madVR - high quality video renderer (GPU assisted)

[Warner306]

Greetings to all:
First of all, I apologize if the question I'm going to do may seem silly, but for more that I've read in this thread yet I've a doubt about the " upscale refinements ".
My question is if these are activated in "any" upscaling, I mean, for example, with "image doubling"; or only when acts an algo from "image upscaling" .
Thanks in advance

Upscaling Refinement is activated whenever a video is resized. If your display is 1080p, then any source below 1080p must be resized and will use the Upscaling Refinement setting.

[noee]

A crash report is always helpful, so yes, I would like to see it. It could be the same as the one from Ver Greeneyes and James Freeman (which is identical), but it could also be a different one. One quick look at it and I know.

Meh.....for grins and tickles, I rebooted and now I can't get the crash to occur again.....hate that.....er....or love it.

Be on Windows for a while so I'll see what else.

Oh yeah, I do see that strobing deint issue mentioned above with 17. 16b does not show it.

[Braum]

Upscaling Refinement is activated whenever a video is resized. If your display is 1080p, then any source below 1080p must be resized and will use the Upscaling Refinement setting.

To be sure : if there is only Chroma Upscaling and Image downscaling, does Upscaling refinement is activated ?

Basically : Source resolution > Monitor resolution

[surgical]

Upscaling Refinement is activated whenever a video is resized. If your display is 1080p, then any source below 1080p must be resized and will use the Upscaling Refinement setting.

Understood :)...........thanks Warner306

[Damien147]

Shall I activate superRes filter in chroma upscaling permanently?Is it for good?

[madshi]

To be sure : if there is only Chroma Upscaling and Image downscaling, does Upscaling refinement is activated ?
No.

Shall I activate superRes filter in chroma upscaling permanently?Is it for good?
I don't know. Try it, and let me know what you think!

[madshi]

madVR v0.88.18 released

http://madshi.net/madVR.zip

* fixed: DXVA deinterlacing resulted in heavy flickering
* fixed: low latency mode was always activated with some media players
* fixed: smooth motion FRC: OSD incorrectly reported empty queues
* fixed: crash when OpenCL wasn't working properly
* fixed: FSE mode was sometimes activated when it shouldn't
Any problems (introduced by v0.88.17) left?

[tobindac]

works now with DVB-T on latest mpc-hc (nightly). thanks.

[XMonarchY]

Any plans for an auto-updater for madVR?

EDIT: Yey! 88.18 fixed my presentation glitches on my 120Hz monitor!!!

[baii]

With current update speed, it will need to separate beta and stable for a updater to work for everyone, which is extra effort that can probably spend else where?

Sent from my 306SH

[TOM_SK]

Any plans for an auto-updater for madVR?



htpc-updater - install or update MPC-HC, LAV Filters and madVR automagically.
https://github.com/nikola/htpc-updater

[KhR0N1K]

awesome! DXVA deinterlacing etc. seems fine now with v .18

i noticed something else im not sure is fixable. but in FSE DX11 its not as Exclusive as FSE DX9 was. for example my kaspersky trial notification still pops up while in FSE DX11 mode or the "Completed" msg pops up after successful burn with ImgBurn the video turns black for about half a second then stays Exclusive. the FSE DX9 version stayed fully Exclusive in both these situations.

not a big deal at all. just not sure if anyone has pointed these specific situations out.

:thanks:

[j5627429]

Is there a way to adjust the size of the OSD text?

[j5627429]

I'm currently doing 4:4:4 8-bit RGB 2160p@24hz.

When new video card arrives I want to do 2160p@60hz so I can use SVP instead of my projector's built-in frame interpolation.

However, my display is not fully hdmi 2.0 compliant and only accepts 4:2:0 2160p @60hz.

So my question is: will outputting 4:2:0 negate any of the benefits of madvr processing, or will it be mostly the same?

[James Freeman]

madVR v0.88.18 released
Any problems (introduced by v0.88.17) left?

Thank you madshi !

Fixed my problems.

[Ava Pug]

v0.88.18 has fixed my 10bit stuttering issue with my Intel GPU again. Thanks.

[ryrynz]

0.88.18 fixed my issue with queues not filling on my HD4000 under D3D11, I can now finally use it.

[VincAlastor]

yesterday i donate SmoothVideo Project (SVP). Today i wanna donate madshi for madVR and eac3to, because you give me an amazing home cinema experience and i wanna say thank you. But i can't find a donate button. Can you show me, please?

