madVR - high quality video renderer (GPU assisted)

[James Freeman]

For some reason only the 960 has the hardware to decode x265/HEVC, the 970 or 980 do not.
https://en.wikipedia.org/wiki/Nvidia_PureVideo#Feature_Set_F

You should wait for next gen (Pascal) if you're a gamer and don't buy the 960.

[Patrik G]

For some reason only the 960 has the hardware to decode x265/HEVC, the 970 or 980 do not.

that was bad.
but ok i dont play any games so i guess a 960 would do the job for movies only.

Edit: i probably wait for the nextgen Nvidia cards.
its a huge dip in performance from a 780 with the 960s.
it feels like an downgrade.

[Stereodude]

The only GPU of current gen that able to decode HEVC is Nvidia GTX960, and yes, the CPU and GPU remain around 0% because the VPU completely takes the full load.
That's very misleading unless you add the 10-bit caveat to it. Intel has full HEVC decoding for 8-bit in Skylake and some of the low power solutions like Braswell too.

[James Freeman]

Yes, thanks for the correction Stereodude.
Intentions are all good.

Still, if I were to build a new HTPC with Nvidia GPU, I'd wait for Pascal which should begin to appear on Q3 of 2016.

[iSeries]

The only GPU of current gen that able to decode HEVC is Nvidia GTX960

Plus the GTX950.

[macycat]

I wrote this a while back:

It looks like I found a problem created between madVR v0.89.12 and v0.89.13. It took some time to track this down since I hadn't upgraded in quite a long time.

The problem occurs with 1080i59 content (recorded with HDPVR), but does not seem to be a problem with 1080p23 content. While playing a 1080i59 video file, if I double click on another 1080i59 video file to play it in the same video player window, I get a blank screen with sound from the new video. I also previously was able to drag the video player window from one screen to another screen with no problems (just a minor pause in playback), but that now usually gives a blank screen with audio playing with an occasional DirectX fails to render crash.

MPC-HC 1.7.10 (64-bit)
LAV Filters 0.67.0
madVR 0.89.12 - ok
madVR 0.89.13 - fail

Windows 7 64-bit
NVIDIA GTX 670 video card

madVR settings:
D3D9 overlay
NV12, 8 bit, 4:2:0
movie 29.970 fps (says source filter)
movie 1920x1080, 16:9
touch window from outside
chroma > Jinc AR
image < Lanczos3 AR
vsync 16.68ms, frame 16.68ms
matrix BT.709 (says upstream)
primaries BT.709 (says upstream)
limited range (says upstream)
deinterlacing on (settings)
...

I haven't updated for quite a while, but with all the great work being done, I thought I would give the 3D playback a try. That is when I noticed the problem described above.

Thanks for all the great features and hard work.


Then v0.90.9 was released:

madVR v0.90.9 released

http://madshi.net/madVR.zip

* slightly improved SSIM quality (even stronger detail reconstruction)
* removed SSimDetailed option, now SSimClean is always used
* removed "use 1.6 gamma instead of 2.2" option
* fixed: moving media player to another monitor resulted in paused playback
* fixed: rendering eventually left some D3D interfaces open
* fixed: madVR shutdown memory leak
* fixed: some problems with live TV (detected by DVBViewer devs)
* fixed: 1080p23 was forced even when not using "auto" (frame packing)

Not sure which one of these fixed my problem, but THANKS!

:thanks:

[jkauff]

Yes actually way more than other countries. German studios will go out of their way to even do a crap post conversion for some crap horror movie and sell it in 3D. Sadly, this could be also why people are disliking 3D.
In case you didn't know, nevcariel lives in Germany. ;)

[Devrim]

madVR v0.90.9 released

http://madshi.net/madVR.zip


* fixed: moving media player to another monitor resulted in paused playback



Thanks! :3

[KoD]

Hi madshi, it looks like vobsub rendering got broken again, but this time different than the last time. It's affecting 1920x1080 vobsubs made from BluRay sups, or the sups themselves. Small DVD vobsubs don't have this issue. The subtitles are visible when the player is in full screen mode on a normal 1080p screen, but they don't show up when the player is not in full screen mode (btw, I am not speaking about full screen exclusive mode). My guess is that the subs are not downscaled when the video is downscaled. EVR custom presenter renderer in MPC-HC in the same scenario works fine.

I put here (http://wikisend.com/download/498420/subs_issue.7z) a sub/idx sample, and my settings.bin file. You can use the subs with any media file, as long as it's longer than 5 minutes (the first spoken line starts ~ 5 minutes, as can be seen in the idx file).

Later edit: it happens the same when playing m2ts files from a BluRay, so it's not only vobsubs, but sups as well.

[XMonarchY]

Could someone compile a package of examples with visual differences between different madVR settings, including most chroma upscaling, image upscaling, image downscaling, image doubling + quadrupling, SuperRes, ER Diffusion and a ton of other settings on different kind of content, such as HQ 1080p, MQ 1080p, LQ 1080p, same for 720p and then for 4xxp... It would be a pretty comprehensive and LARGE package of examples, but shouldn't there be one just like that?

