madVR - high quality video renderer (GPU assisted)

[madshi]

I have PCIe 2.0 and am using 16 neurons on 720p60->1080p60. Was hoping to see an improvement after your initial post of "huge gains" on AMD but that is not the case.

Oh well. :)
Some users seem to have huge gains, others none at all. I'm not sure why that is the case, but that's the way it is. Not much I can do about it.

I hope it will help :)

http://www24.zippyshare.com/v/hFoQpbqw/file.html
Thanks, but this sample is played as BT.709. So no misdetection here. If you want me to look into a misdetection, I need a sample which shows that misdetection... :)

Would be nice to see what build/version of madvr actually is running. If that would be shown on osd that would be good for error hunting. Everyone could easyly see if he is running the latest version or - if he updated the build - that the update worked and new version is running good. That came into my mind because you saw on some error logs that not the actual/userwanted version was running on the system. With this feature all guys could easily see what version is running.
Sorry, but no. You can double click the madVR tray icon to check the madVR version. If I added every bit information to the OSD that anyone ever asked for, the OSD would have 300 lines by now. Since the madVR version can very easily be seen elsewhere I don't think it's necessary to add to the OSD.

On a side note I've noticed something strange to me, when using xysubfilter latest beta, the rendering times in madvr are normal since the first show of subtitles on screen, after that the rendering times got higher with 1-2ms, even with "render to original video size" option selected, is that normal?
I'm not sure if 1-2ms rendering time differences are meaningful. Could just be measurement error? Would also make sense to double check this with older madVR builds to check if it's new, or has always been this way.

[Sunset1982]

Sorry, but no. You can double click the madVR tray icon to check the madVR version.

ok, no problem. You're right. Was just a quick idea. :)

[Siso]

I'm not sure if 1-2ms rendering time differences are meaningful. Could just be measurement error? Would also make sense to double check this with older madVR builds to check if it's new, or has always been this way.

No issues with xy-vsfilter, it only happens with xysubfilter.

[aufkrawall]

I'm not totally sure what you mean? If luma and chroma are both doubled (with the same settings), they're shown as "image doubling", instead of separate "luma doubling" + "chroma doubling". Although see above.

But is it normal that neither Jinc chroma scaling after doubling is shown, but only as pre-doubling chroma scaler?

[huhn]

You're saying you have the same render times regardless of how many neurons you're using? That sounds very strange! What happens if you go back to e.g. v0.88.21, do you have better render times with your NVidia GPU there?

you looked at the logs and SM was eating them for an unknown reason.

i'm pretty sure my old 760 GTX on windows 7 doesn't have this issue.not sure about my 960 GTX. i need nvidia 353.49 to test this right?

[aufkrawall]

I still sometimes have weird flickering/tearing after resizing MPC HC to windowed fullscreen and then resuming playback.
I think it might be related to D3D11, but can't say yet that D3D9 really isn't affected.

[XRyche]

This might have been my fault, maybe I send you a bad/wrong build, I'm not sure. In any case, please forget about the special build. Instead just send me a debug log with v0.89.4, that will serve the same purpose. Thanks!




Here is the log using NNEDI3 chroma upsampling and NNEDI3 image doubling: https://www.mediafire.com/?m6269d5iz4yi671

[Budtz]

I have no searchbar in FSE-mode since 89.0

[mogli]

Played around with the black bars detection:

Found content where sometimes shortly some stuff like snowflakes or raindrops extends into the black bars. In this case I consider zooming out unnecessary. Don't really know what to do here: A small delay before switching would help but probably causes trouble with other content.

Other content has a lot of small changes of bar sizes but enough to trigger a switch. Scaling to the mean image size or the most common one would probably be nice. Of course this may sometimes cuts off a bit of content or shows a bit of the black bars but that's imho better than constant switching.

[v0lt]

madVR v0.89.4
Option "disable desktop composition (Vista and Windows 7)" is not working.

[kasper93]

@madshi I don't know if you are still interested in black bars samples but here is another one black_bars_big_logo.mpg (https://www.dropbox.com/s/3tvbiu1pn72rwf5/black_bars_big_logo.mpg?dl=1) It is probably arguable what to do with "artist of the week" stripe. I'd crop it off since it is static over entire show and only annoy. But since it is pretty big it is probably hard to detect if it isn't some useful data. Anyway I just wanted to share this sample so you have it.

[Asmodian]

you looked at the logs and SM was eating them for an unknown reason.

i'm pretty sure my old 760 GTX on windows 7 doesn't have this issue.not sure about my 960 GTX. i need nvidia 353.49 to test this right?