BTW wich are the best configs for SVP and madVR for 1080p? i just have some artefacts when i use the highest SVP Profile (5g) and the crystal sharp madvr in mpc-hc (against EVR renderer, there are less optical artefacts, but i think i just can't see it, because the renderer isn't that sharp like madVR).

[Braum]

I got a problem :

I have presentation glitches and the render and present queue does not fill when I'm in FSE mode with "present a frame for every VSync" activated.
http://img11.hostingpics.net/pics/868357Sanstitre.png

When I start to play a bluray in FSE this message comes up, but the movie still play nicely.
http://img11.hostingpics.net/pics/777333Sanstitre2.png

When I switch to windowed mode there is no more problem with "present a frame for every VSync" activated.
http://img11.hostingpics.net/pics/752651Sanstitre3.png

If I want to use FSE mode without presentation glitches and queue problems I have to uncheck "present a frame for every VSync".


Intel i5 3570K @4.2GHZ
AMD 7870XT
RAM 8Gb

Windows 8.1 64bits
Potplayer 32bits
Madvr v0.88.18 (same problem with 0.88.16)
LAV Filters 0.65 (using DXVA copy back)
Reclock 1.6

[huhn]

I'm currently doing 4:4:4 8-bit RGB 2160p@24hz.

When new video card arrives I want to do 2160p@60hz so I can use SVP instead of my projector's built-in frame interpolation.

However, my display is not fully hdmi 2.0 compliant and only accepts 4:2:0 2160p @60hz.

So my question is: will outputting 4:2:0 negate any of the benefits of madvr processing, or will it be mostly the same?

chroma up scaling will lose a lot of its meaning.

[chros]

A positive thing I can say about 88.17 with my system is that it cuts the rendering time in half from 23ms to 11ms with DX11, in P8 state (NVIDIA).
Turning D11 off returns to 23ms.
I haven't experienced this neither with v17 nor with 18. Something went wrong on your side.

I got a problem :
When I start to play a bluray in FSE this message comes up, but the movie still play nicely.
http://img11.hostingpics.net/pics/777333Sanstitre2.png

Windows 8.1 64bits
Potplayer 32bits
Madvr v0.88.18 (same problem with 0.88.16)
LAV Filters 0.65 (using DXVA copy back)
Reclock 1.6
Hhmmm strange, I just saw this only in v16b , but it's gone for me with v18 (never saw that before).
What driver and videocard do you have?

[Braum]

I haven't experienced this neither with v17 nor with 18. Something went wrong on your side.


Hhmmm strange, I just saw this only in v16b , but it's gone for me with v18 (never saw that before).
What driver and videocard do you have?
I discovered this problem yesterday, I was in madvr 0.88.17.
So I came back to the .16 version but still got the same problem, and .18 did not sovled it.

AMD 7870XT
Driver version : 15.20 (up to date)

[Dogway]

A question, does NNEDI doubling works on GPU? Can it be offloaded to CPU? I'm kinda trying to figure out if I can play a SD video on 1080p by means of NNEDI doubling (32 and 16 neurons; for luma and chroma) and lanczos 3 + AR upscale on top in a HD 6000 of an Intel NUC. A report here (http://www.missingremote.com/review/intel-nuc-kit-nuc5i5ryk-mini-pc) says not but he could probably by setting 128 neurons or such which would be a waste.

I also just saw super-xbr as a faster alternative, maybe that one can be handled by the iGPU?

[6ari8]

Now *that's* what I call a detailed bug report... :)

Yeah, that doesn't look too good. It's bad that no problems are reported in the OSD, but it still stutters. What happens if you show the madVR seekbar in exclusive mode instead of activating the Ctrl+J OSD? Same stuttering? I suppose you're rendering on the NVidia? What happens if you render on the Intel instead?


It behaves the same way whenever an OSD element is drawn (debug OSD, seekbar, filename, playback time, volume change, etc...) and it still exists in v0.88.18.

My laptop has a mini DisplayPort which is connected directly to the Nvidia GPU so the intel GPU is not involved in the playback. The previous results are with that port.

When I connected my TV to the HDMI port, which is tied to the Intel GPU so the Nvidia CP doesn't even detect an external monitor, I observed strange behavior depending on when I run the media player with the Nvidia GPU or doing everything on the Intel.

