madVR - high quality video renderer (GPU assisted)

[DragonQ]

he multiple time said there will be a 64 bit version at least for 1.0.

he said it would take about 2 days to create a 64 bit version and he like to spend the time on something more meaning full to him like adding feature and things like that.
As long as v1.0 is released before 32-bit emulation is dropped from Windows 15 x128. :D

[aufkrawall]

Could anybody please give me a brief refresh regarding differences of OpenCL <-> D3D-Interop between AMD, Nvidia and Intel?

[huhn]

Could anybody please give me a brief refresh regarding differences of OpenCL <-> D3D-Interop between AMD, Nvidia and Intel?

AMD does something like an copyback making it a lot slower as it should be.

but AMD has so good openCL performence that it still beats most nvidia cards for the same money.

[aufkrawall]

If Nvidia is said to do no unnecessary CB, why is CPU usage significantly higher with NNEDI3 scaling than without?

[huhn]

If Nvidia is said to do no unnecessary CB, why is CPU usage significantly higher with NNEDI3 scaling than without?

I don't notice something like this with both my AMD HTPC r9 270 and my nvidia Gaming/encoding PC 760 gtx.

[flashmozzg]

If Nvidia is said to do no unnecessary CB, why is CPU usage significantly higher with NNEDI3 scaling than without?

Could be that it translates OpenCL calls to CUVID. But I don't know how it would affect CPU load in reality. (btw copyback shouldn't affect CPU, it's just memmory bandwith issue)

[leeperry]

AMD has so good openCL performance that still beats most nvidia cards for the same money.
Only over PCI-E 3.0 and only with the 13.12 drivers in many ppl's experience(including mine with a HD7850 on W7SP1).

[DragonQ]

Yeah my AMD GPU was one of the very worst in terms of performance reported in this thread and it's probably because I have an older PCI-E 2.0 chipset (X58). Since madshi tweaked the OpenCL implementation to reduce the copyback, I can use 16 neurons for some SD material but that's it. Before that tweak I couldn't use NNEDI3 at all.

[kasper93]

1. AMD cards are a lot faster in openCL workloads than NV. And it generally have better implementation of it.
2. AMD has a bug with OpenCL<->D3D Interop. It does copy data to CPU memory it seems. This is possibly driver bug, but we will likely never see a fix for that.
3. Older AMD card are VERY slow when copying data back to CPU memory. (They improved that in HD7xxx+.) Therefor they are very slow when it comes to madVR's openCL processing. Even if GPU raw power is enough to handle certain computing.

Of course then we have other factors that affect memory copy rates. Like PCI-E version and so on. But it all boils down to the fact that AMD does this stupid copy-back, which in theory is not needed, but I think they don't care enough to fix that. (if it is possible on driver level).

Ps. Someone were in contact with AMD guy and they said they look at the problem. But frankly I doubt they ever will. :<

[flashmozzg]

1. AMD cards are a lot faster in openCL workloads than NV. And it generally have better implementation of it.
2. AMD has a bug with OpenCL<->D3D Interop. It does copy data to CPU memory it seems. This is possibly driver bug, but we will likely never see a fix for that.
3. Older AMD card are VERY slow when copying data back to CPU memory. (They improved that in HD7xxx+.) Therefor they are very slow when it comes to madVR's openCL processing. Even if GPU raw power is enough to handle certain computing.

Of course then we have other factors that affect memory copy rates. Like PCI-E version and so on. But it all boils down to the fact that AMD does this stupid copy-back, which in theory is not needed, but I think they don't care enough to fix that. (if it is possible on driver level).

Ps. Someone were in contact with AMD guy and they said they look at the problem. But frankly I doubt they ever will. :<
It's a bug in Dx9<->OpenCL interop. I can hardly imagine any other prgram besides madVR using it. That's why they don't bother.
They said they knew about this problem but it's low priority.

[Fadeout]

Anyone knows what kind of "features" are important on MadVR videocard-wise, considering that decoding is done not in hardware?

I mean, what one should look at to consider scaling performance and presentation? For example a 750 Ti is good or not?

I just want to have an idea about what matters when it comes to MadVR image processing, since I doubt it cares much about 3D performance or GFLOPS...

[flashmozzg]

Anyone knows what kind of "features" are important on MadVR videocard-wise, considering that decoding is done not in hardware?

I mean, what one should look at to consider scaling performance and presentation? For example a 750 Ti is good or not?