On what page should I start looking for SSIM downscaling examples to see difference between Catmull-Rom and SSIM?

I use these settings for SD and HD content (480p, 720p and 1080p) on my 1080p display with MPC-HC + LAV Filters (DXVA 2 Copy-Back Hardware Acceleration):
- 3DLUT
- 10bit 0-255 output on 12bit output display
- Artifact Removal - Low (top setting), Medium (bottom setting), but depends on content quality
- Chroma Upscaling - NNEDI3 64n + SuperRes (Strength 3)
- Image Downscaling - Catmull-Rom + AR + LL
- Image Doubling and Quadrupling (Luma only) set to Always
- Image Upscaling - Jink + AR
- Upscaling Refinement - SuperRes 3
- Dithering - Error Diffusion Type 2 + Colored Noise and Change Dither every frame
- D3D11 Exclusive Mode (3 frames in advance)
- Everything disabled in Trade Quality for Performance

These are obviously GPU-hungry settings for 480p, 720p, and 1080p content even for my 3770K @ 4.8Ghz and GTX 980 @ 1550Mhz. What settings do you recommend to try for:
1. 480p, 720p, and 1080p content upscaled to 4K on 4K display?
2. 4K content @ 4K on 4K display?

Obviously I want the best image quality and no lost or dropped frames and no presentation glitches.

[Patrik G]

madvr crashes with MPC-HC 64bit version but not with 32bit
why is that?

and only when i enter fullscreen not in window mode

crash report

1807 pages of issues and you still isnt able to have a fully working video renderer lol

[JarrettH]

Is SSIM okay to use on a doubled image (super-xbr)? I'm a little lost as to recommendations lately. I use super-xbr 100 doubling + catmull AR + LL. I'd like to improve on that.

I could do Jinc + AR and SR3 or super-xbr 100 + SSIM downscaling. Would either be more beneficial than what I'm running now? I'm concerned that going back to Jinc would cause more aliased lines, which super-xbr doubling helped with.

[baii]

Could someone compile a package of examples with visual differences between different madVR settings, including most chroma upscaling, image upscaling, image downscaling, image doubling + quadrupling, SuperRes, ER Diffusion and a ton of other settings on different kind of content, such as HQ 1080p, MQ 1080p, LQ 1080p, same for 720p and then for 4xxp... It would be a pretty comprehensive and LARGE package of examples, but shouldn't there be one just like that?

On what page should I start looking for SSIM downscaling examples to see difference between Catmull-Rom and SSIM?

I use these settings for SD and HD content (480p, 720p and 1080p) on my 1080p display with MPC-HC + LAV Filters (DXVA 2 Copy-Back Hardware Acceleration):
- 3DLUT
- 10bit 0-255 output on 12bit output display
- Artifact Removal - Low (top setting), Medium (bottom setting), but depends on content quality
- Chroma Upscaling - NNEDI3 64n + SuperRes (Strength 3)
- Image Downscaling - Catmull-Rom + AR + LL
- Image Doubling and Quadrupling (Luma only) set to Always
- Image Upscaling - Jink + AR
- Upscaling Refinement - SuperRes 3
- Dithering - Error Diffusion Type 2 + Colored Noise and Change Dither every frame
- D3D11 Exclusive Mode (3 frames in advance)
- Everything disabled in Trade Quality for Performance

These are obviously GPU-hungry settings for 480p, 720p, and 1080p content even for my 3770K @ 4.8Ghz and GTX 980 @ 1550Mhz. What settings do you recommend to try for:
1. 480p, 720p, and 1080p content upscaled to 4K on 4K display?
2. 4K content @ 4K on 4K display?

Obviously I want the best image quality and no lost or dropped frames and no presentation glitches.
The most expensive transformation right now with current popular contents should be 1080p upscale to 4k. For 4k content, unless you go crazy on nnedi3 chroma or image enhancement, it shouldn't be that much slower than 1080p to 1080p.

For 1080p24 to 4k, with current best GPU, usually it is nnedi3 32/64 + SR + maybe one upscaling refinement, if you opt for jinc or sxbr, you can usually play with all the toys freely like error diffusion, expensive chroma or multiple refinement.

Sent from my 306SH

[Ver Greeneyes]

Wanted to see how the latest version of SSIM compares to Catmull-Rom, which I was using before. Comparison (http://screenshotcomparison.com/comparison/161793). This is downscaling from 1080p to 720p, so not a very big factor and one I actually use a fair bit (both Catmull-Rom and SSIM set to use LL and AR). I have to say I'm impressed, it brings out a lot more detail!

[hu1kamania]

What happens if you try different "deinterlacing modes" in LAV? E.g. try "agressive" or something. Another test would be to try different deinterlacing settings in the NVidia GPU control panel. In any case, if EVR is showing the same problems that it's not really madVR specific, and there might not be much I can do about it.
Congrats on 0.90.9!

The sample, taken from live TV, appears to be a mixed content scenario. Some parts are telecined (Film mode is sublime in these instances). The other part is full 60i and will skip when set to Film mode because of the difference in refresh rate.

