madVR - high quality video renderer (GPU assisted)

[e-t172]

Originally Posted by Arm3nian

Ugh madshi was correct in his statement, just like everything he says. That post is just old, and doesn't apply anymore.

I agree, and in addition, I wonder if it ever applied to anything, even on XP. I think madshi was just being overly cautious by making clear that the clock needs to be high quality for SM to work correctly, but there's no evidence that people encountered bad DirectShow clocks. I certainly don't remember anyone on this thread reporting an issue with SM that could be traced back to a bad clock.

Originally Posted by huhn

nope reclock is not nneded for this but not this tapic again...



rendertimes? FSE shouldn't add noticeable rendertimes.
and there are some questionable settings but this is not a madVR setup guide.

Render time without FSE is about 24.5ms and the same for Max Stats, but with FSE turned on, I also get avg. 25ms, but Max Stats show 260ms and of course constant frame drops...

[dansrfe]

I would use MPC-HC Renderer for WASAPI but I get crackling noises randomly while using it for some reason. Sometimes the nosies go away when skipping around the file.

Using DirectShow for now.

[bcec]

Originally Posted by 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. It does however sometimes show a high milisecond count in the renderer as if it was active. Graphics card is a R9 270X.

I don't have an ATI card, but NNEDI3 is certainly working for me for the x64 stack (I have a particular video where NNEDI3 on/off makes a big difference). How did you verify that NNEDI3 was not working?

[heiseikiseki]

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

[huhn]

Originally Posted by XMonarchY

Render time without FSE is about 24.5ms and the same for Max Stats, but with FSE turned on, I also get avg. 25ms, but Max Stats show 260ms and of course constant frame drops...

there is something not working as intended. hard to say what but this shouldn't happen.
can you try the old FSE mode?

you get the old FSE mode by disabling present several frames in advance under exclusive mode settings.

[huhn]

Originally Posted by heiseikiseki

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

and the PRESENT times rise with settings like jinc 3 ar or other settings?
is aero active?
do you get the same problem in fullscreen exclusive mode?

[heiseikiseki]

Originally Posted by huhn

and the PRESENT times rise with settings like jinc 3 ar or other settings?
is aero active?
do you get the same problem in fullscreen exclusive mode?

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

[huhn]

Originally Posted by heiseikiseki

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

on windows 8 and newer aero is "always" active.(there are ways to still disable it but let's be real...)
present times can rise when aero is disabled that's why i asked.

so yes this looks like a yet another optimus issue. sorry i'm pretty bad on this topic. i hope someone with an optimus laptop knows a work around.

[Anime Viewer]

Originally Posted by heiseikiseki

HI，I have very high present time in my optimus laptop.
about 8~10ms

That make my gtx860m can only handle jinc.

Sorry for my bad English.

Originally Posted by heiseikiseki

Setting seems has no relationship with the PRESENT times.
Also in FSE mode I get the same PRESENT time.
I’m using windows 8.1 so no areo active.

This problem is also mentioned before:
http://forum.doom9.org/showthread.php?t=170937

If you're using MPC-HC did you rename the .exe file to something like mpc-hcnv.exe, mpc-hc64.exe, mpc-hc2.exe, etc, then open Nvidia control panel and associate that file to using the Nvidia GPU, then launch a file open the registry and confirm that that it made an nvidia related directory inside the OpenCL key? (ex: HKEY_CURRENT_USER\Software\madshi\madVR\OpenCL\GeForce GTX 680M). In your case it should read 860m.

What your settings are should definitely effect your present times. Besides telling us all the upscaling/downscaling settings you have set we would also need to know :
What dithering option do you have selected (none, random, ordered, ED1, ED2)? Are you trying to use NNEDI3 for resolution doubling and/or quadrupling? Are you using any hardware acceleration options in LAV Video? (If so, what?)

As a test try setting image upscaling and downscaling both to DXVA2 in madVR settings. That will effectively disable any software processing and NNEDI3 settings you have selected and use the GPU. With that setting you should have very low present times.

Are there any other video/audio altering programs you're trying to run with it (SVP, Reclock, etc)?

While a video is running if you right click on it, and select filters what does it show running? (Ex:
Filters currently loaded:
- Default DirectSound Device
- XySubFilter (=> madVR)
- madVR
- LAV Audio Decoder
- LAV Video Decoder
- LAV Splitter Source
)

[Schwartz]

Originally Posted by bcec

I don't have an ATI card, but NNEDI3 is certainly working for me for the x64 stack (I have a particular video where NNEDI3 on/off makes a big difference). How did you verify that NNEDI3 was not working?

Comparing render ms, looking at the Ctrl-J chroma > and image > lines and zooming in to compare the pictures while the regular upscaler was set to nearest neighbor. I have separate scaler profiles for SD and HD respectively (only SD uses image doubling), and switching between them showed no discernable difference.

There's no error message and otherwise it works flawlessly. Image doubling is just flat-out ignored.