No, as of 89.3 madVR uses a dx11 open cl interop. NNEDI3 now works with the current drivers.

[aufkrawall]

madshi asked him to test with an old madVR build that doesn't support the D3D11 interop, so he has to go back to a driver that still supports the old extensions. :)

[luk008]

The render queue and the present queue are not filling in D3D11 exclusive in 24HZ. I'm using MPC-HC, Win 8.1 x64, GPU is R9 270X.
Does win10 improve madvr performance or is it the same?
Thanks.

[huhn]

You're saying you have the same render times regardless of how many neurons you're using? That sounds very strange! What happens if you go back to e.g. v0.88.21, do you have better render times with your NVidia GPU there?

on 88.21 i get the same behaviour with nnedi3 and SM active at the same time. so it's old.

[leeperry]

I haven't seen that here yet. No idea where it comes from. FWIW, if I removed the digital certificate, it would very likely be exactly the same, or even worse.
Oh, well I run a low-end G3420 dualcore so maybe that's my problem but it certainly was going a lot faster like a year ago and other apps that give me the same authorization window make it appear pretty much instantly.

There definitely seems to be some serious crunching going on in the back before the OS is able to show that window.

videos of 1916 x 1076 or 1912 x 1072 - which for some reason are used by some commercial download sites. This scaling of less than 1% larger, is quite GPU intensive and results would be far better with no scaling.
Well yes and no, I just had a real ball with one of those 1% downscaled VOD movie tbh as I'm now totally hooked to the effect luma SR has on 24p motion blur :cool:

24p fast motion really looks sluggish as hell without luma SR and even chroma SR doesn't fix this(even though it does help), I'm kinda tempted to downscale 1080p by 1% in ffdshow in order to force mVR to run its SR magic on it but ffdshow can't quite process CR AR LL downscaling so I guess I'm SOL....madshi mentioned that he might consider allowing Supersampling (http://avisynth.nl/index.php/Supersampling) at some point and I got plenty of GPU cycles available for that now I've given up on NNEDI3. There's no magic bullet to this atm I presume, apart from getting a 4K display =/

[huhn]

Oh, well I run a low-end G3420 dualcore so maybe that's my problem but it certainly was going a lot faster like a year ago and other apps that give me the same authorization window make it appear pretty much instantly.

There definitely seems to be some serious crunching going on in the back before the OS is able to show that window.

the single thread performance of an pentium isn't that bad.
for me it opens kind of instant and there is no way my CPUs is that much faster.

[Ver Greeneyes]

There definitely seems to be some serious crunching going on in the back before the OS is able to show that window.IIRC it computes a hash of the file and compares it to a database of malware, which takes a while. Here's (http://www.sevenforums.com/tutorials/182353-open-file-security-warning-enable-disable.html) a guide on how to disable it for Windows XP, Vista and 7 at least, dunno if it works on higher versions. I'm not sure if there's anything an application developer can do to speed it up.

[StinDaWg]

I still sometimes have weird flickering/tearing after resizing MPC HC to windowed fullscreen and then resuming playback.
I think it might be related to D3D11, but can't say yet that D3D9 really isn't affected.
Same. Only affects D3D11 for me as well.

[ionutm80]

Hi guys, I need your help with an issue that I have when using madVR.
I still see "1 frame repeat every 0.97 seconds" in madVR stats and dropped frames each sec.
HW config: Core I5 3470s, GTX 750 TI, 8 GB DDR3 1600, 50Hz TV (100Hz option)
SW config: Win 10, MPC-HC, LAV Filters (Intel QS used for HW decoding in LAV Video to save NVidia power for madVR, LAV Audio set for bitstreaming of DTS and AC3), madVR, ffdraw, SVP.
madVR settings: chroma:BC75AR, image up:LZ3AR, image down: CRARLL, Luma Doubling 32 Neurons, smooth motion OFF, Error Difusion 2.
I'm bitstreaming sound through SPIDIF to a Home Cinema capable of DTS decoding so ReClock is not used.
Render time around 7-8 ms for 1080p blu-ray rips way below movie frame interval.
SVP shows 50fps in line with my TV resolution.

What could be the issue?

PS: Please excuse any stupidity if the case ...

[Razoola]

Sounds like the normal anomalies one would see playing 24fps at the wrong refresh rates. Try taking SVC out of the equation and using madVR's smooth motion option instead if your TV cannot do true 24, 48 or 96 refresh rates.

