View Full Version : madVR - high quality video renderer (GPU assisted)
FIX94
8th September 2015, 21:13
AMD is known to be a bit unstable with DXVA2, well, at least I have heard plenty of reports and experienced a bunch of fun myself. You might have found one of those fun occasions.
So I guess for others to reproduce it they might need a AMD card, LAV with DXVA2 and madVR at the same time? No wonder then this issue was not reported. By the way, LAV is awesome, its so good to just have one software instead of some huge codec pack while keeping everything fast, great job on all of that!
huhn
8th September 2015, 21:18
SR on: http://i.imgur.com/l0aNhH1.png
SR off: http://i.imgur.com/maff9s2.png
i use nnedi3 32 for chroma and doubling chroma
you use refine every x2 right?
i get the same and it is not recommended to use superes more than 1 time.
http://abload.de/img/testisthardrgszc.png
Sarasa
8th September 2015, 21:37
@madshi
in the rar, my settings.bin & the file 638*480 used for the test
http://dl.free.fr/getfile.pl?file=/7THddRTo
ADude
8th September 2015, 22:11
See the first image in that post 13 posts back. The black bars I'm talking about are those "encoded letterboxes" described in that image. Also look at all the other images in that post and try to understand what purpose they have in the context of that post, then you should get a pretty good idea what it's all about.
As I mentioned, I know what a "black bar" is.
In the past, the player was dealing with all those issues. (In fact, in response to a request I made months ago, someone replied that "the player should be taking care of that")
What I am missing is what madVR is now doing (when the new feature(s) are used).
It is changing the frame size prior to doing up/down-scaling ?
(Perhaps 6233638 knows what I am asking?)
huhn
8th September 2015, 22:34
if the frame size is 1280x960 (4/3) with a 16/9 picture with hard coded black bars it is changing the picture size to 1280x720.
depending on other settings other things will happen now with that picture.
is there someone else who can't change the anamorphic lens stretch factor ?
Weirdo
8th September 2015, 22:44
I think the image has shifted to the top (not centred any more) on both my media players - haven't touched zoom settings. Happens in full-screen mode only.
ADude
8th September 2015, 23:23
if the frame size is 1280x960 (4/3) with a 16/9 picture with hard coded black bars it is changing the picture size to 1280x720.
depending on other settings other things will happen now with that picture.
Thanks for your reply !
So if I have a madVR profile that says "IF 960 THEN..." will it do the action? Or will it now only satisfy "IF 720 THEN... "
Also, what is "screen masking" ?
ajp_anton
8th September 2015, 23:28
Using which DXVA decoder? LAV Video Decoder? Is it the media player which crashes or madVR? See the comment above this one.
I added more info in later posts (http://forum.doom9.org/showthread.php?p=1735180#post1735180), maybe you missed them. LAV yes, and it was madVR that crashed.
Anyway, I can no longer reproduce the issue, so it might have been a driver bug, because I've updated them since the problem.
huhn
8th September 2015, 23:52
So if I have a madVR profile that says "IF 960 THEN..." will it do the action? Or will it now only satisfy "IF 720 THEN... "
it ignores profiles.
i guess it still take the source size for a simple reason there is no way the algo knows if the whole file is a 4/3 file with 16/9 content.
Also, what is "screen masking" ?
you kind of say madVR how many black pixel should be at each boarder. this can be used for over scanning TV/projector or a not optimal projector screen setup. with other words it adds black boarders.
zvans18
9th September 2015, 00:15
you use refine every x2 right?
i get the same and it is not recommended to use superes more than 1 time.
http://abload.de/img/testisthardrgszc.png
nope, just the once
jkauff
9th September 2015, 03:18
madVR v0.89.2 released
http://madshi.net/madVR.zip
* fixed: one more media player startup crash
* fixed: DXVA deinterlacing and black bar detection didn't like each other
* fixed: DXVA scaling had some problems with v0.89.0 and v0.89.1
* fixed: some media player size/pos zoom wishes were not properly applied
My media player black screen problems (MC21, MPC-HC) didn't go away until I ran "restore default settings.bat".
