madVR - high quality video renderer (GPU assisted)

[Siso]

Crashing on exit also occurs is MPC-BE.1.4.5.364.x86

+1 and also use windowed overlay doesn't work when selected...

[luk008]

Which OS are you using?

I'm using Windows 8.1 x64
My GPU is a R9 270.

[madshi]

I'm using Windows 8.1 x64
My GPU is a R9 270.
In which way does it look bad in windowed mode? A little screenshot might help.

[Moony349]

Crashing is fixed; however, as you mentioned, presentation issues remain on Windows 7.

The render and present queues are 0 or 1-3 and 1-2 (with dx 11 enabled)

EDIT: Overlay seems to work better; no presentation glitches and 7-8 / 16-16.

[SecurityBunny]

If I use 10bit, in windowed mode, image is very dark. In fullscreen mode, it's like as 8 bit.

+1, also experiencing the image being dark with 10 bit output. Going from 10 bit to 9 bit to 8 bit, the image gradually gets lighter back to normal colors. Though the image seems to be dark for me in both windowed mode, fullscreen windowed, and exclusive fullscreen.

MadVR 0.88.1 x64
MPC-HC 1.7.8.162 x64

[Devrim]

Using Direct3D 11 mode crashes MPC-HC when switching the video window from 1 screen to another.

-edit-

I also get stuttering and crashing in non Direct3D 11 mode with switching a video from screen to screen.

With the previous version it was all very smooth.

-edit2-

This has been fixed with v0.88.1

[ikarad]

Fullscreen exclusive mode or fullscreen windowed mode?



Fullscreen exclusive mode

In fullscreen not exclusive mode, no presentation glitch.

[luk008]

In which way does it look bad in windowed mode? A little screenshot might help.

fullscreen exclusive -> http://i.imgur.com/SluCz7U.png
windowed -> http://i.imgur.com/3oCCUzE.png

[luk008]

I uninstalled madvr0.88.0 and I reinstalled 0.88.0. In 10bit ouput image is dark in fullscrren mode and windowed mode. Not only in windowed mode.
It's strange.

Try fullscreen exclusive. It's the only way that 10 bit looks fine here.

[ikarad]

Try fullscreen exclusive. It's the only way that 10 bit looks fine here.
I have always fullscreen exclusive

Summary (I deleted my previous message)
With D3d9 path, In 10bit ouput image is dark in fullscreen exclusive mode and windowed mode. Not only in windowed mode.


With d3d11 path, in 10 bit output image is dark in windowed mode and not dark in fullscreen exclusive mode.

[Devrim]

Does madVR test patterns work with 64bit media players?

[MS-DOS]

Seems to me the presentation glitch problems with D3D11 are probably all caused by Windows 7 behaving differently than Windows 8. Not totally sure, though. On my Windows 8.1 x64 PC everything runs well.

Win 7 64-bit here. BTW, it doesn't report D3D11 in FSE, just "fullscreen exclusive mode, 8 bit". Is that intended? All queues are full in FSE.

Render queue is stuck at 3-4\8 using D3D11 and presenting 4 frames in advance in windowed mode. Increasing presented frames changes nothing, while setting them to 1 fills up the render queue to its maximum and... causes hundreds of presentations glitches :D

Down to 2-3\8 with SM enabled. Well, at least it doesn't cause any problems for now.

[aufkrawall]

Exciting changes, thanks a lot, madshi!

With new DX11 renderer, I'm sometimes getting tearing after seeking in fullscreen new windowed mode (1080p -> 1440p Jinc3 AR, 60fps@75Hz SM on).
Anyone confirm? I'm on Windows 10 10074 with GTX 980 352.63 driver.
Old D3D9 renderer works fine. Haven't tested 10 bit and probably won't do it since my display doesn't support it and I neither like FSE.

And I really like many options to choose from. :D

Edit: Btw: Have you taken a look at the new OpenCL 1.2 drivers of Nvidia? Maybe native DXVA2 now works without quality loss?

[huhn]

Exciting changes, thanks a lot, madshi!

With new DX11 renderer, I'm sometimes getting tearing after seeking in fullscreen new windowed mode (1080p -> 1440p Jinc3 AR, 60fps@75Hz SM on).
Anyone confirm? I'm on Windows 10 10074 with GTX 980 352.63 driver.
Old D3D9 renderer works fine. Haven't tested 10 bit and probably won't do it since my display doesn't support it and I neither like FSE.

And I really like many options to choose from. :D

Edit: Btw: Have you taken a look at the new OpenCL 1.2 drivers of Nvidia? Maybe native DXVA2 now works without quality loss?

i see the tearing too after pause unpause.

just hammer space very easy to reproduce. but such a small issue.

was has DXVA2 to do with openCL?