I just want to have an idea about what matters when it comes to MadVR image processing, since I doubt it cares much about 3D performance or GFLOPS...

750 Ti or amd in that price range is enough for everything but openCL features.

[Asmodian]

I just want to have an idea about what matters when it comes to MadVR image processing, since I doubt it cares much about 3D performance or GFLOPS...

MadVR wants exactly the same thing as games, it loves 3D performance or GFLOPS. It also likes memory bandwidth though above some level more memory bandwidth will not help. I am not sure what this level is now but I believe you want at least 80 GB/sec. Happily the 750 Ti comes in at 86.4 GB/sec.

I can max out the pure GPU power of a Titan with madVR though it takes pretty high NNEDI3 neurons, high resolutions, and/or frame rates.

[Fadeout]

Well, I'm not interested in NNEDI stuff, but the rest is relevant.

Also, I'm on PCI-Ex 2.0

[TheLion]

Today Dynamic Super Resolution hass been announced as new feature for nVidia 2nd generation Maxwell (GTX980/970) -> http://videocardz.com/52362/only-at-vc-nvidia-geforce-gtx-980-final-specifications

Details should be available in a few days. When it is what the name suggests it is this is a major announcement. Think of the concept as the ultimate "sharpness/detail enhancement filter". Depending on the quality of its implementation and whether madVR can also access this feature it may even - the hell freezes - outwheigh all advantages madVR provides and make EVR the preferred choice for nVidia (assuming this feature can only be enabled using EVR).

This could get very interesting.

[nevcairiel]

Scaling is hardly everything madVR offers, even if EVR had better scaling on account of NVIDIAs driver, madVR would still be superior.

Also, the scaling EVR performs can be accessed by using "DXVA" scaling in madVR. :)

[TheLion]

Scaling is hardly everything madVR offers, even if EVR had better scaling on account of NVIDIAs driver, madVR would still be superior.

Also, the scaling EVR performs can be accessed by using "DXVA" scaling in madVR. :)

Sure. But as far as I understand it this is not mere scaling. It is a temporal filter, analyzing multiple frames in order to "add additional detail" to each frame. Detail that is hidden by the given resolution but can be revealed through interpreting multiple (moving) frames. Interesting concept but very hard to implement without serious artifacting. A simplified example: what happens during a static shot where there is no additional information among several frames?

[aufkrawall]

I can max out the pure GPU power of a Titan with madVR though it takes pretty high NNEDI3 neurons, high resolutions, and/or frame rates.
Compute Error Diffusion is also a killer for my 780 Ti (in combination with NNEDI3).
Not a drama, I haven't encountered yet any advantage in real world with it over ordered dithering.

I suppose GM204 will be much better with Error Diffusion and NNEDI3 though.

[GTPVHD]

http://compubench.com/compare.jsp?benchmark=compu20&D1=NVIDIA+GeForce+GTX+980&os1=Windows&api1=cl&D2=NVIDIA+GeForce+GTX+TITAN+Black&cols=2

Would like to see how Maxwell second generation performs with madvr.

[nevcairiel]

http://compubench.com/compare.jsp?benchmark=compu20&D1=NVIDIA+GeForce+GTX+980&os1=Windows&api1=cl&D2=NVIDIA+GeForce+GTX+TITAN+Black&cols=2

Would like to see how Maxwell second generation performs with madvr.

The page seems weird, the 780 Ti performs better in many of the tests as the Titan Black.

[leeperry]

Would like to see how Maxwell second generation performs with madvr.
If AMD boards are stuck with the 13.12 drivers forever, we won't have much choice anyway :sly:

[Seedmanc]

That reminds me of that SR() proprietary filter, also named temporal super-resolution. But the key word is "temporal", it will do nothing on static scenes, unlike NNEDI which does not depend on motion (although SR does give higher detalization, or at least more "nice-looking" noise in textured non-edge areas). Edges are still done better by NNEDI.
Now if only it was possible to combine the powers of two scalers, temporal and spatial...

[DragonQ]

That reminds me of that SR() proprietary filter, also named temporal super-resolution. But the key word is "temporal", it will do nothing on static scenes, unlike NNEDI which does not depend on motion (although SR does give higher detalization, or at least more "nice-looking" noise in textured non-edge areas). Edges are still done better by NNEDI.
Now if only it was possible to combine the powers of two scalers, temporal and spatial...
I'm still waiting for the day that we can solve crimes by upscaling the reflection in a screw in a photo.