[Sunset1982]

Or try using a custom resolution in the nvidia control panel.

[ionutm80]

Sounds like the normal anomalies one would see playing 24fps at the wrong refresh rates. Try taking SVC out of the equation and using madVR's smooth motion option instead if your TV cannot do true 24, 48 or 96 refresh rates.

Hi, below are the resolution supported by my TV:

Supported display resolutions
• Computer formats (via HDMI/DVI input)
Resolution Refresh rate
640 x 480 60Hz
800 x 600 60Hz
1024 x 768 60Hz
1280 x 768 60Hz
1360 x 768 60Hz
1920 x 1080i 60Hz
1920 x 1080p 60Hz
• Video formats
Resolution Refresh rate
480i 60Hz
480p 60Hz
576i 50Hz
576p 50Hz
720p 50Hz, 60Hz
1080i 50Hz, 60Hz
1080p 24Hz, 25Hz, 30Hz
1080p 50Hz, 60Hz

It's because I have only 60Hz for PC no? I know Smooth Motion is recommended for this but I really like SVP effect. Is there any solution to keep SVP without frame repeats?

[ionutm80]

Or try using a custom resolution in the nvidia control panel.

Hi, tried that, it failed to apply. Thanks.

[edigee]

Hi, tried that, it failed to apply. Thanks.
AFAIK NVIDIA has issues with that(custom resolutions) on their latest drivers. Try an older one ,for windows 10 of course.

[huhn]

Hi, below are the resolution supported by my TV:

Supported display resolutions
• Computer formats (via HDMI/DVI input)
Resolution Refresh rate
640 x 480 60Hz
800 x 600 60Hz
1024 x 768 60Hz
1280 x 768 60Hz
1360 x 768 60Hz
1920 x 1080i 60Hz
1920 x 1080p 60Hz
• Video formats
Resolution Refresh rate
480i 60Hz
480p 60Hz
576i 50Hz
576p 50Hz
720p 50Hz, 60Hz
1080i 50Hz, 60Hz
1080p 24Hz, 25Hz, 30Hz
1080p 50Hz, 60Hz

It's because I have only 60Hz for PC no? I know Smooth Motion is recommended for this but I really like SVP effect.

these 100 Hz on TVs is marketing the input for hdmi 1.4 devices is up to 60 hz and nothing more. and you don't need to use a computer format.

Is there any solution to keep SVP without frame repeats?
reclock should be able to fix it.

[ionutm80]

these 100 Hz on TVs is marketing the input for hdmi 1.4 devices is up to 60 hz and nothing more. and you don't need to use a computer format.

reclock should be able to fix it.

Thanks, but by using ReClock I gave up on bitstreaming DTS/AC3 no?
At this moment the sound goes through SPDIF to a Philips Home Cinema which decodes DTS/AC3 ...

[edigee]

..........................................
reclock should be able to fix it.

Reclock works on 32 bit OS only!

[huhn]

Reclock works on 32 bit OS only!

no. it works on 32 bit players only that's a huge difference.

[aufkrawall]

Same. Only affects D3D11 for me as well.
Yes confirmed. It's totally reproducible here when
1. open a file
2. pause it
3. enter windowed fullscreen mode
4. instantly resume playback directly after this

I created a debug log:
http://www.mediafire.com/download/igbz9sayzry2rw7/madVR+-+log.7z

I also noticed that there is a slight delay with D3D9 before playback is resumed, which is not there with D3D11 (but instead it instantly starts tearing and flickering).

[madshi]

But is it normal that neither Jinc chroma scaling after doubling is shown, but only as pre-doubling chroma scaler?
I'm not sure what you mean. If you have a situation where you think the OSD shows something incorrectly, can you make a screenshot (or alternatively write down exactly what the OSD shows for chroma/image/luma/chroma scaling), compared to what you'd expect to see?

you looked at the logs and SM was eating them for an unknown reason.
on 88.21 i get the same behaviour with nnedi3 and SM active at the same time. so it's old.
Ok, not sure what to say/do then. Probably a GPU driver problem? Can't say for sure, though. It seems that most other users don't have this problem?

I have no searchbar in FSE-mode since 89.0
Weird. Is the problem still there with v0.89.5? And you do see a seekbar if you go back to v0.88.21?