[heiseikiseki]

Originally Posted by Anime Viewer

If you're using MPC-HC did you rename the .exe file to something like mpc-hcnv.exe, mpc-hc64.exe, mpc-hc2.exe, etc, then open Nvidia control panel and associate that file to using the Nvidia GPU, then launch a file open the registry and confirm that that it made an nvidia related directory inside the OpenCL key? (ex: HKEY_CURRENT_USER\Software\madshi\madVR\OpenCL\GeForce GTX 680M). In your case it should read 860m.

What your settings are should definitely effect your present times. Besides telling us all the upscaling/downscaling settings you have set we would also need to know :
What dithering option do you have selected (none, random, ordered, ED1, ED2)? Are you trying to use NNEDI3 for resolution doubling and/or quadrupling? Are you using any hardware acceleration options in LAV Video? (If so, what?)

As a test try setting image upscaling and downscaling both to DXVA2 in madVR settings. That will effectively disable any software processing and NNEDI3 settings you have selected and use the GPU. With that setting you should have very low present times.

Are there any other video/audio altering programs you're trying to run with it (SVP, Reclock, etc)?

While a video is running if you right click on it, and select filters what does it show running? (Ex:
Filters currently loaded:
- Default DirectSound Device
- XySubFilter (=> madVR)
- madVR
- LAV Audio Decoder
- LAV Video Decoder
- LAV Splitter Source
)

I'm using MPC-BE and madvr only.
Decode only use CPU,
(becasue if i use DXVA2 on 720p->1080p(jic3AR/jinc3AR) it will cause dropped frames too.)
And I'm sure that I did open the Nvidia control panel to that the MPC-BE to use the Nvidia GPU.
I've checked the registry,that's right.

I am using Ordered dithering .


I’ve tried NEDDI3 16 with
chroma upscaling:Bicubic75
image upscaling:Bicubic75
on sd->1080p often no promblems
but on 720p->1080p
Even the rendering time is only 21ms,
but the PRESENT time is about 10ms
frame drops (render and present queue not getting filled fast enough)





when 720P->1080P



Sorry for my bad English.

[agustin9]

I would try lowering the queues...

[Anime Viewer]

Originally Posted by heiseikiseki

I'm using MPC-BE and madvr only.
Decode only use CPU,
(becasue if i use DXVA2 on 720p->1080p(jic3AR/jinc3AR) it will cause dropped frames too.)

I am using Ordered dithering .


I’ve tried NEDDI3 16 with
chroma upscaling:Bicubic75
image upscaling:Bicubic75
on sd->1080p often no promblems
but on 720p->1080p
Even the rendering time is only 21ms,
but the PRESENT time is about 10ms
frame drops (render and present queue not getting filled fast enough)

Sorry for my bad English.

Things like NNEDI3 can be more taxing on 720p->1080p compared to sd->1080p. What scaling factor do you have set in the doubling screen? Keeping the BiCubic settings you mentioned try with only luma doubling checked, and with the "only if scaling factor is 2.0x (or bigger)" selected.
Does that make any difference?
Your target rectangle is larger that 1080p, so its actually doing more upscaling and downscaling that it would for 1080 resolution. (Is 1920x1140 the native resolution of your screen?)

When you're watching in a default 720p window size what type of numbers are you seeing (half the present ms of fullscreen, or more drastic difference)?
Are you noticing the same numbers in both Full Screen Exclusive and Full Screen Windowed mode, or are they different?
Under trade quality for performance in madVR settings what (if anything) do you have checked?
Are you watching something with subtitles, and (if you are) do you see a difference decoder (aka CPU) at 12, upload and render (aka GPU) at 6, and present at 8. (Usually if your present several frames in advance is too high you'll see presentation glitches, but that doesn't seem to be a problem given your screen shot. However you may want to uncheck the present frames in advance option as a troubleshooting test).

[JarrettH]

Originally Posted by cyberbeing

Back when we were testing the various dithering algorithms, it did seem like more people preferred color vs luma-only dithering noise. Though there was a group of us, including myself, who really disliked the idea of intentionally adding color noise to otherwise pristine chroma, and would rather have the light texture added by luma-only dithering. It's all a trade-off. The type of person who likes grainy movies and hates the appearance of grain-free denoised movies, would likely prefer luma-only dithering. The type of person who likes when their movies have (and keep) a smooth denoised appearance, would likely prefer color dithering.

Personally I like:
HD: Error Diffusion, Option 1, Luma, Static
SD: Error Diffusion, Option 2, Luma, Dynamic
Performance: Ordered Dithering, Luma, Dynamic
Other: Random Dithering, Color, Dynamic


The most popular option favored by others for color dithering I believe was:
Error Diffusion, Option 2, Color, Dynamic

The luma noise lends to smoother gradients IMO.

[omarank]

Originally Posted by cyberbeing

Back when we were testing the various dithering algorithms, it did seem like more people preferred color vs luma-only dithering noise. Though there was a group of us, including myself, who really disliked the idea of intentionally adding color noise to otherwise pristine chroma, and would rather have the light texture added by luma-only dithering. It's all a trade-off.

I too prefer mono dithering.