In MadVR I'll generally opt for Video mode, or Auto when viewing Mixed Content. In this instance, DXVA results in shimmering/aliasing and repeated frames during telecined portions of the video. Usually this is during commercials, but I've seen it happen when graphics appear during the primary broadcast!

Setting LAV "Deinterlacing Mode" to Auto, Aggressive, or Force results in visual shimmering and repeated frames. Setting LAV "Deinterlacing Mode" to Disabled alleviates much of the shimmering and results in 0 repeated frames.

Toggling "Use inverse telecine" in the Nvidia driver doesn't seem to make a difference, I thought this was for CUVID? I'm unaware of any other deinterlace settings in the Nvidia control panel. If anyone knows of another way to manipulate deinterlacing in the control panel or Nvidia Inspector, I'm all ears!

If I disable deinterlacing in MadVR and opt for a software method in LAV, I get double the repeated frames (still zero dropped frames) but shimmering is significantly reduced. Thus my preferred method for watching TV is to disable deinterlacing in MadVR and use one of the software methods from LAV Filters.

I'm unsure if this is an Nvidia driver issue, typical DXVA behavior, misconfiguration on my part, or if there is anything that can be done to manipulate DXVA on your end.

[YxP]

Could someone compile a package of examples

Maybe you could? Wait for 1.0 though. Sounds like a big job and possibly for nothing when settings could change overnight :p

[har3inger]

Oddly enough, latest build trips smartscreen for madhtctrl.exe. It never did before. I ignored/exception'd it and it works as expected.

[XMonarchY]

Wanted to see how the latest version of SSIM compares to Catmull-Rom, which I was using before. Comparison (http://screenshotcomparison.com/comparison/161793). This is downscaling from 1080p to 720p, so not a very big factor and one I actually use a fair bit (both Catmull-Rom and SSIM set to use LL and AR). I have to say I'm impressed, it brings out a lot more detail!

The overall conclusion is that Cat-Rom is better when downscaled to very low SD resolutions and SSIM is better in other cases is what my brief research suggests... When I watch 720p or even 480p content, I often use Luma Image Doubling and Quadrupling (NNEDI32 for both and both set to Always) on my 1080p display. That means that 720p content gets upscaled to 1440p if Doubling is used and to 2880p when Quadrupling is used. Then the image gets downscaled back to 1080p using Cat-Rom or SSIM, correct? In such cases, what would look better for film content - Cat-Rom or SSIM? Should SSIM be used with AR + LL like Cat-Rom for the best quality?

[har3inger]

Okay, I managed to find a good example of a scene where SuperRes introduces a bunch of aliasing. Or rather, what I should requalify as "blobbiness". It's not really aliasing at all, but an interruption in smooth lines.

http://screenshotcomparison.com/comparison/161795

Key areas of note, although the problem is very obvious in almost all parts of the screenshot:
The AOL billboard letters. Also the AOL man's border becomes non-uniform and rings a bit.
All the specular reflections off the cars, especially the thinly shaped ones. They become lumpy/rough. I don't really know how to better describe it.
The Boston Globe sign. Uniform white/grey becomes irregularly blobby white/grey.
Brown wooden fence around the parking lot. The whole thing becomes granular with SuperRes.

The source is not the best, admittedly, and could be entirely responsible for the lumpy/blobby/granular image that superres produces. This is with linear light off, but with it checked on, the strange lumpy quality remains the same.

[FreeFall]

I've found a bug in 0.89.13, so I hope this will be fixed in the next build. If not, please let me know.

Thanks madshi the new build fixed the problem.

Cheers.

[ryrynz]

Should SSIM be used with AR + LL like Cat-Rom for the best quality?

Can't comment on the LL side of things, but as for AR.

http://screenshotcomparison.com/comparison/161801

4K text rendering downscaled by SSIM with AR/NO AR magnified 1000% (pixel resize)

Results should should speak for themselves, the difference though would be basically imperceptible to anyone except for leeperry


When I use madVr(0.90.8) with PotPlayer x86(1.6.58613 - latest stable version) I get a memory leak.

Really pays to read previous posts, you may have just wasted a good deal of your own time, this was reported a couple of pages back.
0.90.9 was released 2 hours before you posted and has likely has fixed the issue.

[MrBonk]

Wanted to see how the latest version of SSIM compares to Catmull-Rom, which I was using before. Comparison (http://screenshotcomparison.com/comparison/161793). This is downscaling from 1080p to 720p, so not a very big factor and one I actually use a fair bit (both Catmull-Rom and SSIM set to use LL and AR). I have to say I'm impressed, it brings out a lot more detail!

More like it aggressively emphasizes aliasing.
Sorry,it just does not look good on real time rendered content that already doesn't have good AA before downscaling. It's going to bring back the aliasing that a good resampling filter should remove.
TW3's built in AA is poor, and 1080p to 720p is only a 2.25x factor so it's not going to look the best anyway.