Thought I'd mention it in case anyone else has the same problem.
fedpul
9th September 2015, 03:30
Many many thanks madshi, 0.89.2 now works with JRiver MC 20 and keep working with MPC-HC. The only thing I did notice is that subtitle positioning is wrong with some movies in JRiver when you choose No Changes in proportion settings, when you change to Expand it is fixed. Another thing i noticed is that with 0.89.2 NNEDI performance has decreased considerably. I was using NNEDI 64 without problems and now i can barely keep up with NNEDI 32.
PD: After more testing JRiver does keeps crashing with some movies.
Bug report link: http://www.megafileupload.com/duS0/bugreport.txt
Thanks in advance.
I really appreciate your hard work.
BrainDedd
9th September 2015, 05:49
madVR v0.89.2 released
http://madshi.net/madVR.zip
* fixed: one more media player startup crash
* fixed: DXVA deinterlacing and black bar detection didn't like each other
* fixed: DXVA scaling had some problems with v0.89.0 and v0.89.1
* fixed: some media player size/pos zoom wishes were not properly applied
Perfect. DXVA scaling is working awesomely. :)
Thanks madshi!
khanmein
9th September 2015, 07:26
image upscaling Jinc tab is grey-ed out. unable to choose 3 taps.
madshi
9th September 2015, 08:13
SR on: http://i.imgur.com/l0aNhH1.png
SR off: http://i.imgur.com/maff9s2.png
i use nnedi3 32 for chroma and doubling chroma
in the rar, my settings.bin & the file 638*480 used for the test
http://dl.free.fr/getfile.pl?file=/7THddRTo
Thanks, will try next weekend to reproduce it.
So I guess for others to reproduce it they might need a AMD card, LAV with DXVA2 and madVR at the same time? No wonder then this issue was not reported.
It's not that rare a combination. Of all the madVR users in this thread most are using LAV, many DXVA2 copyback, and AMD GPUs are quite common, too.
As I mentioned, I know what a "black bar" is.
In the past, the player was dealing with all those issues.
How can the player do that? The player doesn't even know if there are black bars encoded into the video or not.
What I am missing is what madVR is now doing (when the new feature(s) are used).
It is changing the frame size prior to doing up/down-scaling ?
It depends on the settings. Currently madVR simply overwrites the zooming factor and position of the video, to adjust for the detected hard coded black bars. In a future version, if you activate the (not yet available) "clear black bars" option, madVR will probably actually crop the hard coded black bars away completely. But in the end that's just a difference in implementation. The end effect (what is visible on screen) shouldn't differ that much.
Also, what is "screen masking" ?
It's only for front projection. I'll remove these settings for all display types but front projection in a future version.
I think the image has shifted to the top (not centred any more) on both my media players - haven't touched zoom settings. Happens in full-screen mode only.
Totally to the top or just a tiny bit? You're saying the image isn't centered vertically, anymore? That sounds strange. Especially that it only happens in full-screen mode!
I added more info in later posts (http://forum.doom9.org/showthread.php?p=1735180#post1735180), maybe you missed them. LAV yes, and it was madVR that crashed.
Anyway, I can no longer reproduce the issue, so it might have been a driver bug, because I've updated them since the problem.
Ok, thanks.
My media player black screen problems (MC21, MPC-HC) didn't go away until I ran "restore default settings.bat".
Not sure why, though.
Another thing i noticed is that with 0.89.2 NNEDI performance has decreased considerably. I was using NNEDI 64 without problems and now i can barely keep up with NNEDI 32.
That should not be the case, are you sure? Which previous version did you compare to? Make sure you test with the same video. Also try to disable the black bar detection, just to make sure that doesn't affect things (although it shouldn't).
PD: After more testing JRiver does keeps crashing with some movies.
Bug report link: http://www.megafileupload.com/duS0/bugreport.txt
The crash report says you're still using v0.89.1.
Perfect. DXVA scaling is working awesomely. :)
:)
image upscaling Jinc tab is grey-ed out. unable to choose 3 taps.
3 taps is the only option now, that's why the taps option field is greyed out. I've removed 4 taps and 8 taps because they didn't bring any noticeable quality improvement.
wanezhiling
9th September 2015, 11:27
Latest madVR 0.89.2 has an issue with PotPlayer.
Play a file and turn into maximized size, the image will zoom much bigger than normal. http://i1.tietuku.com/2e6e6bab0be6fb69.png
The normal should be: http://i1.tietuku.com/44f1b8a37618c6fb.png
ps: no this issue with 0.89.1 and older versions
madshi
9th September 2015, 11:46
Latest madVR 0.89.2 has an issue with PotPlayer.