1. Run with Nvidia:

http://s24.postimg.org/uusq3krkh/optimus_Run_with_Nvidia_Bicubic.png (http://postimg.org/image/uusq3krkh/)

http://s24.postimg.org/n0nj4fudt/optimus_Run_with_Nvidia_Jinc.png (http://postimg.org/image/n0nj4fudt/)

http://s24.postimg.org/ihfvwo1w1/optimus_Run_with_Nvidia_S_XBR.png (http://postimg.org/image/ihfvwo1w1/)

Do you see the strange present stats? Also, the 23HZ refresh rate changes depending on which GPU I run the player with. As I said, the TV doesn't show up in the Nvidia CP so no custom resolutions are involved.

2. Run with Intel:

http://s24.postimg.org/le2wwy7pt/optimus_Run_with_Intel_Bicubic.png (http://postimg.org/image/le2wwy7pt/)

http://s24.postimg.org/7vw0knvkh/optimus_Run_with_Intel_Lanczos3.png (http://postimg.org/image/7vw0knvkh/)

I'd appreciate it If you can shed some light on why this happens.


And yes, the same stuttering happens in all cases but it's much more random on the Intel and happens only a few times in a couple of minutes.

[nevcairiel]

A question, does NNEDI doubling works on GPU? Can it be offloaded to CPU?

It runs on the GPU, and the CPU would probably be too slow to handle it in realtime at all.

[huhn]

A question, does NNEDI doubling works on GPU? Can it be offloaded to CPU? I'm kinda trying to figure out if I can play a SD video on 1080p by means of NNEDI doubling (32 and 16 neurons; for luma and chroma) and lanczos 3 + AR upscale on top in a HD 6000 of an Intel NUC. A report here (http://www.missingremote.com/review/intel-nuc-kit-nuc5i5ryk-mini-pc) says not but he could probably by setting 128 neurons or such which would be a waste.

I also just saw super-xbr as a faster alternative, maybe that one can be handled by the iGPU?

very unlikely!

super xbr is very cheap that may work.

[Dogway]

Yeah, just finished running some testings, but super-xbr isn't too very nnedi3 alike IMO. Maybe Lanczos + AR is all there is to it.
I wonder what should be the minimum power a GPU should have for NNEDI3? On my PC it works fine with a 750Ti.

[tobindac]

htpc-updater - install or update MPC-HC, LAV Filters and madVR automagically.
https://github.com/nikola/htpc-updater

What's the point of lav when it's included in mpc? Just use mpc nightly.

[Thunderbolt8]

htpc-updater - install or update MPC-HC, LAV Filters and madVR automagically.
https://github.com/nikola/htpc-updaterwould be nice if someone could update this tool to properly work with mpc-hc x64 (nightly) and LAV filter x64. (sorry for OT)

[ashlar42]

I think I'm getting crazy.

I am using a GTX660 (driver 353.30), connected to a Pioneer LX-6090 (Kuro 9G). Using madVR and LAV Filters I set up everything to output 10bit. madVR OSD agrees with me. Nvidia control panel allows me to select 12bit color resolution.

Now the thing that's driving me crazy is that once in a while I get, when the TV switches resolution to 24Hz, the info bit in the top corner (this is a TV small info panel on screen) stating 36bit mode. But this happens once in a while and, for the most part, I get the "regular" 1080p info bit, with no 36bit mention. It appears to be totally random. I just started a movie, got the 36 bit thing, recalled the info panel with the proper remote button, again 36 bit. Stopped the movie, restarted it... just 1080p, no 36 bit. NOTHING changed.

This morning I went out and bought a new HDMI cable, high speed and everything (Panasonic cable)... no difference.

[chros]

What's the point of lav when it's included in mpc? Just use mpc nightly.
Well, I'm overriding only the Video part of the internal lavfilters with the external one: the internal one gave me more GPU usage a while back (e.g. half a year ago, I haven't checked nowdays), don't ask how and why, 'cause I'd think in the opposite way.
(I'm using cuvid in lav video for acceleration.)

[aufkrawall]

Could it be that "image doubling - always" is broken?
It doesn't use NNEDI3 for 1080p -> WQHD here.

Edit: My mistake, stupid Windows Update driver for Windows 10 doesn't include Cuda/OpenCL...

[XMonarchY]

Could it be that "image doubling - always" is broken?
It doesn't use NNEDI3 for 1080p -> WQHD here.