[tuco76]

I'm quite fresh on using MadVR, first off big thanks @ Madshi!
This project and LAV is more than appreciated! :)
On my rather limited gear (check below), I'm trying to find a proper setup providing smooth playback.
For 23fps content, I tried switching my Samsung panel to display mode 24p. Which looked actually worse than 60Hz, I'm assuming it's no "real" 24p, might be internally down-converting 60Hz :sly: !?

So far I get the best results disabling DXVA in LAV options, enabling Smooth Playback in MadVR:

http://s1.directupload.net/images/140917/4vcd5rg2.png (http://www.directupload.net)

Scaling/other options set to default, playback looks really smooth on Madshi's test patterns. But I still get presentation glitches, 1-2 each 10-20 sec. on high bitrate AVC content, ~25MBps. No dropped frames, though.

Are those persistent glitches nothing to worry about, as long as I do not notice them?
Another question, concerning my AMD-CCC settings:
I've disabled all filtering/enhancements etc., should I left "Force Smooth Video Playback" unchecked, too? Does that interfere with MadVR somehow?
Couldn't find reliable info about this option, seems it shuts (GPU-internal?) filtering functions off, if necessary, to avoid dropped frames..!?

Any input or recommendations greatly appreciated :)


[Hardware: Intel Core i3 330M Dual @ 2.13GHz, 4GB RAM, Intel Chipset Havend./Clarkd.02,
AMD Mobility HD 5470
(512MB, DX 11, Pix.Fill: 3.0, Tex.Fil: 6.0, Driver vers. 14.200.)
HDMI to Samsung 46° @ PC-Mode 60Hz]

[tuco76]

A simplified example: what happens during a static shot where there is no additional information among several frames?

As it says "Dynamic", I figure it might utilize elements/methods of static Super Resoloution processing? Sort of repeating the applied SR with xth-frame as reference per each temporal point.
Found that basically explained in this publication's introduction:

http://www.cs.technion.ac.il/~elad/publications/journals/2004/31_DynamicSR_EURASIP.pdf

As DSR claims "4k quality on a 1080p display", I could imagine it delivers improvements mainly on e.g.1080p content created from a 4K source.
Anyway, all stupid speculation yet, let's wait up what's the big deal :)

[QBhd]

@ tuco76:

A couple of things... Why the odd target rectangle? I would disable the DXVA and OpenCL check boxes and enable the "Disable desktop composition" and enable "Fullscreen Exclusive"... As for CCC, I have everything unchecked in the "Video" section.

QB

[ryrynz]

http://compubench.com/compare.jsp?benchmark=compu20&D1=NVIDIA+GeForce+GTX+980&os1=Windows&api1=cl&D2=NVIDIA+GeForce+GTX+TITAN+Black&cols=2

Would like to see how Maxwell second generation performs with madvr.

I'd be interested to see how well it's Dynamic Super Resolution feature works in comparison to madVR's downscaling.

[Xaurus]

The low power draw from the new GTX 970 & GTX 980 is simply fantastic for a HTPC: http://www.techpowerup.com/reviews/ASUS/GTX_970_STRIX_OC/23.html

[tuco76]

@ tuco76:

A couple of things... Why the odd target rectangle? I would disable the DXVA and OpenCL check boxes and enable the "Disable desktop composition" and enable "Fullscreen Exclusive"... As for CCC, I have everything unchecked in the "Video" section.

QB

Thanks for looking into QB!

That odd TR, I just downsized the playerwindow and attached Mad settings to get both into snipping tool.
(fullscreen = boring 1920x1080 :) )

FE Mode was actually the first option I disabled at my very first run, caused trouble. [That was using PotPlayer, maybe I had its own excl. mode enabled on top, or whatev.]
Using MPC-HC by now, tried again (clean forgot about that option), not a single glitch anymore, thanks man :)

For the last three Options, I could hardly notice any differences, so I relied on what that guide recommended:
(THX @ the author BTW)

https://imouto.my/madvr/

The 'use OpenCL to process DXVA NV12 surfaces (Intel, AMD)' should only be enabled if you use GPUs from the two companies, and do not use DXVA renderless video decoding method.
Disable 'use a separate device for presentation (Vista and newer)' option. Enable this though if you want to use the 'smooth motion frame rate conversion' feature below.
Enable 'use a separate device for DXVA processing (Vista and newer)' option.

Admittedly, I don't fully understand yet what those options actually do / effect.
Any feedback/ comments still appreciated, anyway :)

@ all MPC-HC users: Is there an option to keep timeline/tooltip in Full.Excl.Mode I haven't discovered yet?

