madVR - high quality video renderer (GPU assisted)

[madshi]

Originally Posted by mogli

Small bug report:

Most movies finish without any drop or glitch. However since v0.88.x with D3D11 activated (and frame for every VSync off) very rarely randomly in a movie drops and glitches start to occur at about 20 per second (both per OSD and visible on screen). This happens only in FSE mode and doesn't stop by itself but only by switching to windowed mode once and back to FSE. It seems to occur more often with some interlaced NTSC DVDs then with 24p BDs but it isn't reproducible.
I guess some rare conditions can drive the D3D11 logic nuts.

Windows 8.1 x64, NVIDIA 350.12

I suppose you've unchecked the option "present a frame for every VSync"? This option is enabled by default. Re-enable it, and the problem will probably go away.

Originally Posted by luk008

I've reboot and resetted madVR to default settings, but still the same problem.
Maybe a driver issue? Are you using 15.4 as well?

PS: How can I print the screen in FSE to show you the issue?

I don't know which driver version I'm using, it's pretty old. Definitely much older than 15.4. You can't print screen in FSE, unfortunately.

Originally Posted by XMonarchY

Ctrl+alt+shift+P during Rec.601 content playback shows "Source primaries: Auto-detect SMPTE (170M)

SMPTE 170M is not the same as SMPTE-C, I think. madVR at least treats them differently. That also explains why madVR is using the BT.709 3dlut.

Originally Posted by XMonarchY

But that cannot be the case because ctrl+J menu reports BT.601 primaries for my SMPTE 170M content...

That sounds strange that Ctrl+Alt+Shift+P shows something different than the debug OSD! Can I see a screenshot of the debug OSD with Ctrl+Alt+Shift+P overlayed on top of it?

Originally Posted by cyberscott

Just tried 88.8. Previous to this version, the original debanding was beating out Shiandow deband script, even with the clear improvements since its' introduction into madVR
I can now say that Shiandow deband script included in 88.8 has now edged out madVR's deband to my eyes using Power 0.50 and Margin 0.02.

Ok, thanks for the feedback. Are you talking about madVR's "high" preset, only? Or do you find Shiandow's deband also better than madVR's "low" and "medium" settings? If so, which settings would you then suggest for "low" and "medium"?

More votes/opinions, please?

[madshi]

Originally Posted by MysteryX

Since SuperRes is being applied after resizing, shouldn't it have a similar performance hit no matter whether the source material is 288p or 480p?

For 288p material, SuperRes 2 pass causes rendering time to go from 6.2 to 11.9ms on 768p display

For 480p material, same setting causes rendering time to go from 9.3 to 20.2ms!

Good question. I would say your comment makes sense. Don't know why the increase is higher for 480p material. But current priority is on debanding, not SuperRes.

[luk008]

I'll try to downgrade the driver.

I've taken a photo of my issue with FSE 10 bits: http://i.imgur.com/6dntf5u.jpg

[bacondither]

Arrgh... so many new versions in such short time, i'm drowning in them!

I'll do some deep testing with this new debanding majigger this weekend and i will hopefully come to some conclusion.

[MysteryX]

Originally Posted by madshi

Good question. I would say your comment makes sense. Don't know why the increase is higher for 480p material. But current priority is on debanding, not SuperRes.

It's probably the same thing that is causing rendering on 1080p to go from 14ms to 39ms with SuperRes.

Btw, this new version allows me to add SuperRes for SD content on laptop, and Debanding on 1080p content (that needs no upscaling) looks great!

Is there a way to add this SuperRes effect to an AviSynth script? I like the effect it gives for upscaling SD to HD.

[luk008]

Older drivers have worse performance for R9 270 and don't solve the issue.

[cyberscott]

Originally Posted by madshi

Ok, thanks for the feedback. Are you talking about madVR's "high" preset, only? Or do you find Shiandow's deband also better than madVR's "low" and "medium" settings? If so, which settings would you then suggest for "low" and "medium"?

More votes/opinions, please?

Oops, forgot to mention that I have always used High/High in madVR deband settings as this always looked best on the content I watch. Shiandow's deband default settings of Power 0.50 and Margin 0.00 is very close to madVR High/High settings to my eyes.

Originally Posted by madshi

That sounds strange that Ctrl+Alt+Shift+P shows something different than the debug OSD! Can I see a screenshot of the debug OSD with Ctrl+Alt+Shift+P overlayed on top of it?
More votes/opinions, please?