[aufkrawall]

With Intel & AMD, OpenCL can be used to prevent a lossy conversion. Otherwise, a copyback is needed.

[huhn]

about the darker screen in window d11 mode.

did the rest only get this with error diffusion too?

[SecurityBunny]

about the darker screen in window d11 mode.

did the rest only get this with error diffusion too?

Nice catch! I can confirm switching from ED1 (luma, static) to ordered dithering, the darkening from 9/10 bit goes away. Both with the d3d11 presentation checked and unchecked.

[huhn]

is it intentional that upscailing refinement doesn't have a restore defaults button unlike image enhancements?

edit:

finesharp mode 2 and 3 is destroying the image by adding a ton of aliasing and worse things without any sharpening. activing linear light counters this effect.
the rest of the finesharp settings are default.

example:
no finesharp http://abload.de/img/finesharpofffyp5y.png
finesharp mode 1 fine looks nearly original http://abload.de/img/finesharpmode1t9r0r.png
finesharp mode 3 broken http://abload.de/img/finesharpmode3d5oe8.png

[Schwartz]

I noticed that using 64-bit madVR with the appropriate MPC-HC and internal LAV Filters, NNEDI3 upscaling is not working at all. No matter which sub-settings I pick. Milisecond count however (rarely) spikes up in the renderer as if it were active. Graphics card is a R9 270X.
To expand upon that thing I noticed, where render ms appear unusually high or inconsistent with x64 madVR.. that didn't have anything to do with NNEDI3 working or not, but rather something else:

Watching an MPEG-4 video that MPC-HC decodes through software avcodec (using only Spline36/Spline36 for upscaling, Error Diffusion 1, Smooth Motion, all quality tradeoffs disabled), I noticed that CPU usage will rise slowly from 7% to 15% and render ms will likewise go from ~9ms up to ~15ms, which is much more than I've ever seen on a profile that didn't use Image Doubling. I can play 1080p content with Spline36 and not even reach ~9ms, but this 640x480 SD content is bugging out on me. It appears to be less pronounced when watched at 100% zoom, but maximizing the window will produce these effects. The ms increase is not sudden either, but will slowly climb up higher and higher. When the window is maxed and the upscaler kicks in, the values may stay low for as long as 10-20 seconds, suggesting that these extra CPU cycles and render ms aren't actually required to play the movie but come from somewhere else.

This could just as well be a x64 LAV issue I assume, I just don't know.

This still happens in 0.88.1.

[Razoola]

Am I the only one with an issue in the new versions with the exclusive mode seek bar not working correctly. I cannot seek using the mouse and sometimes the seekbar no longer appears when I move the mouse to the bottom of the screen.

[vivan]

I would like to start with evaluating the new "DEBAND" algorithm written by Shiandow. Please could you guys try it with your various sources and let me know how it compares to madVR's other debanding algorithms in terms of quality and performance? Based on your feedback I will either remove the new algorithm again, or I will replace one or multiple of the old options with the new one. I want to end up with only one Deband option with max 3 different settings. Your feedback is very welcome - thanks!1) It's twice slower than "high" preset (or any other without gradient thing). On laptop with Intel HD 3000 I could use old debanding on 1080p video (downscaled to 768p with CR) but not new one (no/old/new - 30/38/46 ms).
2) I can't get better results on anime sources (destroyed by p̶r̶o̶f̶e̶s̶s̶i̶o̶n̶a̶l̶ ̶e̶n̶c̶o̶d̶e̶r̶s Crunchyroll) like this ones: 1 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/akame.y4m), 2 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/terraformars.y4m), 3 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/wixoss.y4m).
Threshold >= 1.0 causes nasty artifacts near lines (on akame even 0.6 is enough to get them) and with detail level set to 1 there's still more banding.

[flashmozzg]

Can confirm that with d3d11 I have more presentation glitches than with dx9.
Although rend./present. times seem to be a tad bit smaller.

[yok833]

Am I the only one with an issue in the new versions with the exclusive mode seek bar not working correctly. I cannot seek using the mouse and sometimes the seekbar no longer appears when I move the mouse to the bottom of the screen.

Nope.. Same thing for me too... Actually seeking with the mouse is now stopping the movies with "pause" on the top left

[Razoola]

Nope.. Same thing for me too... Actually seeking with the mouse is now stopping the movies with "pause" on the top left

Yes I have that also.