[michkrol]

FE Mode was actually the first option I disabled at my very first run, caused trouble. [That was using PotPlayer, maybe I had its own excl. mode enabled on top, or whatev.]

Potplayer does have some kind of override for MadVR's settings if I remember correctly.
https://imouto.my/madvr/
This guide is not really good from what I can see.
Some of the recommendations are just wrong.
I know it's not easy to write a good guide, but the author should do some research before recommending every single setting as he likes it to have.
I especially don't understand why the profiles are recommended like that - one rule based on resolution, others based on filenames. It makes no sense whatsoever.

MadVR has good default settings, so that's a good place to start. Since many settings are based on your hardware or personal preference, a guide won't really cut it and you will need to experiment yourself.
Some settings are also based on the content you play - here the profiles are good to have, but a lot of work to configure.

One advice: if it works good, be sure to backup your setting and start from there or just leave it alone (for now).
@ all MPC-HC users: Is there an option to keep timeline/tooltip in Full.Excl.Mode I haven't discovered yet?
No, there isn't.
Not sure if AMD's GPUs likes this, but you might try using "fullscreen windowed mode (new path)" (as the OSD calls it) instead. Disable "automatic fullscreen exclusive" and enable "present several frames in advance" in rendering->windowed mode settings. The playback should be stable (no glitches). Some GPUs like it others don't so check the OSD after changing settings (and restarting player).

[Fadeout]

So, this 750 Ti isn't working well for me.

I'm looking at steady 5/6ms rendering and less than 2ms present. All queues are rock solid.

But then every 20 seconds or so the rendering time rises up to 28ms, I get a bunch of presentation glitches and then it goes back to normal.

If I go back in the video, it looks like it doesn't happen in the same spot, it just happens at random.

[michkrol]

So, this 750 Ti isn't working well for me.

I'm looking at steady 5/6ms rendering and less than 2ms present. All queues are rock solid.

But then every 20 seconds or so the rendering time rises up to 28ms, I get a bunch of presentation glitches and then it goes back to normal.

If I go back in the video, it looks like it doesn't happen in the same spot, it just happens at random.

I'm on 750ti too, as are a few other users here and don't have any problems.
Please post your config: Windows version, drivers version, player, CPU, RAM, perhaps the GPU manufacturer. Tell us what settings you're using, especially fullscreen exclusive/windowed (new/old path).
Be sure to install the drivers manually from nvidia website. It's best to select "advanced setup" and only install display drivers + physx (if you play games) + audio (if you use hdmi).
I'm using windowed fullscreen (new path), give it a try. You get it with default settings(?) or by selecting "display several frames in advance" in rendering->windowed mode settings.
Check what queues are empty when you get glitches/drops, a screenshot with OSD shown (ctrl+j) could also help.
Be sure to check your GPU temperatures and core frequency during playback (preferably some longer movie) with GPU-Z or HWiNFO. Also check if you don't get some unexpected CPU activity or run out of RAM during playback.

[Fadeout]

Well, it seems I figured it out.

First, the videocard has a wildly variable clock, so it goes up and down on its own (and without any apparent logic).

But what actually caused the glitches is that I set in the Nvidia options the "maximum pre-rendered frames" to 1 to remove lag in games. But this screws up MadVR apparently.

With that option back to default I now STILL have high spikes in rendering times, but no glitch or dropped frame. So I'm guessing the variable clock is fast enough to not screw it, as long it has some headroom presenting frames.

[tuco76]

I especially don't understand why the profiles are recommended like that - one rule based on resolution, others based on filenames. It makes no sense whatsoever.

Missed this chapter, but yeah, doesn't make much sense. I skipped most of this guide, found some useful infos before here:
http://yabb.jriver.com/interact/index.php?topic=80253.0
When testing, I found my current gear doesn't allow much playing around beyond the (really proper) defaults, anyway.
The most essential tipp I found here at Doom somewhere: Disabling DXVA decoding, saving the horsepower for smooth motion. Actually the very first time I see real smooth motion for 23fps :)

Not sure if AMD's GPUs likes this, but you might try using "fullscreen windowed mode (new path)" (as the OSD calls it) instead. Disable "automatic fullscreen exclusive" and enable "present several frames in advance" in rendering->windowed mode settings. The playback should be stable (no glitches). Some GPUs like it others don't so check the OSD after changing settings (and restarting player).