Okay, I managed to find a good example of a scene where SuperRes introduces a bunch of aliasing. Or rather, what I should requalify as "blobbiness". It's not really aliasing at all, but an interruption in smooth lines.

http://screenshotcomparison.com/comparison/161795

Key areas of note, although the problem is very obvious in almost all parts of the screenshot:
The AOL billboard letters. Also the AOL man's border becomes non-uniform and rings a bit.
All the specular reflections off the cars, especially the thinly shaped ones. They become lumpy/rough. I don't really know how to better describe it.
The Boston Globe sign. Uniform white/grey becomes irregularly blobby white/grey.
Brown wooden fence around the parking lot. The whole thing becomes granular with SuperRes.

The source is not the best, admittedly, and could be entirely responsible for the lumpy/blobby/granular image that superres produces. This is with linear light off, but with it checked on, the strange lumpy quality remains the same.
I wouldn't totally call it Aliasing, though it is similar in some aspects.
Bad reconstruction because of a poor source file (Brings out the existing compression). I would have to imagine if the source was lossless it wouldn't happen. Having lossless SD content in a non-interlaced format is a pipe dream though.

[ryrynz]

Having lossless SD content in a non-interlaced format is a pipe dream though.

In this case I'd just use a luma scaler that wasn't as sharp. There's only so much you can do to SD content before it starts looking "off".

[har3inger]

More like it aggressively emphasizes aliasing.
Sorry,it just does not look good on real time rendered content that already doesn't have good AA before downscaling. It's going to bring back the aliasing that a good resampling filter should remove.
TW3's built in AA is poor, and 1080p to 720p is only a 2.25x factor so it's not going to look the best anyway.


I wouldn't totally call it Aliasing, though it is similar in some aspects.
Bad reconstruction because of a poor source file (Brings out the existing compression). I would have to imagine if the source was lossless it wouldn't happen. Having lossless SD content in a non-interlaced format is a pipe dream though.

TW3 has an in-engine sharpen filter that can look exactly like that image with SSM. In fact, the game is basically never that soft-looking/un-aliased at native res like it seems in catmull rom. This is especially true if there is also no AA being applied, which is likely. VSR or FSSSAA forced in drivers are the only real AA techniques that would look like the catmull rom image. tl'dr, because the source is probably super aliased, you can't fairly judge the merits of SSM.

And yeah, superres is likely screwing up the image because the source is bad, but even on pretty clean sources (2gb BR rips down to DVD resolution... though encoding settings and actual source may be much less good), I tend to see the same problems.

[Qotscha]

0.90.9 seems to have introduced a new SSIM related bug at least in my system (Win 10 x64, Radeon HD 6850, Catalyst 15.11.1). When video deinterlacing is on and downscaling factor is such that, instead of downscaling the whole image, only luma is downscaled, something strange happens (http://imgur.com/a/JTNPa).

[k4rma]

Hi, ive noticed an issue that has appeared since i upgraded to the 90.x version: for HEVC decodes when i go full screen and match the content frame rate (to 1080p23) the video starts stuttering. This does not occur when the video is in a window and desktop is at 50Hz, and does not occur for x264 content.

I have noticed it in x265 videos ive watched previously (before upgrade) that didnt have this problem, but now do.

Im using MPC-HC.

Has anyone else seen this type of issue?

[Ge'in]

0.90.9 seems to have introduced a new SSIM related bug at least in my system (Win 10 x64, Radeon HD 6850, Catalyst 15.11.1). When video deinterlacing is on and downscaling factor is such that, instead of downscaling the whole image, only luma is downscaled, something strange happens (http://imgur.com/a/JTNPa).

Same issue, with Win 7 x64, Radeon 7970 and Catalyst 16.1.
Video : 1440*1080 @ 25 fps with deint=ON
Downscale with SSIM : 1440*1080 to 1238*693 it's broken

http://reho.st/preview/self/5c0a714f63e625f1c6d094d1660e92760384499c.png (http://reho.st/view/self/5c0a714f63e625f1c6d094d1660e92760384499c.png)

[Ver Greeneyes]

madshi, bug report for you:

Using a 3DLUT, I'm getting some pretty bad clipping on particularly bright colors, and I think it's due to some precision loss in madVR. I say that because I turned off the 3DLUT, saved a screenshot, then loaded it up in MPC-HC and turned the 3DLUT back on, and the clipping is gone! I have all trade quality options disabled, though I did notice that enabling the "use 10bit" options shifts the clipping around slightly. Could you please take a look?

Here's an archive (https://mega.nz/#!OAcSwAZC!gikqrVHWF0vajYa4dFKcwsyZu1TZrmwt1kduHpvW0jE) containing the 3DLUT, a sample video, and several images to show what I'm seeing. With the 3DLUT enabled, I see clipping on the sky - the wrong colors look like the colors without the 3DLUT. I took a screenshot of frame 256 with ("video-frame-256-with-3DLUT.png") and without ("video-frame-256-no-3DLUT.png") the 3DLUT enabled to show the clipping. "saved-image-with-3DLUT.png" shows what I see if I open "video-frame-256-no-3DLUT.png" in MPC-HC with the 3DLUT enabled.