I edited my thread like 20 times to make it as clear as possible, so I am not sure which edition of my reply you're replying to, lol!

Just to recap, my content was labeled SMPTE 170M, which madVR treated differently from content labeled SMPTE C. When I toggled and enabled SMPTE C through ctrl+shift+alt+p, proper Rec.601 3DLUT in SMPTE C line was applied. 3DLUT's in SMPTE C line simply do not apply to SMPTE 170M content...

Here's a screenshot of the original ctrl+j menu WITHOUT toggling. It is how madVR "auto-detected" content I guess. madVR applied Rec.709 3DLUT from BT.709 line to this content.



I can't find how SMPTE 170M and SMPTE C differ in terms of primaries. They both use Rec. 601 primaries, gamma and D65 white point, don't they? I wonder if ArgyllCMS can create SMPTE 170M 3DLUT. Any ideas? I know this is madVR forum (not Display Calibration forum), but I figured you may know...

[leeperry]

Is 10bit supposed to work via DVI-SL on AMD with 13.12 and W7SP1?

I tried 88.7 and 88.8, enabled desktop composition in W7SP1, D3D11 10bit output in mVR with 8 frames in advance and I don't have desktop composition disabled in mVR.

I also tried to reboot after enabling Aero, reset mVR's defaults and run the latest nightly build of MPC-HC and I'm still stuck in DX9 land

The only reason I could think of is that I'm using a DVI-SL cable to my HDMI Sammy TV but 10bit is apparently supported via DVI huh:

my Eizo CG243W does indeed accept full 10bit input over a single-link DVI-D connection (it doesn't have dual-link).

I didn't try 10bit earlier as I don't like Aero Peek, Aero's reflections effects on windows and when I switch from FSE to FSW in PotP with Aero off I get flashing black frames but when it's enabled I get zombie previous frames from the movie which is plain annoying.

Originally Posted by madshi

NEDI has much stronger fractal artifacts than NNEDI3. However, SuperRes really helps there. So a combination of NEDI+SuperRes *may* be able to compete with NNEDI3. However, you can also combine NNEDI3 with SuperRes, too! So there's a lot of things to try out to find the best quality. But as I said, it's too early to discuss that. We'll come to SuperRes discussion sooner or later, but we're not there yet.

Using SR, EE is just too strong with NNEDI3 to my eyes as in moving objects feeling like cel shading, the picture screams DIGITAL(using ED2@8bit). Downscaling it with SR enabled on 720p content using CR AR LL is seriously too sharp whatever anime or live action, J3AR looks the most natural but is also way too soft.....NEDI makes for a nice in-between to my eyes, me like

I guess I could tame down SR'ed NNEDI3's über-sharpness with a softer downscaler as CR AR LL is too sharp and Softcubic50 AR LL obviously too soft but CR AR LL looks great with NEDI, the latter requiring far less ressources and also taking care of chroma kinda being the icing on the cake

Originally Posted by QBhd

I had figured this out quite quickly when using D3D11... pressing the TAB key to pull up the OSD of PotPlayer would sort of kick madVR out of whack and it couldn't recover without a playback restart. Selecting "display messages on image itself" helped with the TAB OSD message... put pausing still borks madVR, even just clicking on the other desktop can do it. So I have resorted to not using D3D11.

What's the "other desktop"? I only use PotP's D3D GUI, no other OSD.

I wouldn't need to break a sweat about 10bit no workee on my box if PotP becomes unusable to begin with.

Originally Posted by huhn

i'm pretty sure this has nothing to do with seeing banding.
you can easily get banding on a gray ramp.

That's the thing, we're judging colorful banding and colorblinds see less tints.

[XRyche]

Originally Posted by madshi

madVR v0.88.8 released

http://madshi.net/madVR.zip

* updated to latest Shiandow deband script
* fixed: D3D11 didn't activate with "frames presented in advance" set to 16
* fixed: ConfigureDisplayModeChanger(allowResolutionChanges = false) bug

So here's the latest Deband update again. Please retest - thank you!

(P.S: And you can set the number pre presented frames in advance back to 16 for D3D11, if you like. madVR will then automatically reduce it to 15.)

