't attach to this forum, upload it somewhere else and link here), which shows the "too fast playback". Also, please compare which refresh rates madVR lists in both windowed and fullscreen mode, with this build compared to v0.89.19. I suppose if you get the "too fast playback", probably the refresh rate isn't shown correctly?

[MSL_DK]

Could be running out of RAM

Yes, it is true in my case.

[BetA13]

Can you please try this build:

http://madshi.net/madVRBetA13.rar

Does it work? If not, please create a debug log for me with this build (don't attach to this forum, upload it somewhere else and link here), which shows the "too fast playback". Also, please compare which refresh rates madVR lists in both windowed and fullscreen mode, with this build compared to v0.89.19. I suppose if you get the "too fast playback", probably the refresh rate isn't shown correctly?

hy there,

:( nope this build also has the problem occuring..

umm, sorry for teh noobish question, but how do i do an debug report? i did try to look for an key combo but couldnt find one.

What i did do is compared teh OSD and refresh rate from windowed to fullscreen and from 0.89.19 to your Beta13 build..

And yes, you are correct, theres is something wrong with the "hz"

i made some screenshots:

madVRBetA13 build:

windowed mode:
http://s10.postimg.org/hqsygy105/Unbwenannt.jpg (http://postimg.org/image/hqsygy105/)

Fullscreen mode:
http://s28.postimg.org/bh0b96iop/sdfsdfsdf.jpg (http://postimg.org/image/bh0b96iop/)

0.89.19 Build (working Build):

windowed mode:
http://s18.postimg.org/q85k3v2g5/sfeegfsegfsf.jpg (http://postimg.org/image/q85k3v2g5/)

Fullscreen Mode:
http://s22.postimg.org/c04ixia3x/sefwfwfsf.jpg (http://postimg.org/image/c04ixia3x/)


Got the error report also, its quite big lol..

Steps i did:

Started debug log - started movie in windowed mode - after 3-5 sek. i changed to fullscreen mode - 2 sek. later i changed to windowed and quit the player, enden debug mode..

Link to Log:
http://filehorst.de/d/bAuhGbwH


i hope this will help you a bit... :)

greetings

[MSL_DK]

In the madVR folder run the debug.bat Can't remember the exact name of the file.

Then play a movie. On the desktop you will find the debug result, a txt file.

[BetA13]

In the madVR folder run the debug.bat Can't remember the exact name of the file.

Then play a movie. On the desktop you will find the debug result, a txt file.

ok, found teh bat. started it. left it open (teh cmd window) and started teh movie..etc.

now, wheres teh LOG?
i cant find it in the madvr folder or in c:/

[MSL_DK]

Right click the .bat run as admin, play a movie. After that, run the .bat again for leaving debug-mode

[BetA13]

Right click the .bat run as admin, play a movie. After that, run the .bat again for leaving debug-mode

right worked as intended. debug mode was started, video was played by me, video was stopped player exited and run the bat again to release debugmode again.

i feel very noobish about now since i still cant find a file .. theres no new file made , not in the madvr folder or c:

im pretty confused now haha

aha, you edited your comment, its on the desktop.. THANK you a lot mate.. haha

[apgood]

right worked as intended. debug mode was started, video was played by me, video was stopped player exited and run the bat again to release debugmode again.

i feel very noobish about now since i still cant find a file .. theres no new file made , not in the madvr folder or c:

im pretty confused now haha

aha, you edited your comment, its on the desktop.. THANK you a lot mate.. haha
It should be on your desktop

[BetA13]

hehe, yeah i got it now :)
Thanks for all the help from you guys ;)

@madshi

Got the error report also, its quite big lol..

Steps i did:

Started debug log - started movie in windowed mode - after 3-5 sek. i changed to fullscreen mode - 2 sek. later i changed to windowed and quit the player, enden debug mode..

Link to Log:
http://filehorst.de/d/bAuhGbwH


i hope this will help you a bit... :)

greetings

[clsid]

Next time zip/rar the log before you upload it. Then it will be much smaller, and the upload/download will be quicker.

[XMonarchY]

Does GeForce GTX 960 have any advantages over GTX 970 and 980 series? I know it supports newer CUVID decoding and is much faster at decoding HEVC than GTX 970/980 and Titan X.

- Can LAV Filter CUVID decoding also produce artifacts on GTX 960 like it can on GTX 970/980?
- Does GTX 960's faster CUVID HEVC decoding have any impact on DXVA Copy-Back decoding?