[ADude]

5) Please note that currently there are tons and TONS of options. This definitely is overkill and needs to be cleaned. This is a task we have to work on together. For each algorithm I'd like to define max 3 options (e.g. low, mid, high) and that must be sufficient. Otherwise the sheer amount of options will shy new users away.

Actually, I would suggest using the popular Basic/Advanced options paradigm. In other words, have few options shown by default ("Basic") and then have an Advanced button to click to show all options.

[huhn]

i have a huge issue with d11 fullscreen mode.
it is switching to 59p in fullscreen with a 60p source. the composition rate is still shown with 60 hz and window mode is using the correct 60p mode.

same issue when i play a 24p source. full screen uses 23p not 24p and i see a ton of multi frame tearing.
no issue with 23p source and 59p sources i haven't test anything else.

[madshi]

With new DX11 renderer, I'm sometimes getting tearing after seeking in fullscreen new windowed mode (1080p -> 1440p Jinc3 AR, 60fps@75Hz SM on).
i see the tearing too after pause unpause.

just hammer space very easy to reproduce. but such a small issue.
There isn't much I can do about it. In windowed mode Aero/DWM is supposed to take care of tearing. If it doesn't, it's on OS/GPU bug.

Edit: Btw: Have you taken a look at the new OpenCL 1.2 drivers of Nvidia?
No.

about the darker screen in window d11 mode.

did the rest only get this with error diffusion too?
Ah, error diffusion, yeah...

is it intentional that upscailing refinement doesn't have a restore defaults button unlike image enhancements?
Yes, no space for such a button there. You can see the defaults in the image enhancements page and copy them from there manually, if you want to go back to the defaults.

finesharp mode 2 and 3 is destroying the image by adding a ton of aliasing and worse things without any sharpening.
Confirmed.

Am I the only one with an issue in the new versions with the exclusive mode seek bar not working correctly. I cannot seek using the mouse and sometimes the seekbar no longer appears when I move the mouse to the bottom of the screen.
Nope.. Same thing for me too... Actually seeking with the mouse is now stopping the movies with "pause" on the top left
I don't have that here. Which OS? Do you only have that in D3D11 mode, or also with D3D9?

1) It's twice slower than "high" preset (or any other without gradient thing). On laptop with Intel HD 3000 I could use old debanding on 1080p video (downscaled to 768p with CR) but not new one (no/old/new - 30/38/46 ms).
2) I can't get better results on anime sources (destroyed by p̶r̶o̶f̶e̶s̶s̶i̶o̶n̶a̶l̶ ̶e̶n̶c̶o̶d̶e̶r̶s Crunchyroll) like this ones: 1 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/akame.y4m), 2 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/terraformars.y4m), 3 (https://dl.dropboxusercontent.com/u/16254258/android/viavr/rendering/debanding/wixoss.y4m).
Threshold >= 1.0 causes nasty artifacts near lines (on akame even 0.6 is enough to get them) and with detail level set to 1 there's still more banding.
Thanks. So you prefer the "old" algorithm?

Can confirm that with d3d11 I have more presentation glitches than with dx9.
Which OS?

[madshi]

i have a huge issue with d11 fullscreen mode.
it is switching to 59p in fullscreen with a 60p source. the composition rate is still shown with 60 hz and window mode is using the correct 60p mode.

same issue when i play a 24p source. full screen uses 23p not 24p and i see a ton of multi frame tearing.
no issue with 23p source and 59p sources i haven't test anything else.
Is this fullscreen windowed or fullscreen exclusive mode? If you're talking about FSE, is there any difference depending on whether you switch your display to 8bit or 10bit?

Does madVR test patterns work with 64bit media players?
No, not yet.

Win 7 64-bit here. BTW, it doesn't report D3D11 in FSE, just "fullscreen exclusive mode, 8 bit". Is that intended?
Yes. Had "D3D11" in there, it made the OSD so big.

[nevcairiel]

I get loads of presentation glitches in D3D11 FSE mode, on Windows 8.1, NVIDIA Titan X (350.12). On a multi-monitor setup if that maybe matters.
The number sky-rockets, almost like every frame (or every second) produces a glitch counter.

Seems fine in windowed D3D11, or D3D9 FSE.