The frames in advance set to 16 was precisely the issue that I was having. Thank you for fixing it. When I got home i started tinkering with it and found that this was the exact issue and was just about to post it . It is now working for me exactly the way you described it. Set FSE pre-presented frames to 16 and it automatically sets it to 15.

[aufkrawall]

I still think that the old deband implementation is more efficient in terms of quality:
At high, it softens the image less than Shiandow's at 0.5 and fights banding still partially better (but also partially worse).

imho there's not much room for highering the values when even 0.5 softens the image more than old algo@high.

Anyway, thank you two for your continuous great work!

[XRyche]

Originally Posted by luk008

I've reboot and resetted madVR to default settings, but still the same problem.
Maybe a driver issue? Are you using 15.4 as well?

PS: How can I print the screen in FSE to show you the issue?

I am currently using 15.3 and didn't have a que issue in FSE D3D11 10 bit. I'm downloading 15.4 now and will verify whether 15.4 is giving me the same issue as you, as soon as I restart.

[madshi]

Originally Posted by XMonarchY

The problem was that SMPTE 170M aws not detected as SMPTE C, even though ctrl+j menu reported that my SMPTE 170M content had "matrix BT.601". Its all a bit confusing...

Not at all. The matrix is a totally different thing from the primaries. E.g. EBU/PAL and SMPTE-C have different primaries, but use the same BT.601 matrix.

Originally Posted by XMonarchY

Do you think you could add SMPTE 170M standard support for SMPTE C line or make an extra line for SMPTE 170M? AFAIK SMPTE C is the color gamut that uses Rec.601 primaries, while SMPTE 170M is the actual standard, not just the color gamut name. I mean its no big deal really, since I can toggle now.

Hmmmm... Have checked the numbers. It does seem as if the primaries are the same. So probably I should use the SMPTE-C 3dlut for SMPTE-170M files. Btw, SMPTE-240M seems to use the same primaries, too - but a different matrix <sigh>.

Originally Posted by leeperry

Is 10bit supposed to work via DVI-SL on AMD with 13.12 and W7SP1?

I tried 88.7 and 88.8, enabled desktop composition in W7SP1, DCD11 10bit output in mVR with 8 frames in advance and I don't have desktop composition disabled in mVR.

Do you get D3D11 mode working at all? In windowed mode? Or with 8bit display bitdepth? I rather think that the problem is D3D11 mode in general, and not the 10bit.

Do you have the Windows 7 Platform Update installed? That's a requirement for madVR's D3D11 presentation path.

Originally Posted by leeperry

Using SR, EE is just too strong with NNEDI3 to my eyes as in moving objects feeling like cel shading, the picture screams DIGITAL(using ED2@8bit). Downscaling it with SR enabled on 720p content using CR AR LL is seriously too sharp whatever anime or live action, J3AR looks the most natural but is also way too soft.....NEDI makes for a nice in-between to my eyes, me like

There are different SR defaults for NNEDI3 compared to NEDI. Have you adjusted them? Also you can tame SR further down, if you still find it too sharp.

Originally Posted by luk008

I'll try to downgrade the driver.

I've taken a photo of my issue with FSE 10 bits: http://i.imgur.com/6dntf5u.jpg

Looks like the presentation times are the problem. They probably slow rendering down so much that the rendering queue doesn't fill. Don't know why that happens - especially only in 10bit! Strange...

Originally Posted by bacondither

Arrgh... so many new versions in such short time, i'm drowning in them!

I'll do some deep testing with this new debanding majigger this weekend and i will hopefully come to some conclusion.

Great - looking forward to your feedback!

(Use only the latest version, ignore the older versions. Maybe that will stop you from drowning... )

Originally Posted by cyberscott

Oops, forgot to mention that I have always used High/High in madVR deband settings as this always looked best on the content I watch. Shiandow's deband default settings of Power 0.50 and Margin 0.00 is very close to madVR High/High settings to my eyes.

Ah ok. So probably no vote from you about old low/medium vs Shiandow?

Originally Posted by XRyche

The frames in advance set to 16 was precisely the issue that I was having. Thank you for fixing it. When I got home i started tinkering with it and found that this was the exact issue and was just about to post it . It is now working for me exactly the way you described it. Set FSE pre-presented frames to 16 and it automatically sets it to 15.

Good to hear!

Originally Posted by aufkrawall

I still think that the old deband implementation is more efficient in terms of quality:
At high, it softens the image less than Shiandow's at 0.5 and fights banding still partially better (but also partially worse).

imho there's not much room for highering the values when even 0.5 softens the image more than old algo@high.