Found content where sometimes shortly some stuff like snowflakes or raindrops extends into the black bars. In this case I consider zooming out unnecessary. Don't really know what to do here: A small delay before switching would help but probably causes trouble with other content.
This sounds like really weird content. Delaying a black bar detection change might be an option. But then the question is for how long? "Shortly" is a very relative term. Could be a couple of frames, or a couple of seconds. Delaying also means that in other situations for some frames you'd be losing real image content.

Other content has a lot of small changes of bar sizes but enough to trigger a switch. Scaling to the mean image size or the most common one would probably be nice. Of course this may sometimes cuts off a bit of content or shows a bit of the black bars but that's imho better than constant switching.
Most be more than 4 pixels difference then. I guess I could make the number of pixels which are considered "not enough to switch" adjustable. Alternatively have you looked into the "if black bars change pick one zoom factor" option? That would do something near to what you're suggestion. However, it would also become active for large aspect ratio changes (e.g. switching between cinemascope and 16:9 in IMAX Blu-Rays). I suppose you'd want a different setting for relatively "small" changes in black bar size?

madVR v0.89.4
Option "disable desktop composition (Vista and Windows 7)" is not working.
Strange, haven't changed anything in that code area for *ages*. Could you check which is the last madVR version which still has that working? I'm on Windows 8.1 x64 now, so it's hard for me to check. Old madVR builds available here:

http://www.videohelp.com/tools/madVR/old-versions#download

@madshi I don't know if you are still interested in black bars samples but here is another one black_bars_big_logo.mpg (https://www.dropbox.com/s/3tvbiu1pn72rwf5/black_bars_big_logo.mpg?dl=1) It is probably arguable what to do with "artist of the week" stripe. I'd crop it off since it is static over entire show and only annoy. But since it is pretty big it is probably hard to detect if it isn't some useful data. Anyway I just wanted to share this sample so you have it.
Weird samples are always useful, thanks. Currently I'm not really looking at "static" vs "dynamic" because it would require a whole lot more CPU power to do that. I'd have to actually compare different frames pixel by pixel for that. The "artist of the week" is sized too wide to be considered a TV station logo, so madVR considers it as part of the video frame. Probably not much I can do about it...

The render queue and the present queue are not filling in D3D11 exclusive in 24HZ. I'm using MPC-HC, Win 8.1 x64, GPU is R9 270X.
Is that with 10bit output or 8bit output? Some users have that problem with 10bit output, but usually it works fine with 8bit output. How about D3D9? How about windowed mode?

Does win10 improve madvr performance or is it the same?
I don't know, but I wouldn't expect a performance improvement.

I still see "1 frame repeat every 0.97 seconds" in madVR stats and dropped frames each sec.
Please show a cropped screenshot of the OSD (just the OSD). Please don't attach it to this forum (takes too long to approve).

Yes confirmed. It's totally reproducible here when
1. open a file
2. pause it
3. enter windowed fullscreen mode
4. instantly resume playback directly after this

I created a debug log:
http://www.mediafire.com/download/igbz9sayzry2rw7/madVR+-+log.7z

I also noticed that there is a slight delay with D3D9 before playback is resumed, which is not there with D3D11 (but instead it instantly starts tearing and flickering).
No problems here. Does the OSD show anything? Frame drops, presentation glitches, anything? Does the problem go away after some time or does it stay "forever"?

Here is the log using NNEDI3 chroma upsampling and NNEDI3 image doubling: https://www.mediafire.com/?m6269d5iz4yi671
The good news is that it's a log from the right build now. The bad news is that although it's still the same as before, I'm more confused than ever about why it's happening. It's a slightly different location than I had anticipated. Will need to create another test build to get to the bottom of this. But not today. Hopefully you'll be around for testing next weekend?

[madshi]

madVR v0.89.5 released

http://madshi.net/madVR.zip

* cleanup image borders can now optionally only crop black bar borders
* implemented "move OSD and subtitles into active video area" option
* madVR OSD now respects screen masking
* optimized OSD APIs to allow media players to respect screen masking
* optimized subtitle positioning for internal MPC-HC/MPC-BE subtitle renderer
* added APIs for media players to set zoom/position modes
* added IMadVRCommand::SendCommandInt("keyPress") support (J.River MC)
* some render code cleanup (fully removed AMD interop hack etc)
* fixed: one cause for shutdown crash with native DXVA decoding
* fixed: OSD fused different luma/chroma NNEDI3 neurons into "image"
* fixed: image enhancements "restore defaults" set LumaSharpen "clamp" wrong
Had planned to include some more changes, like some new file tags, or an up-to-date AdaptiveSharpen build, but ran out of time. Tried to update XySubFilter subtitle positioning, but will need a new XySubFilter build for that. Hopefully I can resurrect the developer for that at least, but I have my doubts. We'll see...