Seems she likes it! Render.time slightly increased, yet no glitches!
Thanks mich, awesome having my timeline back :)

[nonowaimas]

Hello; first time here. I have a very specific question for which the FAQs has no answer. Is there any reason Haali downscaling and MadVR Catmul Rom downscaling would look the exact same? They're both rather sharp and the aliasing is kinda evident.

[cyberbeing]

Haali downscaling utilizes hardware Anisotropic interpolation, so how it appears will likely vary based on GPU model. That said, Catmull-Rom is just your standard Bicubic scaler, nothing really exotic about it which should make it stand out unless you've enabled Linear Light or Anti-Ringing. If you want something softer, you could try Mitchell-Netravali instead, but usually high sharpness is desired when downscaling.

[Seedmanc]

Anisotropic interpolation
Could somebody please explain what does it mean in application to video rendering and scaling and how does it work? I know what this option does in 3D games, but here with resizing a flat 2D image I can't quite understand it.

[wolfman2791]

Is Madshi on vacation?

[ryrynz]

Is Madshi on vacation?
He's working on his other projects. If we're lucky we'll see a another version later this year.

[DragonQ]

The low power draw from the new GTX 970 & GTX 980 is simply fantastic for a HTPC: http://www.techpowerup.com/reviews/ASUS/GTX_970_STRIX_OC/23.html

I've been thinking of turning my HTPC into a HTPC/gaming PC hybrid at some point. The main issue for me is the fact that I want my HTPC to be quiet but a gaming PC is usually fairly loud. This type of card could be ideal if it's able to run passively (or at very low RPMs) when playing video but also able to handle games at 60 fps.

[ryrynz]

I've been thinking of turning my HTPC into a HTPC/gaming PC hybrid at some point.

As much as we'll all love the new Maxwell bits we shouldn't derail this thread regarding them :P

[madshi]

Ah, I figured out the bug. When I hit a button to turn on screen in full screen mode (exclusive mode is off) when coming back from the screen turning off (automatically by Windows after x minutes), I get "creating Direct3D device failed (80070005)"
not from even standby, and only happens in full screen and doesn't happen when I just move the mouse
Can you give me a detailed step-by-step guide to reproduce this problem? Which OS and GPU?

Thanks, here's the log.
Looks like your GPU/driver doesn't deliver any vsync scanline information. I'm not sure why. Currently madVR depends on that information, so without vsync scanline information playback won't work. Maybe an option would be to ask a different device for the scanline information, but I'm not sure how to do that. I'd probably need access to an Optimus PC myself, which has these problems, in order to find a way to solve them...

Whenever I play a file in MPC-HC using MadVR I get a black screen with sound playing. I've had this problem a long time now but it's gotten worse and worse, not sure but I think my problems started with Win8. It used to be that I had to restart a few times before it would work or wait maybe 10-15s before it would play but lately no file plays ever. I reinstalled a fresh Win8.1, it was time anyway. Same problem!

However, if I press pause the first frame shows. If I press CTRL-J, play the file again and press pause I can see that it says "Display 0.00000Hz" and vsync interval unknown. If I try EVR Custom it says 0Hz too unless I enable vsync, then it shows the right refresh rate (well nearly).

I've tried all kinds of different settings, drivers and all combinations thereof but I can't figure out what's wrong. I mean it used to work, at least most of the time. I guess MadVR can't detect the display refresh but why? Reclock finds the correct refresh.

I'm using a laptop, with a 60/120Hz display, Win 8.1, AMD Radeon 6650M, Catalyst 14.7 WHQL right now but I've tried a lot of diffrent versions. Tried resetting MadVR to defaults, LAV to defaults etc.
60Hz or 120Hz? 120Hz is more "difficult". It seems that also for you vsync scanline information isn't coming through reliably enough. I don't know why this happens. It seems to work fine for most other users - except for some NVidia Optimus users...

Hi.
Will be possible to add keyboard short cuts for:
-next audio stream
-next subtitles stream
-enable/disable (show/hide) subtitles ?
I'm not sure if madVR can automatically figure out how to show/hide subtitles, but I'll put this on my long term to do list.

I noticed that the madvr binaries are not flagged as dynamic base (ASLR) or nx compatible (DEP).

So which is the culprit here, an ancient Visual Studio or mingw (binutils)?
At least in the case of madVR.ax, it appears to be linked with Visual Studio 2005 defaults (/subsystem 4.0, aslr & nx disabled).