I'm testing this in windowed mode, but I see the clipping pretty much regardless of the settings. It's visible regardless of chroma doubling, superres, D3D11 versus D3D9, the native display bitdepth or trade quality options. My graphics card is an Nvidia GeForce GTX 970, with the 361.75 drivers, and I'm on 64-bit Windows 10 (using 64-bit MPC-HC and 64-bit madVR). I'm using the latest version of madVR, but I've been seeing this for a long time - it's just different every time I do a new calibration on my monitor.

[Telion]

Found it, will be fixed in the next build. Thanks for letting me know.
Unfortunately it's not completely fixed on my side. Now it leaks not on every file close - sometimes it doesn't leak when there are several files opened in a row, though more often it does than does not. And on the first file it seems to leak always.
I'm still on XP if that matters.

[madshi]

What do you think how material usually is downscaled?
My wild guess would be that with digital processing, downscaling in GL is most common. But I might be totally wrong.
Yes, I would agree that GL is probably most common. But it's hard to be sure.

When I use madVr(0.90.8) with PotPlayer x86(1.6.58613 - latest stable version) I get a memory leak.
Please try v0.90.9 or newer, should be fixed in there.

With 0.90.9 + Kodi Dsplayer i'm getting black screen when playing (sd, 720p or cropped 1080p) > UHD while 1080p & UHD > UHD is fine.
switching to D3D11 presentation fix the problem but unfortunetly my IGPU struggle with D3D11.
none of the above problems occures with MPC-HC X64.
X86 bug?
Is this a new problem with 0.90.9? If so, please try to find out which exact madVR version introduced it. You can download old versions from here:

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

madvr crashes with MPC-HC 64bit version but not with 32bit
why is that?

and only when i enter fullscreen not in window mode

crash report

1807 pages of issues and you still isnt able to have a fully working video renderer lol
Crash reports attached to this forum don't ever get approved, or at least very late. Try uploading them somewhere else.

The overall conclusion is that Cat-Rom is better when downscaled to very low SD resolutions and SSIM is better in other cases is what my brief research suggests...
I disagree.

Hi, ive noticed an issue that has appeared since i upgraded to the 90.x version: for HEVC decodes when i go full screen and match the content frame rate (to 1080p23) the video starts stuttering. This does not occur when the video is in a window and desktop is at 50Hz, and does not occur for x264 content.

I have noticed it in x265 videos ive watched previously (before upgrade) that didnt have this problem, but now do.

Im using MPC-HC.

Has anyone else seen this type of issue?
Can I see a screenshot with the Ctrl+J debug OSD turned on, please?

And can you please try to find out which exact madVR build introduced the problem? You can download old builds here:

The sample, taken from live TV, appears to be a mixed content scenario. Some parts are telecined (Film mode is sublime in these instances). The other part is full 60i and will skip when set to Film mode because of the difference in refresh rate.

In MadVR I'll generally opt for Video mode, or Auto when viewing Mixed Content. In this instance, DXVA results in shimmering/aliasing and repeated frames during telecined portions of the video. Usually this is during commercials, but I've seen it happen when graphics appear during the primary broadcast!

Setting LAV "Deinterlacing Mode" to Auto, Aggressive, or Force results in visual shimmering and repeated frames. Setting LAV "Deinterlacing Mode" to Disabled alleviates much of the shimmering and results in 0 repeated frames.

Toggling "Use inverse telecine" in the Nvidia driver doesn't seem to make a difference, I thought this was for CUVID? I'm unaware of any other deinterlace settings in the Nvidia control panel. If anyone knows of another way to manipulate deinterlacing in the control panel or Nvidia Inspector, I'm all ears!

If I disable deinterlacing in MadVR and opt for a software method in LAV, I get double the repeated frames (still zero dropped frames) but shimmering is significantly reduced. Thus my preferred method for watching TV is to disable deinterlacing in MadVR and use one of the software methods from LAV Filters.

I'm unsure if this is an Nvidia driver issue, typical DXVA behavior, misconfiguration on my part, or if there is anything that can be done to manipulate DXVA on your end.
Well, I don't know what else to suggest. Sounds all weird to me. But in the end, I don't really have much control over what DXVA does internally.

0.90.9 seems to have introduced a new SSIM related bug at least in my system (Win 10 x64, Radeon HD 6850, Catalyst 15.11.1). When video deinterlacing is on and downscaling factor is such that, instead of downscaling the whole image, only luma is downscaled, something strange happens (http://imgur.com/a/JTNPa).
Same issue, with Win 7 x64, Radeon 7970 and Catalyst 16.1.
Video : 1440*1080 @ 25 fps with deint=ON
Downscale with SSIM : 1440*1080 to 1238*693 it's broken

http://reho.st/preview/self/5c0a714f63e625f1c6d094d1660e92760384499c.png (http://reho.st/view/self/5c0a714f63e625f1c6d094d1660e92760384499c.png)
Should be fixed in the next build.

Unfortunately it's not completely fixed on my side. Now it leaks not on every file close - sometimes it doesn't leak when there are several files opened in a row, though more often it does than does not. And on the first file it seems to leak always.
I'm still on XP if that matters.
I was able to reproduce the leak pretty reliably, and now it's gone for me. If the problem is still there for you I need some more information. Ideally I'd need to be able to reproduce it somehow. If that doesn't work out, can you please check if the problem is really 100% not there for you with v0.90.3, while it's definitely there with v0.90.9 (just to be safe). And then please create a debug log with v0.90.9 which captures the leak. Thanks.