Edit: My mistake, stupid Windows Update driver for Windows 10 doesn't include Cuda/OpenCL...

Never-ever use Windows Update drivers for anything other than Windows' native drivers (System Devices, USB Hubs, etc). I still have no idea why Windows updates do not include proper GPU drivers from the manufacturer.

[XMonarchY]

chroma up scaling will lose a lot of its meaning.

Would Chroma Upscaling also lose meaning if TV is set to output only to 4:2:2? 4:4:4 looks awful on my TV due color luminance issues.

[huhn]

yes it is downscaled again same as with 4:2:0 so the benefit from chroma upsampling is way lower on a 4:2:0 or 4.2:2 screen compared to a 4:4:4 screen.

[huhn]

Never-ever use Windows Update drivers for anything other than Windows' native drivers (System Devices, USB Hubs, etc). I still have no idea why Windows updates do not include proper GPU drivers from the manufacturer.

there is no other "official" way to get the newest nvidia driver for windwos 10.

[Braum]

yes it is downscaled again same as with 4:2:0 so the benefit from chroma upsampling is way lower on a 4:2:0 or 4.2:2 screen compared to a 4:4:4 screen.

I have a 4:4:4 screen and when I play a bluray I got this :

h264 8bit 4:2:0 -> NV12 8bit 4:2:0

It make sense in a way because the source is 4:2:0 but shouldn't chroma upscaling output in 4:4:4 ?

[huhn]

I have a 4:4:4 screen and when I play a bluray I got this :

h264 8bit 4:2:0 -> NV12 8bit 4:2:0

It make sense in a way because the source is 4:2:0 but shouldn't chroma upscaling output in 4:4:4 ?

that what madVR get's as an input the output of madVR is 8 or 10 bit RGB which is 4:4:4

[6ari8]

there is no other "official" way to get the newest nvidia driver for windwos 10.

What do you mean by "official"?

Nvidia has been releasing drivers for W10 for a while now.

http://www.nvidia.com/download/driverResults.aspx/86512/en-us

[Braum]

that what madVR get's as an input the output of madVR is 8 or 10 bit RGB which is 4:4:4

Thank you, indeed it makes more sense :)

[huhn]

What do you mean by "official"?

Nvidia has been releasing drivers for W10 for a while now.

http://www.nvidia.com/download/driverResults.aspx/86512/en-us

that's 353.30 not 353.50 which is now in windows update and currently only there.

i have the 353.50 installer now on my disc so I could share this and by doing this it would be unofficial.

i would like to stop this topic now maybe the driver will be released soon on the nvidia homepage (and i hope with openCL/CUDA) or maybe not.

[6ari8]

that's 353.30 not 353.50 which is now in windows update and currently only there.

i have the 353.50 installer now on my disc so I could share this and by doing this it would be unofficial.

i would like to stop this topic now maybe the driver will be released soon on the nvidia homepage (and i hope with openCL/CUDA) or maybe not.

Oh I see. Sorry for misunderstanding.

Yep 353.50 is still not up on Nvidia. MS just released their first RTM build so that may have a relation with the driver they put up in Windows Update today. I think Nvidia will release it soon.

[XMonarchY]

353.49 Hotfix is the official latest Windows 10 NVidia driver that can be downloaded from NVidia website. I do not understand all this arguing. MS never provided proper NVidia drivers that contain CUDA, OpenCL, and PhysX because MS is stupid like that. If NVidia does not have an official link to these new 353.50 drivers, then you should not be using them.

As far as 4:2:2 chroma upscaling goes - I can undeniably see the difference between selecting Jinc and selecting NNEDI3 when I zoom in slightly, so there is definitely some improvement, even with 4:2:2 output.

Another thing I wanted to ask Madshi was about dithering on a 12bit display, using 10bit+ madVR setting. Does it actually make sense to use ED or whichever dithering on 12bit display? Disabling dithering on 12bit display gets rid of quite a bit of noise and yet gradations remains very smooth. I noticed another difference too. When I toggle between ED dithering and no dithering, using 12bit NVidia CP setting and 10bit+ D3D11 Exclusive mode setting, the actual image colors change as if a different colorspace is used. That makes no sense to me because AFAIK ED or whichever dithering is supposed to create a perception of more colors by reducing visible gradations (same as using higher color bit depth), but is it supposed to actually show/create DIFFERENT colors? That goes against creating a 3DLUT/calibrating TV for the purpose of seeing the image as it was mastered...