The rest of the madVR binaries are Delphi, and a quick google search suggests it also has both options disabled by default.
^ cyberbeing is correct. Is there any practical benefit (worth mentioning) to have these flags enabled?

I found something that are concered with madTPG and ArgyllCMS. When using dispcal.exe and madTPG to calibration the monitor,the madVR settings "how many video frames shall be presented in advace"could not set more than 10, otherwise it will show the error message with dispcal.exe and finish the work. Windowed mode and FSE mode have the same issue. It works well when setting <= 10.
"The error message"? What error message is that exactly? Does normal video playback work with more than 10 frames?

I've discovered that my TV (Sony 55W905) can only display 4:4:4 chroma in Game or Graphics modes but only properly displays 24fps input without judder in Cinema mode. So I need to make a choice between the two unfortunately.

Am I correct in thinking that if I sacrifice 4:4:4 chroma there's no point in using NNEDI3 for chroma upscaling with madVR as it'll just be blurred by the TV anyway?
I would generally value native 24Hz mode (if it works correctly, double check whether it's equal or better to smooth motion FRC with 60Hz) over 4:4:4 support. However, whether NNEDI3 is still worth it when the display only supports 4:2:2 I can't answer. You'll need to trust your eyes on that one, cause it might also depend on what your display is doing exactly...

I have a ridiculous problem and one of you might be able to help me.
Basically I want to use the APL pattern using madTPG. But, my television, panasonics sloppiest plasma ever made, prevents me from using it. Reason is, that the lines containing the measurement field MUST be black or white. Otherwise they will have an IRE depended colorshift.

Is there a way I can black out the sides of the pattern? An extra application for example?
Hmmmm... Does this problem apply to all Panasonic plasmas? Or many of them? Or just one model?

The rules one can setup in madVR aren't bad at all. I even wonder if the next step might be to open it up a bit more such that madVR could call an external program upon exit that looks at the last file's recorded final stats (same as CTRL+J) and tries to intelligently make up rules based on that and previous user configuration. A start would be to let madVR record these stats per-file in a simple text file, then have someone make a parser that suggests rules. I hesitate to suggest this quality-balancing feature to be added outright to madVR, since I bet it would be an absolute hell to support, partly as it's ultimately up to user preference.
What if the rules are selected too optimistically? Playback would judder as a consequence. The only proper way to intelligently adjust settings would IMHO be if madVR did that itself, based on the state of the render queue. But that's not something I'm planning to look into in the near future.

Oh, and I'd like to query which is better to code for in OpenCL and for video decoding --- Nvidia or AMD?
That depends on whether you need Direct3D9 interop or not. Both have their pros on cons.

Although I understand that the underlying "black/frozen screen" problem with integrated/dedicated gpu machines and madVR may be the fault of an external library, is it possible to at least pinpoint where the problem is rather than ignore it altogether?

Multiple people with 7/8xx GPUs have cited a problem with getting madVR/MPC-HC to render out with the NVidia GPU selected and no one really knows where the problem is occurring. DirectX is backwards compatible with DirectX 9.0 and Windows 8.1 comes with DirectX 11.2 so attempting to install an older version of this (which can't be done anyway) doesn't seem to be the right solution.

Is there any information that I and others experiencing this problem can provide that would aid in pinpointing the problem and closing in on a potential solution? Will remote access to a machine with this problem help?
First of all the problem is not an external library, but most probably the GPU driver. About D3D9: You don't seem to have understood the basics here, so I will ignore that part of your post. Remote access to such a machine could maybe help a little (or maybe not, I'm not sure), but it would only help if I would have remote access whenever I want without having to ask every time, and without any time limits/contraints etc. Much better than remote access would be direct physical access, of course.

There is nothing to indicate either the OS or the gpu drivers. People who've had the black screens in games and MPC have tried the last resort clean OS installs, and it doesn't work. Its not a case of a few users having the black screen, so far reports are that every 800 series user who tries to use their Optimus Nvida gpu (and possibly new ATI switchable GPU systems (http://forum.doom9.org/showthread.php?p=1692218#post1692218) have the issue. 800 users have tried driver versions old and new, and (unlike like the Nvidia OpenCL bug) currently none of the versions tested have resolved the issue. The 800 users are using the same (ex: 340.52 default Nviida) drivers as 600 (and other series) users. If the drivers were the problem people with models other then the 800 would have an issue just like what happened with the Nvidia OpenCL bug.
You do know that drivers often contain different code paths for different GPUs? If the problem only applies to 800 users but not to 600 users then the problem can only be either a driver issue or a hardware issue. Everything else is very unlikely, IMHO. Personally, I believe it's a driver issue.