Hi madshi, it looks like vobsub rendering got broken again, but this time different than the last time. It's affecting 1920x1080 vobsubs made from BluRay sups, or the sups themselves. Small DVD vobsubs don't have this issue. The subtitles are visible when the player is in full screen mode on a normal 1080p screen, but they don't show up when the player is not in full screen mode (btw, I am not speaking about full screen exclusive mode). My guess is that the subs are not downscaled when the video is downscaled. EVR custom presenter renderer in MPC-HC in the same scenario works fine.

I put here (http://wikisend.com/download/498420/subs_issue.7z) a sub/idx sample, and my settings.bin file. You can use the subs with any media file, as long as it's longer than 5 minutes (the first spoken line starts ~ 5 minutes, as can be seen in the idx file).

Later edit: it happens the same when playing m2ts files from a BluRay, so it's not only vobsubs, but sups as well.
Seems to work fine here. But you've not provided a lot of information. E.g. which subtitle renderer are we talking about? If XySubFilter, is it the latest version or some old build? And which resolution does the video have you've tested this with?

madshi, bug report for you:

Using a 3DLUT, I'm getting some pretty bad clipping on particularly bright colors, and I think it's due to some precision loss in madVR. I say that because I turned off the 3DLUT, saved a screenshot, then loaded it up in MPC-HC and turned the 3DLUT back on, and the clipping is gone! I have all trade quality options disabled, though I did notice that enabling the "use 10bit" options shifts the clipping around slightly. Could you please take a look?

Here's an archive (https://mega.nz/#!OAcSwAZC!gikqrVHWF0vajYa4dFKcwsyZu1TZrmwt1kduHpvW0jE) containing the 3DLUT, a sample video, and several images to show what I'm seeing. With the 3DLUT enabled, I see clipping on the sky - the wrong colors look like the colors without the 3DLUT. I took a screenshot of frame 256 with ("video-frame-256-with-3DLUT.png") and without ("video-frame-256-no-3DLUT.png") the 3DLUT enabled to show the clipping. "saved-image-with-3DLUT.png" shows what I see if I open "video-frame-256-no-3DLUT.png" in MPC-HC with the 3DLUT enabled.

I'm testing this in windowed mode, but I see the clipping pretty much regardless of the settings. It's visible regardless of chroma doubling, superres, D3D11 versus D3D9, the native display bitdepth or trade quality options. My graphics card is an Nvidia GeForce GTX 970, with the 361.75 drivers, and I'm on 64-bit Windows 10 (using 64-bit MPC-HC and 64-bit madVR). I'm using the latest version of madVR, but I've been seeing this for a long time - it's just different every time I do a new calibration on my monitor.
Looks like your 3dlut handles BTB/WTW values badly. That sort of data is clipped when you make an RGB screenshot (which is stored in PC levels!).

Try adding this custom shader:

sampler s0 : register(s0);

float4 main(float2 tex : TEXCOORD0) : COLOR
{
return saturate(tex2D(s0, tex));
}
I'm not sure if this video file is really TV levels, btw. It might be PC levels. Try adding "levels=full" as a file name tag.

Wanted to see how the latest version of SSIM compares to Catmull-Rom, which I was using before. Comparison (http://screenshotcomparison.com/comparison/161793). This is downscaling from 1080p to 720p, so not a very big factor and one I actually use a fair bit (both Catmull-Rom and SSIM set to use LL and AR). I have to say I'm impressed, it brings out a lot more detail!
Would be interesting to see the original frame, because only that we we could judge which downscale is a better representation of the original frame.

Okay, I managed to find a good example of a scene where SuperRes introduces a bunch of aliasing. Or rather, what I should requalify as "blobbiness". It's not really aliasing at all, but an interruption in smooth lines.

http://screenshotcomparison.com/comparison/161795
Unfortunately screenshotcomparison doesn't seem to work atm. Will try to remember to check this out later...

[sat4all]

Is this a new problem with 0.90.9? If so, please try to find out which exact madVR version introduced it. You can download old versions from here:

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


Yes it's a 0.90.9 problem, all previous versions work ok.
:thanks:

[madshi]

Thanks for the new release.
With 0.90.9 + Kodi Dsplayer i'm getting black screen when playing (sd, 720p or cropped 1080p) > UHD while 1080p & UHD > UHD is fine.
switching to D3D11 presentation fix the problem but unfortunetly my IGPU struggle with D3D11.
none of the above problems occures with MPC-HC X64.
X86 bug?
Does the Ctrl+J OSD work? Do you see anything on screen? Is your TV even able to sync to the GPU? Have you tried resetting the madVR settings as a test?

[Ver Greeneyes]

Looks like your 3dlut handles BTB/WTW values badly. That sort of data is clipped when you make an RGB screenshot (which is stored in PC levels!).