[aufkrawall]

No problems here. Does the OSD show anything? Frame drops, presentation glitches, anything? Does the problem go away after some time or does it stay "forever"?

OSD doesn't show anything interesting (I think there's just a small bug with Windows 10 or NV driver which makes it sometimes fail to read out the render queue right, it's then half of the present queue).
I also sometimes notice that there is flickering for one or two frames after pausing in windowed mode.
It might be still a NV driver bug, the flickering disappears when showing a menu or the seekbar of MPC HC. We already had this in an even worse shape during Windows 10 beta.

[ashlar42]

madVR v0.89.5 released

http://madshi.net/madVR.zip


* implemented "move OSD and subtitles into active video area" option
* optimized subtitle positioning for internal MPC-HC/MPC-BE subtitle renderer
Had planned to include some more changes, like some new file tags, or an up-to-date AdaptiveSharpen build, but ran out of time. Tried to update XySubFilter subtitle positioning, but will need a new XySubFilter build for that. Hopefully I can resurrect the developer for that at least, but I have my doubts. We'll see...
Do you mean this allows for subtitles rendering on black bars or that you couldn't do it due to support from XySubFilter's dev?

[SweetLow]

If I added every bit information to the OSD that anyone ever asked for, the OSD would have 300 lines by now.
Are you remember my wishes of text mode control for this? Or maybe hotkey for copy of this info into clipboard in text form? ;)

[kongfl888]

Thanks for another release.

[madshi]

OSD doesn't show anything interesting (I think there's just a small bug with Windows 10 or NV driver which makes it sometimes fail to read out the render queue right, it's then half of the present queue).
I also sometimes notice that there is flickering for one or two frames after pausing in windowed mode.
It might be still a NV driver bug, the flickering disappears when showing a menu or the seekbar of MPC HC. We already had this in an even worse shape during Windows 10 beta.
Can't reproduce it here on Windows 8.1. Suppose it's probably a Windows 10 NVidia GPU driver problem. Hope they'll sort it out soon. But D3D9 works ok?

Do you mean this allows for subtitles rendering on black bars or that you couldn't do it due to support from XySubFilter's dev?
Couldn't do any changes due to XySubFilter's API support being incomplete. Need a new XySubFilter version. I've already implemented all necessary code in madVR, but can't activate it yet, doesn't work with currenty XySubFilter build.

Are you remember my wishes of text mode control for this?
Text mode control? Not sure what you mean.

[madshi]

@media player devs

Please have a look at this PDF:

http://madshi.net/89notesForDevs.pdf

[aufkrawall]

Can't reproduce it here on Windows 8.1. Suppose it's probably a Windows 10 NVidia GPU driver problem. Hope they'll sort it out soon. But D3D9 works ok?

Yes, so far no problems with D3D9.
And FSE seems to be working fine with D3D11, so it's not a big issue after all. But people should be aware of this.
I suppose it would disappear if I installed a pre r352 Nvidia driver, but they are from quite some time before the Windows 10 release.

Edit: With version 89.5, my issue regarding shown chroma scaler with doubling/quadrupling has been solved. :)

[XRyche]

The good news is that it's a log from the right build now.

Well at least there is that. I thought is was progressively getting stupider every-time I want to the restroom lately.



The bad news is that although it's still the same as before, I'm more confused than ever about why it's happening. It's a slightly different location than I had anticipated. Will need to create another test build to get to the bottom of this. But not today. Hopefully you'll be around for testing next weekend?

Yes, I'll be around next weekend for testing. No problem.

P.S. With the new build NNEDI3 chroma upsampling is visually working as it should, no discolouration (green tinge) but still heavy fps skipping before I get an image. NNEDI3 image doubling remains unchanged.

EDIT: Actually NNEDI3 is visually correct for chroma upsampling only when you activate the SuperRes filter.

[zveroboy]

madVR v0.89.4
Option "disable desktop composition (Vista and Windows 7)" is not working.


Strange, haven't changed anything in that code area for *ages*. Could you check which is the last madVR version which still has that working?
The problem started with v.0.88.6
http://forum.doom9.org/showthread.php?p=1722055#post1722055

Ups... Sorry. I checked again - the problem started with madVR-0.88.2

madVR-0.88.1 -- OK
madVR-0.88.2 -- "disable desktop composition" is not working

[Stan]

I updated from .2 to madVR v0.89.5 today and since then I cannot move an already playing video frame (mpc-hc) to a second monitor. The picture just freezes but the sound continues to play.