Thanks for your feedback!

I've done a quick comparison myself and also found that the old high algo works partially better and partially worse at debanding, while keeping slightly higher detail. Wasn't sure which I preferred overall...

[sneaker_ger]

Earlier someone pitched the idea of adding a fourth option ("ultra"/"very high") instead of replacing the current "high". Maybe that's viable solution?

[madshi]

Originally Posted by sneaker_ger

Earlier someone pitched the idea of adding a fourth option ("ultra"/"very high") instead of replacing the current "high". Maybe that's viable solution?

As a last resort maybe, but only if that ultra option would deliver better debanding in all situations. Currently Shiandow's deband seems to be sometimes a bit stronger and sometimes a bit weaker in debanding strength compared to madVR's old "high" preset.

I would still prefer to keep just low/medium/high, and just replace some or all of the old presets with the new algorithm - of course only if the new one is really better. So, still waiting for more feedback!

[aufkrawall]

Maybe both algos can be combined at "ultra"?
There is hardly an additional detail loss when I enable old high and Shiandow's at 0.1/0 at the same time, but Shiandow's algo clearly reduces banding left by madVR's old one.

[BetA13]

Originally Posted by madshi

Maybe you need to reapply the fix suggested by cyberbeing, which IIRC was clearing the madVR OpenCL registry storage key and then recreating the kernel with the 64bit madVR build.


Which modes do you have listed in the madVR display mode changer? And which exact mode does either D3D9 or D3D11 switch to? (Using the same video file)

wooow, that openCL reg fix worked flawless Perfect.

to the second part..
i have no modes listed there, i use the automatic detection. i think...

Picture...


ill try to explain it a bit better...
i have an toshiba hd tv connected via HDMI.
Res. is 1920x1080 (i) ?, at 30hz. BUT in FSE its at 60hz/fps(with d3d9 new path)
Also, i had to resize the resolution with nvidia panel, when i set 1920x1080 i have stuff missing on top, sides, etc.. so i have to resize with nvidias tool...i dont know if this resizing is important to mention, so i just wrote it also...

pic:

[img=http://s10.postimg.org/l8ihbnqw5/dsfsdfsdfsdfsdfsdf.jpg]

its an old TV, but it works fine and the picture is awesome thanks to your fine work..

anyway, when i go into fullscreen exclusive mode usually what happens is that it displays 60hz/60fps also in OSD.



And thats what im used to, works perfect..

now when i enable this d3d11 mode, it goes to 25hz, changes the system resolution to TRUE 1920x1080 @ 25hz, wich is not what should happen, i guess.. and therefore teh picture isnt as fluid anymore..

greetz

edit:

i see tehres an new update, i shall try this first and report back

[XRyche]

Originally Posted by XRyche

I am currently using 15.3 and didn't have a que issue in FSE D3D11 10 bit. I'm downloading 15.4 now and will verify whether 15.4 is giving me the same issue as you, as soon as I restart.

Catalyst 15.4 fills up the ques to their appropriate levels for me so I can't verify that it is a driver issue unless it's hardware related since I don't have an R9 270 but an R7 265.

[XRyche]

As far as debanding goes I actually prefer Shiandow's over MadVR's for anime; and madVR's over Shiandow's for movies and DVR files. Shiandow's is more aggressive and kills some details, which isn't really that big of an issue for anime imho. MadVR's preserves details better while still doing a very good job of overall debanding which for me, works better for movies. This is just my opinion and isn't based on any thorough concrete testing so "take it with a grain of salt".

Added bonus (at least for me) Shiandow' debanding is a good for cleaning up old, badly compressed, and just general a mess TIVO rips and such for the early 2000's.

[MysteryX]

I replaced by AMD 12.200 Beta drivers with the latest 15.4 (Leshcat version).

Great news!! Finally a driver that provides better performance! Especially for lower resolution videos so it most likely better process Jinc upscaling.

Before/After
288p: 16/15ms
720p: 14/13.4ms
1080p: 11/11ms

As far as DirectX 11 presentation, it still doesn't work. It still gives a black screen.

NNEDI3 still doesn't work either. But this time the setting slightly alters rendering times, but not enough to show NNEDI3 usage.

This is with a Radeon HD 7670M on Windows 7 x64