Try adding this custom shader:Hey, that works! I added it as a pre-resize shader, is that right? It doesn't seem to matter for the result. The 3DLUT file was just produced by ArgyllCMS's collink, and I don't think that offers any way to clip BTB/WTW. I can report it as an ArgyllCMS bug if you'd like. Does 'badly' here mean that the 3DLUT includes output values outside the TV level range, or that it does weird things if given an input value outside the TV level range?

I'm not sure if this video file is really TV levels, btw. It might be PC levels. Try adding "levels=full" as a file name tag.
I'm pretty sure the stream layout looks right by default - with levels=full it looks grey. Maybe the encode is sloppy in some way (this was a livestream).

Would be interesting to see the original frame, because only that we we could judge which downscale is a better representation of the original frame.True, I tried to upload it but kept running into size caps and gave up. I can try compressing it a little more (or putting it on Mediafire / Mega) if you'd still like to see it.

[sat4all]

Does the Ctrl+J OSD work? Do you see anything on screen? Is your TV even able to sync to the GPU? Have you tried resetting the madVR settings as a test?

no, nothing at all. but i can hear audio of the playing files.
yes tv do sync with the gpu.
tried settings reset and it didn't help.
one more finding: disabling FSE fix the problem.

[madshi]

The 3DLUT file was just produced by ArgyllCMS's collink, and I don't think that offers any way to clip BTB/WTW. I can report it as an ArgyllCMS bug if you'd like. Does 'badly' here mean that the 3DLUT includes output values outside the TV level range, or that it does weird things if given an input value outside the TV level range?
I don't know. I'm not really an expert in 3dlut matters. Might make sense to ask that question in the AVSForum madVR/ArgyllCMS thread.

no, nothing at all. but i can hear audio of the playing files.
yes tv do sync with the gpu.
tried settings reset and it didn't help.
one more finding: disabling FSE fix the problem.
Does the media player still react to mouse and keyboard events? Then a debug log may help. If not, a freeze report might help.

[Ver Greeneyes]

I don't know. I'm not really an expert in 3dlut matters. Might make sense to ask that question in the AVSForum madVR/ArgyllCMS thread.Okay, I applied a quick dirty fix to the collink source. I'll see if it works, then post to the ArgyllCMS mailing list or something (my fix is very ugly, but also very simple, so it shouldn't be hard for Graeme to address if my fix works).

[KoD]

Seems to work fine here. But you've not provided a lot of information. E.g. which subtitle renderer are we talking about? If XySubFilter, is it the latest version or some old build? And which resolution does the video have you've tested this with?

It happens with 1080p videos, displayed at less than 1920 x 1080 resolution, as long as the resolution of the vobsub or the sup is also large (1920 x 1080). The subtitle renderer is xysubfilter 3.1.0.746 (the latest, as far as I know). Happens both in 32 and 64 bit.

Here's another sample (http://wikisend.com/download/509278/pgs_sample.mkv) [38 Mb], a mux directly from a BluRay, with only the video and the PGS subs.

Player in full-screen mode:
http://s29.postimg.org/mgh3ze9tv/shot1.png (http://postimg.org/image/mgh3ze9tv/) http://s29.postimg.org/4ce3erc5f/shot3.png (http://postimg.org/image/4ce3erc5f/)

Player not in full-screen mode:
http://s29.postimg.org/5tzjqbgw3/shot2.png (http://postimg.org/image/5tzjqbgw3/) http://s29.postimg.org/wb895mds3/shot4.png (http://postimg.org/image/wb895mds3/)

[baii]

A little comparison for those interested in hardware, I recently acquired 980ti and a full unlocked fury(think furyx). They fury(x) is slightly faster(about 1ms render time, avg ~14 vs ~15ms) for one of my 1080i60 clip upscale to 4k with following setting.
980ti 1460hz
Fury 1060hz
3dlut , banding on, order dithering
Chroma : lanczo ar
Luma nnedi3 32
Chroma doubling ( image upscaling used when nnedi3 chroma doubling is not used)
Lanczo ar
Upscale refinement, SR str1.

Realistically, the performance difference won't matter for 24 or 30fps content, but 60fps content is really tight on performance.

If anyone want test other comparison, please let me know, before selling off one of the cards.


Sent from my SM-T700 using Tapatalk

[Ver Greeneyes]

Actually, is a file format specification of madVR's 3DLUT files available anywhere? I'm a bit confused by some of the things in the collink source (in particular, are output values supposed to be stored in the range 0x0000 - 0xff00? I would expect a range of 0x1010 - 0xebeb for video levels, but that probably just shows my ignorance).

[JarrettH]

Anyone have an image preference? 720p to 1080p

jinc, SR3
https://c2.staticflickr.com/2/1578/24660833309_4aab322d61_o.png

sxbr100, catmull
https://c2.staticflickr.com/2/1719/25002112806_ef924442f2_o.png

jinc, SR3
https://c2.staticflickr.com/2/1633/24401620403_8600faff57_o.png

sxbr100, catmull
https://c2.staticflickr.com/2/1598/24660835439_28ac622d4d_o.png

Trying to decide if using SR3 is an objective improvement.