It's still possible I could find a way to work around it. But that's very hard without having direct access to such a problem PC.

@madshi: Question from Blight. Are mouse-move events no longer filtered in recent versions? They are supposed to be routed through a call-back function.
Nothing should have changed there. If something has changed (unintentionally) please create an issue in the madVR bug tracker, thanks.

i am upgrading my system & decided to get a graphics card.i know that even high end cards struggle with 1080p/i24fps to 1080p60Hz screen but my sources are mostly 480p/720p 23.976fps to be played on 1080p60Hz 32" led tv through hdmi & most of them are animation(japanese anime to be precise).would a card in the league of R7 250 be able to cut it or do i need a more powerful card to ultra-enhance my 480p/720p animation videos using 128 neuron nnedi3 to make it look good?
128 neuron nnedi3 requires a lot of power. If you really need that many neurons, I'm not sure if an R7 250 will be able to do it. You might need something faster. But I don't really know for sure. If you're looking for "buyers guide" kind of information, it might be better to read renethx's posts on AVSForum. He's posted a lot of benchmarks and information on which GPUs can do what with madVR there.

How do I configure madvr with the goal of reducing lag it seems to introduce and at the same time maintain stability? I'm using it to output and upscale video I'm capturing in realtime from PS2 emulator window to watch it on a bigger screen, so minimizing lag is important as I'm playing the game.
Ouch. madVR is optimized for video playback where lag doesn't matter *AT ALL*. The only thing important for video playback is that audio and video are in sync. If you want to reduce lag as much as possible, I'd suggest that you lower the GPU queue size to the lowest possible value and also try disabling the "present several frames in advance" feature. But there will still be some lag. And to be brutally honest, I don't plan to look into making madVR have less lag, because for video playback that's really not important at all.

Another problem is that sometimes with too high settings for madvr or too much processing being applied to my realtime setup (I'm also processing the video in avisynth before output) madvr's framerate severely drops after a while and almost grinds to a halt, only relaunching the player helps, but only for a while. Why does this happen? It's like it clogs instead of just having lower but steady framerate. I tried looking at OSD, there doesn't seem to be any queue overflows.
Not sure what you mean with "queue overflows"? Are all the queues full when that problem occurs? Does the OSD report any frame drops or presentation glitches?

What is the biggest cause for high deinterlace stats? I have a 7850, and deinterlacing is set to vector adaptive. Sometimes max stats (5s) will get into the 20-27 range which causes occasional dropped frames. Movie frame interval is 20.85ms. Is there any settings I can change to lighten the load?
I've no idea. DXVA deinterlacing is kind of a black box. I feed frames in, I get some out, and it takes as long as it takes. Have you tried increasing the GPU queue size? Maybe that could carry you through such spikes? Of course you'll need to have enough GPU RAM for that to work...

If Nvidia is said to do no unnecessary CB, why is CPU usage significantly higher with NNEDI3 scaling than without?
IIRC NVidia sometimes (or always?) pushes the thread/CPU which waits for an OpenCL/CUDA rendering to finish to 100%. I think what NVidia is doing there is an endless loop like "dear operating system, if there's anybody else needing this CPU, give it to them, otherwise I'll continue to use it". This way the CPU goes to 100%, which looks bad in the taskmanager and which also costs power and heats up the CPU, but it actually doesn't slow anything down much, because as soon as the CPU is needed for other tasks, it's available...

Today Dynamic Super Resolution hass been announced as new feature for nVidia 2nd generation Maxwell (GTX980/970) -> http://videocardz.com/52362/only-at-vc-nvidia-geforce-gtx-980-final-specifications

Details should be available in a few days. When it is what the name suggests it is this is a major announcement. Think of the concept as the ultimate "sharpness/detail enhancement filter". Depending on the quality of its implementation and whether madVR can also access this feature it may even - the hell freezes - outwheigh all advantages madVR provides and make EVR the preferred choice for nVidia (assuming this feature can only be enabled using EVR).

This could get very interesting.
Seems like this is just another sort of game-rendering mode, with no consequence to and no benefit for video playback.

For 23fps content, I tried switching my Samsung panel to display mode 24p. Which looked actually worse than 60Hz, I'm assuming it's no "real" 24p, might be internally down-converting 60Hz :sly: !?
That's quite possible, several other users reported similar issues with various TVs, too. It seems some TVs do proper 24fps playback and others don't...