[JPulowski]

Hi madshi,

Thanks for including LumaSharpen. I did some testing and here is some feedback:

Looks like currently you are using the old version of LumaSharpen, which is 1.4.1. Current version is 1.5.0, which is included with Reshade + SweetFX bundle. I have already modified it for MPC-HC use, you can find it here (http://pastebin.com/rxDjqmGg).
Current madVR implementation uses Pattern 2 (9 tap gaussian, 4+1 texture fetches), although it looks nice on video games it doesn't look that good when it comes to videos (http://forum.doom9.org/showthread.php?p=1673897#post1673897). In my opinion, it would be better if you'd include the other patterns as options.
This is more of a personal request: I am normally used to MPC-HC's shader management. For instance I do always apply LumaSharpen, no matter if there is resizing or not. With madVR even though I can use profiles it gets complicated in this case. Because sometimes I watch videos downsized in windowed mode, which LumaSharpen does not apply unless it is pre-resize. And pre-resize and post-resize(upscale) applications of LumaSharpen is quite different from each other and requires additional fiddling. It would be nice if there would be an option to apply LumaSharpen always as post-resize.

[huhn]

Is this fullscreen windowed or fullscreen exclusive mode? If you're talking about FSE, is there any difference depending on whether you switch your display to 8bit or 10bit

yes d11 FSE. and i tried both 8 and 10 bit both have this issue.
not sure if it matters at all i'm on windows 10 x64 10074.

[flashmozzg]

There isn't much I can do about it.
Which OS?
W8.1 x64
It wasn't that bad (by the time I finished watching ~20min movie I had like 14-15 presentation glitches. Usually it's around 2-4). Could be because I stiched in and out of FSE more than usual.

[Telion]

Actually, I would suggest using the popular Basic/Advanced options paradigm. In other words, have few options shown by default ("Basic") and then have an Advanced button to click to show all options.

Second this, as madVR's flexibility is one of its main advantages. The ability to fine-tune almost everything is crucial for such a feature-rich renderer that is used in many different environments and scenarios. Besides this Basic/Advanced paradigm, I would suggest to add explanation tooltips for some options - that would definitely make things much less confusing.

[huhn]

Hi madshi,

Thanks for including LumaSharpen. I did some testing and here is some feedback:

Looks like currently you are using the old version of LumaSharpen, which is 1.4.1. Current version is 1.5.0, which is included with Reshade + SweetFX bundle. I have already modified it for MPC-HC use, you can find it here (http://pastebin.com/rxDjqmGg).
Current madVR implementation uses Pattern 2 (9 tap gaussian, 4+1 texture fetches), although it looks nice on video games it doesn't look that good when it comes to videos (http://forum.doom9.org/showthread.php?p=1673897#post1673897). In my opinion, it would be better if you'd include the other patterns as options.
This is more of a personal request: I am normally used to MPC-HC's shader management. For instance I do always apply LumaSharpen, no matter if there is resizing or not. With madVR even though I can use profiles it gets complicated in this case. Because sometimes I watch videos downsized in windowed mode, which LumaSharpen does not apply unless it is pre-resize. And pre-resize and post-resize(upscale) applications of LumaSharpen is quite different from each other and requires additional fiddling. It would be nice if there would be an option to apply LumaSharpen always as post-resize.

isn't "refine the image only once after upscaling is complete" post resize?

[MS-DOS]

Enabling windowed overlay or switching to old path for windowed\FSE modes doesn't seem to do anything when D3D11 is activated. Apparently it's as it should be, but still a bit awkward.

Disabling D3D11 while windowed is set to old path causes the video to stuck in a few frames loop cycle.

Also, just had a video playing on the background in D3D11 mode while browsing and it hanged after a few seconds. Pausing\unpausing resumed the sound, but the picture remained frozen.

[luk008]

I'm testing the Shiandow's debanding algorithm against the "old" reduce banding artifacts.
I can't give a veredict, but Shiandow's seems to reduce the banding artifacts a little better than the old algorithm.
In the efficiency side, Shiandow's is 1-2ms slower here (r9 270 + i3-3250).
I've also used crunchyroll anime to compare.
I'll select some comparison screenshots to post later.

[Schwartz]

Pausing an XVID on DX11 made the video repeat a fraction of a second over and over again like one of those old VCRs. Resuming play didn't work either.

[Devrim]

It would be nice if we could manually select frames in MPC-HC so we could make decent comparisons. (Like in avspmod) Anyone who knows a way to do this?

If it would be possible we could makes comparisons, upload it to http://screenshotcomparison.com/

[nevcairiel]

It would be nice if we could manually select frames in MPC-HC so we could make decent comparisons. (Like in avspmod) Anyone who knows a way to do this?

Navigation -> Go to
Remember to pause before.

[huhn]

It would be nice if we could manually select frames in MPC-HC so we could make decent comparisons. (Like in avspmod) Anyone who knows a way to do this?

If it would be possible we could makes comparisons, upload it to http://screenshotcomparison.com/

load on PNG or use control + g and copy paste the frame.