Play a file and turn into maximized size, the image will zoom much bigger than normal. http://i1.tietuku.com/2e6e6bab0be6fb69.png
The normal should be: http://i1.tietuku.com/44f1b8a37618c6fb.png
ps: no this issue with 0.89.1 and older versions
Is that with the black bar detection stuff enabled or disabled? If it's enabled, does it make a difference if you disable it? On a quick check I wasn't able to reproduce the issue here with the PotPlayer version I have installed. It's a rather old build, though.
Weirdo
9th September 2015, 11:48
Did a settings reset (or sth. else I didn't notice...) and this issue is solved, thanks. Yes, it wasn't centred vertically, being fully at the top with all video still visible (on screen).
Totally to the top or just a tiny bit? You're saying the image isn't centered vertically, anymore? That sounds strange. Especially that it only happens in full-screen mode!
wanezhiling
9th September 2015, 11:53
@madshi
I use default madVR settings and didn't touch any config yet, it's a fresh installation. So it's disabled, right?
I use latest PotPlayer 1.6.56209, released today.
madshi
9th September 2015, 12:36
I use default madVR settings and didn't touch any config yet, it's a fresh installation. So it's disabled, right?
I use latest PotPlayer 1.6.56209, released today.
Yes, it should be disabled then. Can you create a debug log for me from the incorrect zoom situation?
wanezhiling
9th September 2015, 12:44
@madshi
https://www.sendspace.com/file/4irlv0
madshi
9th September 2015, 12:55
https://www.sendspace.com/file/4irlv0
Thanks, I can see the problem, but I probably won't have time to fix it until next weekend.
wanezhiling
9th September 2015, 13:11
That's fine, thank you.
Hmm.. I reported the issue to PotPlayer dev before posting here becasue I thought that was a player's problem...0.0
madshi
9th September 2015, 13:17
No, it's my fault.
videonerd
9th September 2015, 15:11
Ok, I've double checked. The problem is that playrate changes are done by completely stopping the graph and then restarting it. This 2 second timeout I have for seeks only works because I know in advance when a seek occurs. But I don't know when playrate changes. MPC-HC doesn't tell me that playrate changes. I just see that the graph is stopped. And I think it would not be good if I waited multiple seconds then until I resumed rendering. So at the moment there's not so much I can do about it. I'll put it on my to do list, though, maybe I'll find a better solution in the future.
Hi madshi - thank you. Since an enhancement to madvr is done to satisfy a subset of users (certain players rendering during pause) but results in an undesired side effect (blank during undetectable graph restarts) with no ready solution, how about adding a variable (ini is fine, need not be in GUI) to turn on/off this new feature? Thank you once again.
aufkrawall
9th September 2015, 15:23
Stupid question: When you said you'd like to look into the NNEDI3 issue, did you mean the coming weekend or the weekend after it? :o
ibius
9th September 2015, 18:39
Hi madshi - thank you. Since an enhancement to madvr is done to satisfy a subset of users (certain players rendering during pause) but results in an undesired side effect (blank during undetectable graph restarts) with no ready solution, how about adding a variable (ini is fine, need not be in GUI) to turn on/off this new feature? Thank you once again.
Same here, would love to be able to turn that thing off (like creating empty file system object "ShowBlackBars").
Meanwhile I'm on v0.88.16 and looking forward to "clear black bars" option.
Thank you.
nussman
9th September 2015, 18:56
Stupid question: When you said you'd like to look into the NNEDI3 issue, did you mean the coming weekend or the weekend after it? :o
Oh man ... :devil:
edigee
9th September 2015, 19:15
Speaking about the NNEDI3 issue ,is this one present on windows 10 only? I'm on W7 64 bit ,GTX 960 ,madVR v0.89.2 ,MPC-HC. Clicking Ctrl-J it shows NNEDI3 working OK on chroma upscaling. Driver is 355.83
aufkrawall
9th September 2015, 19:18
You alrady have your answer. ;)
Yes.
Different topic:
Shouldn't madVR pick BT.2020 for UHD 10/12 bit content?