[huhn]

wrong thread...

the 960 doesn't use hybrid decoding and can decode 10 bit HEVC using an full function decoder that doesn't need GPU or CPU processing power.

the hybrid decoder from the 970 and higher is totally useless in comparison.

and forget CUVID plz...

[Thunderbolt8]

the 960 doesn't use hybrid decoding and can decode 10 bit HEVC using an full function decoder that doesn't need GPU or CPU processing power.is it save to assume that the GTX960 will be sufficient for 4K HEVC downscaling to 1080p with madvr in the future?

[huhn]

is it save to assume that the GTX960 will be sufficient for 4K HEVC downscaling to 1080p with madvr in the future?


depends on the algorithm. the 960 is not really powerful. i would set my eyes on the upcoming polaris or pascal GPUs.

[Shiandow]

is it save to assume that the GTX960 will be sufficient for 4K HEVC downscaling to 1080p with madvr in the future?

Should be reasonably safe. It depends a bit on how you handle the chroma, but really the most sensible thing to do would be to not scale the chroma which is slightly more complicated, but still doable. Downscaling the image also isn't particularly expensive, even this (http://screenshotcomparison.com/comparison/158879) new experimental downscaler runs quite comfortably on my GTX 960 (although strangely enough I take a bigger performance hit than on my late 560 Ti, memory speed seems to be the limiting factor). I can even run that algorithm in combination with use NNEDI3 (16 neurons) chroma scaling, which is a ridiculously inefficient way to shift chroma half a pixel.

[madshi]

Should be reasonably safe. It depends a bit on how you handle the chroma, but really the most sensible thing to do would be to not scale the chroma which is slightly more complicated, but still doable. Downscaling the image also isn't particularly expensive, even this (http://screenshotcomparison.com/comparison/158879) new experimental downscaler runs quite comfortably on my GTX 960 (although strangely enough I take a bigger performance hit than on my late 560 Ti, memory speed seems to be the limiting factor). I can even run that algorithm in combination with use NNEDI3 (16 neurons) chroma scaling, which is a ridiculously inefficient way to shift chroma half a pixel.
Did you have time to look into the potential patent issues yet? If there are no problems, I'd love to add your algo to madVR ASAP. I expect downscaling 4K content to 1080p will be a big thing, so good quality downscaling will be more important than ever. Thanks.

[Shiandow]

Did you have time to look into the potential patent issues yet? If there are no problems, I'd love to add your algo to madVR ASAP. I expect downscaling 4K content to 1080p will be a big thing, so good quality downscaling will be more important than ever. Thanks.

Somewhat. I'm quite sure the method I'm using is different enough from the method they're describing that it won't fall under such a patent application. Also, at the moment it is their responsibility to give sufficient warning of the potential existence of any patent, and given that the content their paper is quite frankly unpatentable it's somewhat unclear what exactly they are trying to patent. At the very least there don't seem to be any patents titled 'perceptually based downscaling of images'.

I could try to see if there's any way to contact them ask them for the patent number (if there even is one), although legally, I don't have to. In fact, I don't gain any legal advantage by making myself aware of any potential patents at this point.

[iSunrise]

@Shiandow
Wow, the results of that algorithm are quite astonishing. Sharpness of fine details is so much better, it's not even funny. With that algorithm it should be a lot less necessary to even use any sharpener at all when downscaling.

Does it have any side-effects like pronouncing edges with moving images or is it all about getting most of the detail transfered to lower resolution but NOT adding any kind of artificially sharpened edges?

It would be a BIG win if you and madshi could work together on this one.

I would also be highly interested to test it against madshi's "Catmul-Rom with linear-light" downscaling combination. If this is even better than that, then my hats go off to the inventor.

How does it compare performance-wise to say Bilinear or Catmull-Rom?

[madshi]

Somewhat. I'm quite sure the method I'm using is different enough from the method they're describing that it won't fall under such a patent application. Also, at the moment it is their responsibility to give sufficient warning of the potential existence of any patent, and given that the content their paper is quite frankly unpatentable it's somewhat unclear what exactly they are trying to patent. At the very least there don't seem to be any patents titled 'perceptually based downscaling of images'.

I could try to see if there's any way to contact them ask them for the patent number (if there even is one), although legally, I don't have to. In fact, I don't gain any legal advantage by making myself aware of any potential patents at this point.
Sounds good to me. So would it make sense to include your algo into this weekend's madVR build?