[Ver Greeneyes]

Anyone who knows a way to do this?Yep, open the Navigate menu and select Go To..., then use the bottom field to select the desired frame :)

[Devrim]

Navigation -> Go to
Remember to pause before.

Thanks found it

When I got time I will make some comparisons.
Is it better to use grainy videos or better to use clean videos?

[Morku]

I enabled D3D11 mode and get presentation glitches and dropped frames after these steps:

I switch from window to fullscreen exclusiv mode. Now when I jump in movie (by clicking on the bottom line) = that problem.

When I am in window mode and jump in movie = no problems. When I switch to fullscreen exclusive mode and don't jump in that mode, also no problems.

W8.1 x64, MPC-HC x64, NVIDIA GTX480 newest drivers. smooth motion is enabled, 59Hz, 23.976fps movie source. 8bit Output.
I also get a dark image with 10bit Output in window mode and normal colors in fullscreen mode, but I am not sure, if it is really a problem, because the panel is just 8bit. Just wanted to say.

[Werewolfy]

madVR v0.88.0 released

Now I need FEEDBACK from you guys. But since there are so many new features/options, it doesn't make sense to just collect feedback from everyone about everything. That just wouldn't work well. So please allow me to coordinate testing and feedback:

I would like to start with evaluating the new "DEBAND" algorithm written by Shiandow. Please could you guys try it with your various sources and let me know how it compares to madVR's other debanding algorithms in terms of quality and performance? Based on your feedback I will either remove the new algorithm again, or I will replace one or multiple of the old options with the new one. I want to end up with only one Deband option with max 3 different settings. Your feedback is very welcome - thanks!

Thanks for all your work!

The "old" debanding algorithm seems to be better than the new one. To maintain the same level of details as the low preset, you need to use "treshold : 0.3" and "detail level :5" but in that case the debanding is weaker than the old one.

I have also a darker screen when I use 10bit in windowed or fullscreen mode. Only FSE mode is correst. But I noticed that my TV only detects a 10bit signal in FSE mode so it could be the explanation.

[huhn]

I have also a darker screen when I use 10bit in windowed or fullscreen mode. Only FSE mode is correst. But I noticed that my TV only detects a 10bit signal in FSE mode so it could be the explanation.

disable error diffusion and the dark screen issue should go away.

FSE is needed for 10 bit output.

[aufkrawall]

There isn't much I can do about it. In windowed mode Aero/DWM is supposed to take care of tearing. If it doesn't, it's on OS/GPU bug.

I doubt this, since D3D11 works well with MPDN in windowed mode.

[JarrettH]

Edit: don't need to know anymore

[Stan]

When using DX11 mode, the sound continued to play when I closed MPC-HC. I had to kill the process to close it for real. this was with the .0 not the .1 version though, if that was fixed already, nvm me!

[JarrettH]

I'm also getting a tonne of presentation glitches in the DX11 path and my render times are fine.

[Sparktank]

Version: 0.88.1
EDIT2: Using x86-only MadVR only on MPC-HC (KLCP).
Installed KLCP and replaced existing MadVR with madshi's 0.88.1 fix.
I don't use x64 MPC-HC (thus no x64 codecs/renderers).

Mostly default settings except for "Rendering-General" (all ticked (except D3D11)) and LL enabled for downscaling and AR enabled for all scaling (no double, no up ref.).

with D3D11:
Turning on D3D11 and playing in Windowed Mode causes any video to repeat only a few frames (like a really short GIF, probably whatever frames is set to "pressented in advance") while the audio continues in window mode. This applies to any bit depth (7-10 bit).
Switching to FSE, the video plays normally without repeated frame (short gif).

no D3D11:
9bit, plays normally.
10bit, plays normally.

Smooth Motion:
SM + any bit + D3D11 + Ordered Dithering:
Presentation Glitches increase by 22-26 per second if time skipped to a different point in video and continues to increase.
Initially playing video with SM+D3D11 doesn't cause any PG and will continue play smoothly. Until time is skipped.

SM + any bit + D3D11 + Error Diffusion 1/2:
Instant presentation glitches in FSE.

Any bit, no D3D11:
Plays normally.
No glitches.
Skipping time has no negative effect.

Dither:
Error Diffusion 1/2 will cause video to darken in bit depth 9/10.
ED1/2 with D3D11 in FSE will correct brightness.

ED1/2 in 8 bit depth works correctly (no D3D11; with D3D11 requires FSE).

Non-GPU dither does not darken video in any bit depth.

For now I use Ordered Dithering, and no D3D11.

EDIT1:
The time it took to write everything out and double-confirm it...

disable error diffusion and the dark screen issue should go away.

FSE is needed for 10 bit output.

Ah, this summarizes quickly.