With 10 bit it guesses BT.709 here.
huhn
9th September 2015, 20:08
back in the days UHD content wasn't using BT 2020 that's why it was changed.
the best way is to stop this guessing by simply putting the colorspace information in the encodes...
aufkrawall
9th September 2015, 20:09
Unfortunately, streams often aren't flagged right.
From my perspective, it seems to be cleaner to guess like the specs expect.
If a stream deviates from the specs, it should be flagged that way. That's usually how specs and their exceptions are supposed to work, aren't they?
baii
9th September 2015, 20:19
I think most if not all still use bt709? Can't seem to find sample encoded/mastered in bt2020.
If madvr use bt709 for proper bt2020 video , the video should look wash out i believe?
edit: downloaded life of pi demo, it detects bt2020 fine. Those in youtube are... idk will test after finish download.
obviously they are bt709 if they are from youtube...
the creator for the youtube bt2020 demo say put a eizo coloredge in bt2020 mode to see it, yea yea yea....
kasper93
9th September 2015, 20:21
All 1920x1080 files are reported as 1918x1078 in recent build. Something is wrong.
aufkrawall
9th September 2015, 20:24
They are correctly reported as 1920x1080 here. But is it normal that the target rectangle isn't shown by the OSD, even when scaling the picture up or down?
kasper93
9th September 2015, 20:34
Hmm, something about settings. It works fine after restoring defaults. And I can't seem to find which one is the cause.
"Clean up image borders" option was the issue. Even though black bars detection was disabled it still was cropping.
kolak
9th September 2015, 20:50
You alrady have your answer. ;)
Yes.
Different topic:
Shouldn't madVR pick BT.2020 for UHD 10/12 bit content?
With 10 bit it guesses BT.709 here.
No, because there is no real BT.2020 content neither display to show it. Forget about BT.2020 at the moment- it's more theory than real thing.
If you want bigger gamut than Rec.709, use P3 and one of the latest TVs, as they are close to be able to display it.
ADude
9th September 2015, 20:50
It depends on the settings. Currently madVR simply overwrites the zooming factor and position of the video, to adjust for the detected hard coded black bars. In a future version, if you activate the (not yet available) "clear black bars" option, madVR will probably actually crop the hard coded black bars away completely. But in the end that's just a difference in implementation. The end effect (what is visible on screen) shouldn't differ that much.
What if the area of the video that is not black, is not the same frame size as the screen? Does that new smaller frame size trigger the madVR scaling profiles and algorithms ?
For example, suppose the screen is 1920x1080 and the area inside the black bars is 1280x720. Will madVR using the same scaling profiles and algorithms that would be used if the actual stated frame size of the video was 1280x720 ?
aufkrawall
9th September 2015, 21:07
No, because there is no real BT.2020 content neither display to show it. Forget about BT.2020 at the moment- it's more theory than real thing.
Ok, if there really is no content at all, it makes of course sense to not slavishly stick to the specs.
baii
9th September 2015, 21:30
Ok, if there really is no content at all, it makes of course sense to not slavishly stick to the specs.
there is one on the web and madvr detect it just fine.
XMonarchY
9th September 2015, 21:33
Yey me! I made NVidia listen, I think...
Here's my "Bug Report":
Good morning, NVidia!
I would like to contact a higher-level technician to communicate a problem that affects GeForce drivers for Windows 10.
At the moment, there is only one Windows 10 driver (353.49 Hotfix driver) that supports required OpenCL extensions to run all the features of madVR renderer, which is a very popular, important, prominent, and well-known mainstream renderer in the field of HTPC's. Other drivers, such as the latest 353.62 WHQL driver does NOT come with appropriate OpenCL extension - "cl_nv_d3d9_sharing" or "cl_khr_dx9_media_sharing" , which are required for madVR's important features, such as NNEDI3 chroma upscaling, to work. Basically, the latest NVidia Windows 10 drivers lack D3D9 <-> OpenCL interop support!!!
NVidia Windows 7/8/8.1 drivers always came with these important OpenCL extensions and it is vital that NVidia re-introduces these extensions and D3D9 <-> OpenCL interop support in Windows 10 drivers.
Please understand that the issue is in NVidia drivers and lack of proper OpenCL extensions and D3D9 <-> OpenCL Interop support. The issue is NOT within madVR, which simply relies on OpenCL to provide the best quality rendering.