BTW, does your algo ring?

[TMDaines]

Ok. In some situations the default location might be ok. But often it is not. E.g. look here, playing a DVD (using an LAV test build with the new LAV DVD subtitle export to madVR) with MPC-HC with the "notify media player" zoom control option activated:

default position (http://madVR.com/doom9/stuff/MpcNotifyDefault.png) - | - move up = down (http://madVR.com/doom9/stuff/MpcNotifyMove.png)

Or here with the "notify media player" zoom control option deactivated:

default position (http://madVR.com/doom9/stuff/MpcDefault.png) - | - move up (http://madVR.com/doom9/stuff/MpcMoveUp.png) - | - move down (http://madVR.com/doom9/stuff/MpcMoveDown.png)

Isn't any of the move options always better (in this example at least)? I think having one subtitle line in the active video area and one subtitle line in the black bar is rather bad, and it happens quite often with the standard location.
I am using LAV Nightlies and the latest madVR but cannot get this to work with DVD subtitles. I have a DVD (the German DVD of Die innere Sicherheit) (http://www.amazon.de/Die-innere-Sicherheit-Special-Edition/dp/B00005U8PS) which is non-anamorphic and has a 1.66:1 film windowboxed into the 4:3 space. The subtitles on the DVD itself are partly in the video and partly in the black bars underneath. No matter what combination of settings I use, the DVD subtitles do not move. If I load a separate .srt file into XySubFilter, I can easily get the black bars cropped and the subtitles managed correctly.

Do I somehow need to activate the "new LAV DVD subtitle export to madVR"?

[Shiandow]

Sounds good to me. So would it make sense to include your algo into this weekend's madVR build?

BTW, does your algo ring?

That should be doable, the method itself is simple enough.

Haven't seen any ringing yet, or aliasing. Mostly it seems a pretty solid algorithm, but there might be some specific cases where things go wrong.

[Sunset1982]

That should be doable, the method itself is simple enough.

Haven't seen any ringing yet, or aliasing. Mostly it seems a pretty solid algorithm, but there might be some specific cases where things go wrong.

So let's check it out! :goodpost:

[Sunset1982]

The combination of Nevcairiel, Shaindow and madshi is epic!

Because of guys like these I love my hobby to get the best picture quality possible even more!

Just want to say thank you guys for your hard work!!!

[Thunderbolt8]

yeah, please implement this algorithm in madvr. Ive mailed the guy who made the algorithm paper about adding some 4k to 1080p downscaled comparisons which he said he wants to add, but so far nothing has happened yet at his website.

[har3inger]

4k-> 1080p may be big in the future, but for many users, downscaling is already being used all the time for 720p->1440p->1080p. I think it's a great idea to implement it ASAP for testing.

[Nevilne]

There's plenty of lower hanging fruit for downscaling, such as box or negative b bicubic as well.

[bcec]

@madshi: I noticed that on my nVidia 980 Ti, 1080p frame packed 3D 23Hz translates to really ~23.000fps (so about every 3-4minutes there's a glitch). If I remember correctly, it was mentioned that madVR forces 3D 1080p at 23Hz. Is there a way to change that to 24Hz instead? Both are supported by nVidia at 1080p frame packed 3D, and 24Hz will be a lot closer to 23.976 than 23.

AFAIK, there doesn't seem to be a way to override/create a custom 3D resolution the same way it exists for non-3D resolutions.

[Asmodian]

@madshi: I noticed that on my nVidia 980 Ti, 1080p frame packed 3D 23Hz translates to really ~23.000fps (so about every 3-4minutes there's a glitch). If I remember correctly, it was mentioned that madVR forces 3D 1080p at 23Hz. Is there a way to change that to 24Hz instead? Both are supported by nVidia at 1080p frame packed 3D, and 24Hz will be a lot closer to 23.976 than 23.

AFAIK, there doesn't seem to be a way to override/create a custom 3D resolution the same way it exists for non-3D resolutions.

23Hz means ~23.976Hz, if you really had 23.000Hz there would be a glitches continuously, not every 3-4 minutes.

A glitch every 3-4 minutes is a real refresh rate of ~23.971Hz or ~23.981Hz. My 980 Ti runs at about 23.9715Hz so I think you are exactly as expected. Sorry. :(

You will have to run Reclock or similar for 3D for now because smooth motion does not work with 3D (yet ;) ).