[sneaker_ger]

I get some major back and forth jumping with v0.88.17 and v0.88.18. No problems prior to this.
http://217.160.126.132/madVRv0.88.18.0%20back%20and%20forth%20jumping.7z

D3D9
Win 7 x64 (Aero off)
MPC-HC x64 (LAV)
Radeon HD5850 Catalyst 15.7

[MistahBonzai]

When I toggle between ED dithering and no dithering, using 12bit NVidia CP setting and 10bit+ D3D11 Exclusive mode setting, the actual image colors change as if a different colorspace is used. That makes no sense to me because AFAIK ED or whichever dithering is supposed to create a perception of more colors by reducing visible gradations (same as using higher color bit depth), but is it supposed to actually show/create DIFFERENT colors?
Did you carry out this test on an Eizo Foris FG2421 (as listed in your sig)? If so then the FG2421 has 8-bit MVA panel and it utilizes its FRC stage to create a pallet of 1 billion colors. I'm not sure how adjusting dithering as if it were a 10 (or 12?) bit panel affects color accuracy. I looked into the FG2421 as a 2nd display based on favorable reviews regarding calibration results. In the end I passed in part due to the lack of true 10-bit.

[Dogway]

Is it normal that playback (initialization) delays 3 or 4 seconds for every video (even windowed)? I tested LAV Video with CUVID, DXVA and CPU, then also with madVR, EVR and VMR-9. I say because I just did some tests using the iGPU (for some performance benchmarks) and all videos played fine within a second.

-----------------------

I did some benchmarks concerning the iGPU Intel HD 4600 to extrapolate with UCFF systems like Intel NUC. Actually an Intel HD 6000 is very close to a HD 4600 according to passmark. The goal is to upscale using NNEDI. The miliseconds in brackets are the average rendering time.

Using DXVA2 (copy-back) in LAV Video:
(30ms) You can set all setting tuned for quality with NEDI (no neuron) doubling, this is probably better than only lanczos3 but still feels like droping frames (not reported in stats)
(44ms) NNEDI3 (16) doubling only luma with everything else to minimum (optimizations) is still 3 frame drops + 3 reps off from ideal, probably needs Iris Pro to make this work, or skylake iteration of iGPU.
(18-23ms) 1080p content is fine since no scaling is needed
(21ms) 720p content (internally lanczos upscale)

[Prolesious]

there is no other "official" way to get the newest nvidia driver for windwos 10.

So there is a newer driver available other than the what's on nvidia's page for windows 10?

Nvidia's page has one there which is version 353.30 released on, June 22, 2015

[madshi]

i noticed something else im not sure is fixable. but in FSE DX11 its not as Exclusive as FSE DX9 was. for example my kaspersky trial notification still pops up while in FSE DX11 mode or the "Completed" msg pops up after successful burn with ImgBurn the video turns black for about half a second then stays Exclusive. the FSE DX9 version stayed fully Exclusive in both these situations.
I don't think there's much I can do about that. That's probably a different behaviour between DX9 and DX11.

Is there a way to adjust the size of the OSD text?
Not at this time. Maybe some time in the future.

I'm currently doing 4:4:4 8-bit RGB 2160p@24hz.

When new video card arrives I want to do 2160p@60hz so I can use SVP instead of my projector's built-in frame interpolation.

However, my display is not fully hdmi 2.0 compliant and only accepts 4:2:0 2160p @60hz.

So my question is: will outputting 4:2:0 negate any of the benefits of madvr processing, or will it be mostly the same?
chroma up scaling will lose a lot of its meaning.
^

Chroma upscaling is one of the less important things. So the loss might not be dramatic. But it's a loss, in any case.

yesterday i donate SmoothVideo Project (SVP). Today i wanna donate madshi for madVR and eac3to, because you give me an amazing home cinema experience and i wanna say thank you. But i can't find a donate button. Can you show me, please?
At this point I'm not accepting donations just yet. I plan to, when I reach version 1.0, though. Thanks for your offer, though!

BTW wich are the best configs for SVP and madVR for 1080p? i just have some artefacts when i use the highest SVP Profile (5g) and the crystal sharp madvr in mpc-hc (against EVR renderer, there are less optical artefacts, but i think i just can't see it, because the renderer isn't that sharp like madVR).
Can't help with that, don't have much experience with SVP.