Scaling/other options set to default, playback looks really smooth on Madshi's test patterns. But I still get presentation glitches, 1-2 each 10-20 sec. on high bitrate AVC content, ~25MBps. No dropped frames, though.

Are those persistent glitches nothing to worry about, as long as I do not notice them?
In windowed mode that can happen, although in theory it shouldn't. If you don't see glitches, then it might not be something to lose sleep over. You could try whether increasing the size of the GPU queue and the number of prepresented frames helps. What should definitely help there is enabling fullscreen exclusive mode.

I've been thinking of turning my HTPC into a HTPC/gaming PC hybrid at some point. The main issue for me is the fact that I want my HTPC to be quiet but a gaming PC is usually fairly loud. This type of card could be ideal if it's able to run passively (or at very low RPMs) when playing video but also able to handle games at 60 fps.
The Asus STRIX 970/980 look promising. The fans automatically turn themselves off when the GPU is idle or not doing much. And when the GPU does a lot of work, the fans reportedly are still pretty silent:

http://www.techpowerup.com/reviews/ASUS/GTX_970_STRIX_OC/

[RainyDog]

I've been thinking of turning my HTPC into a HTPC/gaming PC hybrid at some point. The main issue for me is the fact that I want my HTPC to be quiet but a gaming PC is usually fairly loud. This type of card could be ideal if it's able to run passively (or at very low RPMs) when playing video but also able to handle games at 60 fps.

I've just set up my new MSI GTX 970 and had a quick look it how it fairs with madVR.

Good news is that the MSI card stays in passive mode and the fans don't even spin up when using reasonably heavy madVR settings which is great coming off my R9 290 which got quite audible. My typical madVR settings are nnedi3 64 luma doubling, Jinc3 chroma, Jinc3 upscaling and spline36 down scaling, plus I use a bit of LumaSharpen in MPC-BE.

Uses about 40-50w less power than the R9 290 during playback too, though render times are showing about 5ms higher.

Only potential issue is that I can see the 15 and 16 IRE bars on the AVS calibration black level test pattern. I'd read that Nvidia cards can have issues with video levels over HDMI on certain TV's and not allow 0-255 output on capable displays so I was ready to use the FullRGB tweaker tool. Though when I installed the latest drivers black levels did appear to be correct by eye, but looking at the black level pattern bars 15 and 16 are visible so maybe I do need to use the tweaker tool?

[madshi]

With NVidia GPUs and HDMI, always use the madLevelsTweaker, unless you want to use the PC for gaming and desktop/applications, too.

[RainyDog]

With NVidia GPUs and HDMI, always use the madLevelsTweaker, unless you want to use the PC for gaming and desktop/applications, too.

Thanks madshi. I do use my PC for gaming too though, so what negative effects will the levels tweaker have on that?

I've also noticed that using CUVID for decoding in LAV is quite a bit more efficient than either DXVA2 or software. Is there any downsides from an IQ perspective using CUVID?

[nevcairiel]

If your TV supports 0-255/PC Range input, then its fine for gaming.

[Neet009]

Quote:
"The error message"? What error message is that exactly? Does normal video playback work with more than 10 frames?[/url]

It seems ArgyllCMS 1.7 beta has fixed the issue.Now it works well even if I set to 16 frames.:thanks:

[Nui]

Hmmmm... Does this problem apply to all Panasonic plasmas? Or many of them? Or just one model?
Actually its even confined to a certain size range of a specific model of one year and not even all examples. So only very few people could possible profit from an option helping here.

[Plutotype]

I've just set up my new MSI GTX 970 and had a quick look it how it fairs with madVR.

Good news is that the MSI card stays in passive mode and the fans don't even spin up when using reasonably heavy madVR settings which is great coming off my R9 290 which got quite audible. My typical madVR settings are nnedi3 64 luma doubling, Jinc3 chroma, Jinc3 upscaling and spline36 down scaling, plus I use a bit of LumaSharpen in MPC-BE.

Uses about 40-50w less power than the R9 290 during playback too, though render times are showing about 5ms higher.


Thanks for this report.
Did you test the 1080p 24fps upscaling to 2160p please?
Thanks

[RainyDog]

Thanks for this report.
Did you test the 1080p 24fps upscaling to 2160p please?
Thanks

Do you mean the new Dynamic Super Resolution feature? It looks like it's just simple way to down sample for gaming from what I can see.