Normally madVR does dithering at the last step when the video is converted to 8 bits (or lower bitdepth). However, when doing DXVA deinterlacing on a video that is more than 8 bits, madVR dithers the video to 8 bit before it is deinterlaced. I was wondering whether madVR uses the selected dithering algorithm for this dithering step or it uses Random dithering (Colored Static?).

[heiseikiseki]

Originally Posted by Anime Viewer

Things like NNEDI3 can be more taxing on 720p->1080p compared to sd->1080p. What scaling factor do you have set in the doubling screen? Keeping the BiCubic settings you mentioned try with only luma doubling checked, and with the "only if scaling factor is 2.0x (or bigger)" selected.
Does that make any difference?
Your target rectangle is larger that 1080p, so its actually doing more upscaling and downscaling that it would for 1080 resolution. (Is 1920x1140 the native resolution of your screen?)

When you're watching in a default 720p window size what type of numbers are you seeing (half the present ms of fullscreen, or more drastic difference)?
Are you noticing the same numbers in both Full Screen Exclusive and Full Screen Windowed mode, or are they different?
Under trade quality for performance in madVR settings what (if anything) do you have checked?
Are you watching something with subtitles, and (if you are) do you see a difference decoder (aka CPU) at 12, upload and render (aka GPU) at 6, and present at 8. (Usually if your present several frames in advance is too high you'll see presentation glitches, but that doesn't seem to be a problem given your screen shot. However you may want to uncheck the present frames in advance option as a troubleshooting test).

The resolution of my display is 1920*1200. (I also tried a FHD display. That would be the same result.）

And I set the scaling factor to 1.5X.

YES，I've tried the exclusive mode too. Nothing changed，the Present time is still about 8~9ms.（I don't know how to do a printscreen in exclusive mode so I use windows mode（new path）to attach a picture.）

I've tried to choose all the Trade quality for performance setting， seems no relationship of the PRESENT times.
So now I only checked "don‘t analyze gradient angles for debanding".


I've tried to change the numbers of present frames and lower the queues，too. Also tried old path. and other option like overlay.
Also nothing changed with the PRESENT time.

The full screen mode seems double the present time.

When default 720P window size the PRESENT time is about 4ms，and the 1080P is about 8~10ms.

I dont't use subtitles.

Here are some attched picture with the 720P.



720P in window size

720P in full screen（no scaling to 1080p）
seems strange. The PRESENT time also doubled.

[6ari8]

Originally Posted by heiseikiseki

The resolution of my display is 1920*1200.

And I’ve set the scaling factor to 1.5X.

YES，I've tryed the exclusive mode too. Nothing changed，the Present time is still about 8~9ms.（I don't know how to do a printscreen in exclusive mode so I use windows mode（new path）to attach a picture.）

I've tride to choose all the Trade quality for performance setting， seems no relationship of the PRESENT times.
So now I only checked "don‘t analyze gradient angles for debanding".


I've tried to change the numbers of present frams and lower the queues，too. Also tried old path. and other option like overlay.
Also nothing changed with the PRESENT time.

The full screen mode seems double the present time.

When default 720P window size the PRESENT time is about 4ms，and the 1080P is about 8~10ms.

I dont't use subtitles.

Here are some attched picture with the 720P.



720P in window size

720P in full screen（no scaling to 1080p）
seems strange. The PRESENT time also doubled.

Just want to mention something that might help you.

I have an M14x R2 Optimus laptop with an HDMI and a miniDP port. I remember having a similar problem to yours when I was using the HDMI port but not when using the miniDP port. It turned out that the HDMI port is wired to the Intel GPU while the miniDP was wired to the Nvidia GPU. I'm not sure why it had a high present time only when connected to the iGPU but I solved my issue by using the miniDP.

[heiseikiseki]

Originally Posted by 6ari8

Just want to mention something that might help you.

I have an M14x R2 Optimus laptop with an HDMI and a miniDP port. I remember having a similar problem to yours when I was using the HDMI port but not when using the miniDP port. It turned out that the HDMI port is wired to the Intel GPU while the miniDP was wired to the Nvidia GPU. I'm not sure why it had a high present time only when connected to the iGPU but I solved my issue by using the miniDP.

Thanks, but I'm using the DISPLAY port and I also tried the HDMI port.
Nothing changed.Seems In my computer(ASUS ROG GR8), Both of them are wired to IGP controller.

[Anime Viewer]

Originally Posted by heiseikiseki

I'm using MPC-BE and madvr only.
Decode only use CPU,

Are you using the latest Nvidia drivers (350.12) with OpenCL 1.2, or an older driver with OpenCL 1.1?

You could try downloading, installing, and setting up (disabling internal and setting external to preferred) the latest LAV filters. Extending on that tell us exactly what your filters are named when you're running a video and right click it selecting filters. Instead of none for hardware decoding try with CUVID selected.

If you haven't already try putting a check in the use a separate device for presentations box in the general settings.

Do you have v-sync forced to enabled or disabled in Nvidia control panel settings general and mpc-be linked areas?