I have presentation glitches and the render and present queue does not fill when I'm in FSE mode with "present a frame for every VSync" activated.

When I start to play a bluray in FSE this message comes up, but the movie still play nicely.

When I switch to windowed mode there is no more problem with "present a frame for every VSync" activated.

If I want to use FSE mode without presentation glitches and queue problems I have to uncheck "present a frame for every VSync".
You can try v0.88.19. But if you get good playback with "present a frame for every VSync", then by all means use it!

This whole problem seems to be related to DXVA scaling. You could also try using a madVR scaler instead of DXVA.

It behaves the same way whenever an OSD element is drawn (debug OSD, seekbar, filename, playback time, volume change, etc...) and it still exists in v0.88.18.

And yes, the same stuttering happens in all cases but it's much more random on the Intel and happens only a few times in a couple of minutes.
Please try again with v0.88.19. If you still get the issue, try disabling both D3D11 and the "use a separate device for presentation" options. That might "solve" the issue. At this point I think if you still get the issue with v0.88.19, it's most probably some sort of GPU driver issue, unfortunately...

I am using a GTX660 (driver 353.30), connected to a Pioneer LX-6090 (Kuro 9G). Using madVR and LAV Filters I set up everything to output 10bit. madVR OSD agrees with me. Nvidia control panel allows me to select 12bit color resolution.

Now the thing that's driving me crazy is that once in a while I get, when the TV switches resolution to 24Hz, the info bit in the top corner (this is a TV small info panel on screen) stating 36bit mode. But this happens once in a while and, for the most part, I get the "regular" 1080p info bit, with no 36bit mention. It appears to be totally random. I just started a movie, got the 36 bit thing, recalled the info panel with the proper remote button, again 36 bit. Stopped the movie, restarted it... just 1080p, no 36 bit. NOTHING changed.
Does madVR report 10bit output mode? If so, it must be the GPU drivers which do these weird things. All madVR can do it output 10bit to D3D11. The rest is up to the GPU driver.

Another thing I wanted to ask Madshi was about dithering on a 12bit display, using 10bit+ madVR setting. Does it actually make sense to use ED or whichever dithering on 12bit display? Disabling dithering on 12bit display gets rid of quite a bit of noise and yet gradations remains very smooth. I noticed another difference too. When I toggle between ED dithering and no dithering, using 12bit NVidia CP setting and 10bit+ D3D11 Exclusive mode setting, the actual image colors change as if a different colorspace is used. That makes no sense to me because AFAIK ED or whichever dithering is supposed to create a perception of more colors by reducing visible gradations (same as using higher color bit depth), but is it supposed to actually show/create DIFFERENT colors? That goes against creating a 3DLUT/calibrating TV for the purpose of seeing the image as it was mastered...
Dithering should not affect colors. If they do, something weird is going on. IMHO dithering should still be beneficial even when using 10bit output mode in madVR. But let your eyes be the judge and choose what looks best to you.

I get some major back and forth jumping with v0.88.17 and v0.88.18. No problems prior to this.
Does it still occur with v0.88.19? I fear that this might be a bug in the GPU drivers. Does unchecking the "use a separate device for presentation" option help? FWIW, on my PC, everything works fine with Intel, AMD and NVidia GPUs.

Is it normal that playback (initialization) delays 3 or 4 seconds for every video? I tested LAV Video with CUVID, DXVA and CPU, then also with madVR, EVR and VMR-9.
So you tested this and that, but you didn't really tell us whether the delay occurs always, or only in some of these tests (and in which). Because of that I can't really say anything meaningful.

[madshi]

madVR v0.88.19 released

http://madshi.net/madVR.zip

* fixed: using the sharpening algos sometimes produced startup crashes
* fixed: forced film mode sometimes produced crashes
* fixed: weird resolution videos resulted in black screen with some old GPUs
* optimized OSD rendering performance slightly
* when running in a debugger, keyboard hook is not set, anymore
Some more bugfixes.

[Dogway]

So you tested this and that, but you didn't really tell us whether the delay occurs always, or only in some of these tests (and in which). Because of that I can't really say anything meaningful.

I tested with this and that and I always get 3-4 second delays (with nvidia GPU), 1 second then using iGPU. Out of curiosity, can I tell what GPU to use for madVR (just like I can with LAV Video; DXVA, QS, CUVID...)?