[sat4all]

Does the media player still react to mouse and keyboard events? Then a debug log may help. If not, a freeze report might help.

After uninstall/reinstall kodi i can't reproduce anymore.:stupid:
Thanks alot for your precious time.
Btw i've decided to jump for kodi + mpc-hcfr.x64 as external player instead of dsplayer. Madvr x64 is a huge boost, cropped 1080p to uhd was impossible for me without ticking "use 10bit chroma/image buffer instead of 16bit" but thankfully, with x64 chain i can do it without ticking a single thing in trade quality for performance.

[Kevin Gilmore]

MadShi,

Please advise: wondering about cpu-load<PCI-e bus load from the renderer through MadVR

or weather any A-situational processing occurs that you/we know of where lots of CPU-GPU copy-back or
CPU post processing occurs.. that may or may not load the pci-bus moreso than otherwise.

Basically, is it a viable or good idea? Particularly @ higher desktop class MadVR options/settings...
Thunderbolt 3/usbC laptop now to.. tap the e-desktop gpu for its video decoders package/8k video output, etc..

The 8k teli could process the video signal [120/240-etc interpolation] from there.

https://www.techpowerup.com/reviews/NVIDIA/GTX_980_PCI-Express_Scaling/22.html
[gtx980] Fairly recent pci-article if somehow relevant
a) This is of course presuming intermediary dock, decode/recode or Laptop-Thunderbolt III chipset are not ever an issue-they do not ever throttle.
1a) Though of course there could be slight delay/input lag like this e-gpu rep mentions here (https://www.youtube.com/watch?v=rs_Ky4yCz0k).

but I'd like your take on how MadVR coterminates with the CPU or loads the PCI-ebus through some of its settings..
If you'd care to provide a thoughtful answer I'll probably go for it and report what I find when new GPUs' are out..
Thanks MadShi

[aufkrawall]

madVR shouldn't copy back anything via PCIe, only LAV should do this with DXVA2 copyback mode.

[baii]

I thought it was said that pcie speed may make a difference for really extreme nnedi3 ?

Sent from my SM-T700 using Tapatalk

[Kevin Gilmore]

Aufkrawall, Baii, thanks for responding.
Are there any overt or outstanding bus-hogs in typical MadVR use? Or is running MadVR at full-tilt quite
manageable like any/many other a program on a somewhat lesser bus?

Under some active MadVR capabilities, must the CPU liaise multiple, or compounding data,
or bi-directional communication (I guess this would be post processing -- which heavily loads PCI-bus)? Than say, a typical Video-game render pipeline will have to, or requires?

..Absolutely no idea weather the GPU-Z-bus sensor is accurate though.
If the bus ceiling were met the result is a minimum fps drop, or microstutter?

This would be a good thing to know if we do not already have an idea :)

[kasper93]

When NNEDI3 quadruple is in use, OSD reports that luma is downscale with Bicubic150 AR instead of SSIM AR. Any reason for that?

[raymondjpg]

Is there a way to change the active device in madVR settings?

[ryrynz]

When NNEDI3 quadruple is in use, OSD reports that luma is downscale with Bicubic150 AR instead of SSIM AR. Any reason for that?
I found when I upgraded to. 9 after. 8 it defaulted to bicubic when I had SSIM detailed set in .8 what with it being deprecated and all. I actually tested quadrupling with SSIM the night before and the osd was correct for me so perhaps just change your Luma Scaler to something that's not SSIM and back again and that might fix it.

[Telion]

I was able to reproduce the leak pretty reliably, and now it's gone for me. If the problem is still there for you I need some more information. Ideally I'd need to be able to reproduce it somehow. If that doesn't work out, can you please check if the problem is really 100% not there for you with v0.90.3, while it's definitely there with v0.90.9 (just to be safe). And then please create a debug log with v0.90.9 which captures the leak.
http://hugefiles.net/4eeyrwm5p5xz
I've found out that if I stop the playback before closing the file or opening another, the leak is never there (and it is so in both v0.90.4 and v0.90.9 builds). But if I do this while playing or on pause - sometimes it leaks, sometimes not. It seems to be completely random - I haven't been able to identify any pattern or dependencies. Also, the "stop" case doesn't depend on the presence of "DontRenderAfterStop" file.
So I created two debug logs - one with stop&close case (so no leak) and the other with immediate close after playback started (and with a leak of course). I have a rather slow rig, so maybe that matters and it is some sort of race condition. But definitely there were no such problem with v0.90.3 and before.

[baii]

Aufkrawall, Baii, thanks for responding.
Are there any overt or outstanding bus-hogs in typical MadVR use? Or is/are the MadVR protocol quite manageable like any/many other a program on a somewhat lesser bus?

This would be a good things to know anyway, if we do not already have an idea :)

Never heard anyone having problem with pcie bandwidth. The nnedi3 thing was a artificial copying benchmark or w.e and never had been seen in normal use.
And if you are interested in egpu, techinferno is the go to place. They may have more knowledge on scalling or actual madvr use on egpu.
I don't think there would be much problem if it reorganized as native card.

Sent from my SM-T700 using Tapatalk