[Asmodian]

I updated from .2 to madVR v0.89.5 today and since then I cannot move an already playing video frame (mpc-hc) to a second monitor. The picture just freezes but the sound continues to play.

Any more specifics? This works fine on my system:

Win10 x64, 980Ti with 355.98, MPC-HC 1.7.9 x64, D3D9 overlay, madVR v0.89.5. I also tested D3D9 windowed and had no issues with it either.

I can drag it back and forth while playing without any trouble.

[Stan]

Any more specifics?

Win8.1 x64, 980Ti with 355.98, MPC-HC 1.7.9.165 (nightly) x64, madVR v0.89.5, not sure about the overlay mode, can't find that setting atm. I tried with the last mpc-hc final release and it still happens with it. I reset madVR settings, still happens. Using mpc-hc's EVR works. madVR 0.89.3 and .4 work so I'm pretty sure I nailed it down to .5 breaking it for me.

[6233638]

Can't reproduce it here on Windows 8.1. Suppose it's probably a Windows 10 NVidia GPU driver problem. Hope they'll sort it out soon. But D3D9 works ok?There is definitely some weirdness with D3D11 in Windows 10.
I couldn't say whether it's the NVIDIA drivers, DWM changes, the introduction of virtual desktops/aero snap changes, or something else which is causing it.
I had to reduce the queue sizes to 6 to at least minimize how often things would occur.
I would end up with out-of-order frame presentation (looked like movement was jumping back and forth), screen tearing, or dropped frames.
And when playing multiple videos in succession in JRiver, it would "lose focus" after the first video and play in FSW rather than FSE - despite being in front of all the other applications.

None of this was consistently reproducible, which is why I haven't filed a bug report.
Due to those issues with madVR, and a lot of issues with other applications - including the OS itself - I've had to return to Windows 8.1.
10 really feels like a beta release right now, and I don't have the time to spend troubleshooting it at the moment.
There are definitely some things that it improves upon compared to 8.1, but it just doesn't feel finished.

[FreeFall]

* implemented "move OSD and subtitles into active video area" option

Is this a user configurable option? I can't find it in the zoom control settings tab or anywhere else, using Windows 7 x64 Pro.

Thanks.

[chros]

Guys, can I test chroma upscaling with image comparison like madshi did with luma in this post? http://forum.doom9.org/showthread.php?p=1730855&highlight=ground+truth#post1730855
If yes, then can you provide an image for this? Or a test clip with the same image which was compressed lossless?

Regarding to this question, I have been using Jinc+AR for chroma upscaling:
1. do you recommend super-xbr+AR for this instead?
2. do you recommend to turn on superes 1 pass (I don't have the power for 2 passes) for chroma upscaling?

And thanks for the new version!

[mrbinky]

After upgrading to v0.89.5, when I try to use Madvr to scale any video, the process quickly eats up my available memory, and then the video stops playing. At which point, my video player (latest 64 bit MPC-HC and 64 bit LAV) or video driver (latest catalyst beta) might crash.. If the later happens, then I need to reboot to recover. I am running Windows 10, 16 GB with plenty of free memory.

I tried deleting the OpenCL key in regedit to no avail.

I tried going back to v0.89.4, to no avail. I suppose I could try deleting OpenCL key again, but I was impatient to watch my movie, so I switched to EVR, and everything was fine.

I noticed that I can play a video at it's native resolution with Madvr active, but as soon as I start scaling it up, even just a bit, memory starts getting eaten. Even if I turn off NNEDI3 everyplace.

Hopefully next update will fix this. I might be able to do some additional troubleshooting or get some logs after I get some from work, but since the whole family relies on this for watching movies, and they are impatient, I cannot promise anything. Might just need to stay away from Madvr until it is usable again.

[aufkrawall]

It's bug of the latest Catalyst 15.9 beta driver, happens also with browsers:
http://www.forum-3dcenter.org/vbulletin/showpost.php?p=10795845&postcount=11

[mrbinky]

It's bug of the latest Catalyst 15.9 beta driver, happens also with browsers:
http://www.forum-3dcenter.org/vbulletin/showpost.php?p=10795845&postcount=11

I forgot to mention that I had also tried downgrading my video driver to the official release, and it made no difference.

EDIT: I forgot that you have to actually uninstall AMD's beta upgrade. Now it seems to be working. Thanks!