At the moment AMD drivers appropriately support D3D9 <-> OpenCL Interop in Windows 10, while NVidia Windows 10 drivers do not. This may affect people's decisions and push them towards buying AMD cards due to more complete OpenCL driver support for madVR.
I cannot stress enough the importance of inclusion needed OpenCL extensions in NVidia Windows 10 drivers.
Doom9, HTPC, and madVR communities would at least like to know whether NVidia plans of re-introducing appropriate D3D9 <-> OpenCL Interop support and extensions in future Windows 10 drivers.
Please visit this thread for more information - http://forum.doom9.org/showthread.php?t=146228
Sincerely,
Response:
Hello ,
Thank you for contacting NVIDIA Customer Care. We're extremely sorry for the delay in getting back to you .
Your case is being escalated to our Level-2 team. The technician from the Level-2 team will review the case notes and will contact you as soon they can to assist or point you in the right direction.
Best Regards,
Abrar
NVIDIA Customer Care
And THEN:
Hi Sergey,
The case was escalated to me. We have received reports of this issue and have replicated the failure in our lab. The bug has been escalated to our development team to investigate. Thanks for bring this to our attention.
Best regards,
Ray
I know I am not a beneficial contributor in this thread, but I'm trying :helpful:, although I think madshi has a more direct communication with NVidia developers.
huhn
9th September 2015, 21:44
Ok, if there really is no content at all, it makes of course sense to not slavishly stick to the specs.
that was about two years ago.
i'm pretty sure this will change again when the UBD is out.
and if you see a stream that's not correctly flag bring it up to the maker of the stream
XRyche
10th September 2015, 08:48
Does using a different (older or newer) GPU driver help, maybe?
I've updated to a new Windows 10 Beta driver and it didn't make a difference. Still getting inordinately long wait times when any type of image doubling is used. It actually skips 500-700 frames before I actually get an image on screen including using NNEDI3. With NNEDI3 I am still getting an embossed effect with the image. Also, I didn't notice until someone posted about it but, madVR isn't using FSE mode in FS only FS windowed. I tried switching to from using D3D11 to using D3D9 and that didn't make a difference either.
I also tried using a very, very old build of MPC-HC (actually JanWillem32's) and I am getting the exact same behavior as with the latest stable main branch build of MPC-HC.
Also thank you for not leaving Windows 10 users till later and addressing our issues now.
fedpul
10th September 2015, 12:34
madshi, I have been doing several tests and finally I was able to identify what was causing the crashes with JRiver MC 20. It was NNEDI. The instant I choose it as Luma Doubling, then stop the movie, then click play and it crashes instantly. Also I find another problem, DX11 windowed mode is very slow (at the point i can't even use Super-XBR) in comparison with DX9 windowed, there is no such difference in FSE.
OS: Windows 7 x64
My hardware is:
CPU: Intel i5 3570k @ 4.5 GHz
RAM: 16 GB DDR3 @ 1600 MHz
GPU: Sapphire HD 7870 OC @ 1100/1375 MHz
If you need more info, or want me to test more things just tell me.
Thanks
Sunset1982
10th September 2015, 14:22
@ madshi
I updated madvr from 0.88.21 to 0.89.2 today. Now madvr crashes everytime I start a movie. Send you the crashlog via madvr.
when I reset to defaults, movie starts but I got a big red "out of memory" message on screen. When I set up my prefered options in madvr it crashes again.
reverting back to 0.88.21 does not cure the problem even if I reset to default settings...
Edit: problem solved. Wrong bios setting (PCI-E Lanes "16/0" instead of "Auto"; changed it to Auto solved the problem) Now everything works great with 0.89.2
daert
10th September 2015, 14:29
I have a general question: does overlay work in d3d11? I mean: if I use d3d9, osd shows d3d9 fullscreen overlay, but if I choose d3d11, it shows d3d11 fullscreen windowed (without the "overlay")
ibius
10th September 2015, 14:36
It does not.
Sunset1982
10th September 2015, 14:40
someone has a link to the registry hack or the thread page to force madvr to use nvidia gpu instead of onboard gpu?
daert
10th September 2015, 14:52
It does not.
Thanks
CiNcH
10th September 2015, 17:01
I still have the failing DXVA processing with .2 when switching channels (1080i) with a subsequent switch-off of the deinterlacing.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.