madVR - high quality video renderer (GPU assisted)

[xabregas]

Hi

I ahev a doubt with MPC subtitle renderer. Every time i use it gives me better PQ than ffsdhow raw and xyvsfilter(xysub is a no option for me because it doesnt let it extend to 16:9). That said ffdshow raw filter with only subtitles enabled changes the color and xyvsfilter do the same for worst, this happens with the same settings in madvr for the 3 solutions... so whats the problem? why im not happy with mpc subtitle renderer??

Its simple, with xyvsfilter and ffdshow i get rendering times of 4ms in averge and max stats, while with mpc subtitle renderer every time subtitles appear the rendering time in max stats rise to 20ms then drops to 4ms again when there is no subs. This is because im not crazy in scaling algorithms chosing.

When i chose nnedi3 in chroma with 32 neurons i can play every video with no frame drops with xyvsfilter and ffdshow but not with mpc subtitle renderer and the worst part is WHY ffdshow and xyvsfilter give me different color, too bright......

What should i do?

TIA

[huhn]

what type of subtitle?

[xabregas]

what type of subtitle?

.srt external

[cyberbeing]

What exactly are you doing with the FFDShow Raw Video Filter & xy-VSFilter?

If you wanted xy-VSFilter subtitle quality similar to XySubFilter & MPC-HC ISR for .srt only, you'd need to:

Resize the video to desktop resolution (for higher resolution subtitles) with FFDShow
Output 4:4:4 AYUV or RGB (improved color, sharpness) with FFDShow.

Doing both would of course would eliminate the primary benefits of using madVR though, which is why XySubFilter and the MPC-HC ISR exist. There shouldn't be anything like 'different color, too bright' unless you have a madVR 3DLUT or color/gamma processing enabled (these apply to xy-VSFilter/XySubFilter but not MPC-HC ISR), or some other misconfiguration in FFDShow or xy-VSFilter.

I can't help you with any madVR performance issues with the MPC-HC ISR though, since the render path is entirely different from XySubFilter. All I know is with the MPC-HC ISR, subtitles are uploaded and processed entirely within the timeframe of madVR's render step and treated as a quasi OSD(?) element. If you are pushing your GPU to the limit with heavy madVR processing like NNEDI3 or there are otherwise delays in the MPC-HC ISR providing subtitles, experiencing spikes in render times are to be expected.

[kasper93]

Hi

I ahev a doubt with MPC subtitle renderer. Every time i use it gives me better PQ than ffsdhow raw and xyvsfilter(xysub is a no option for me because it doesnt let it extend to 16:9). That said ffdshow raw filter with only subtitles enabled changes the color and xyvsfilter do the same for worst, this happens with the same settings in madvr for the 3 solutions... so whats the problem? why im not happy with mpc subtitle renderer??

Its simple, with xyvsfilter and ffdshow i get rendering times of 4ms in averge and max stats, while with mpc subtitle renderer every time subtitles appear the rendering time in max stats rise to 20ms then drops to 4ms again when there is no subs. This is because im not crazy in scaling algorithms chosing.

When i chose nnedi3 in chroma with 32 neurons i can play every video with no frame drops with xyvsfilter and ffdshow but not with mpc subtitle renderer and the worst part is WHY ffdshow and xyvsfilter give me different color, too bright......

What should i do?

TIA

Use latest build from http://nightly.mpc-hc.org/. Subtitle queue was (re)enabled for madVR so you will no longer see any performance spikes. Moreover soon we will be shipping another batch of improvements which should close the gap between ISR and XySubFilter for heavily animated subtitles. Anyway I personally recommend ISR as it is easier to use and out of the box. Unless of course you have problems (report them to http://trac.mpc.org) which XY may resolve. But the choice is yours, pick whatever works better in your work flow.

[MrGlasspoole]

Hi,

after a long time i updated madVR and LAV.
But now, no matter what settings i make in madVR/LAV, deinterlacing does not work.
I can't remember what settings i had in madVR but i'm sure that LAV always was at "aggressive".
I also never used the "deint=" flags in my filenames and it was working.
If i set "force film mode" the dropped frames go away but it still stutters and i see the interlacing.

AMD FirePro V3900
Driver: 13.352.1014
Win8.1 x64

[vivan]

What OSD (Ctrl+J) says about deinterlacing?

[xabregas]

Use latest build from http://nightly.mpc-hc.org/. Subtitle queue was (re)enabled for madVR so you will no longer see any performance spikes. Moreover soon we will be shipping another batch of improvements which should close the gap between ISR and XySubFilter for heavily animated subtitles. Anyway I personally recommend ISR as it is easier to use and out of the box. Unless of course you have problems (report them to http://trac.mpc.org) which XY may resolve. But the choice is yours, pick whatever works better in your work flow.

well it solved the renderer problem but the playback was not as before...


Well, i reverted again to ffdshow raw video filter for the subs, its the best one that can give me stability and less colour change...

now i have a question, ffdshow raw video filter says on tray icon while playing a video that input colour space is yv12 and output is nv12, madvr also says its outputing nv12 with both lav and ffdshow.

i dont understand why the output is nv12 when the input is yv12, why changing?? madvr works better with nv12?

TIA

[huhn]

well it solved the renderer problem but the playback was not as before...


Well, i reverted again to ffdshow raw video filter for the subs, its the best one that can give me stability and less colour change...

now i have a question, ffdshow raw video filter says on tray icon while playing a video that input colour space is yv12 and output is nv12, madvr also says its outputing nv12 with both lav and ffdshow.

i dont understand why the output is nv12 when the input is yv12, why changing?? madvr works better with nv12?

TIA

nv12 or yv12 are both 4:2:0 YCbCr convertaion is lossless.

so looks like ffdshow raw as for yv12 and lavfilter send this to it.
and madVr ask for nv12 so ffdshow convert this to nv12.

[Eino]

What sort of GPU do I need for smooth motion?

I have an Intel Core i7-3770 in my HTPC, but no dedicated graphics card (just the Intel HD4000 on chip graphics). If I enable "don't use linear light for smooth motion frame blending" the video becomes very choppy during pans.

I want to get a fanless GPU to boost the performance, but the fastest passively cooled GPU seems to be the GeForce GT 730. Would that be fast enough to enable the above setting with high bit rate Hi10P 1080p content?

If not, would a GTX 750 do? I've seen articles about a fanless Zotac GeForce GTX 750 Zone Edition, but it's not currently available anywhere. One German retailer seems to say the end of August.

I'd be grateful for any advice on what I should get.

[MrGlasspoole]

What OSD (Ctrl+J) says about deinterlacing?
I did reset all the settings and now its working again. Maybe after some years and a view updates that was needed.

But there is another thing i wonder.
I have one file that is progressive but madVR says and handles it as interlaced.
I thought only interlaced files can be treated wrong?
So how to handle that? Turning of automatic interlacing in madVR and set a flac in every interlaced movie file name?
But what happens then to discs?

The other file that is funny is interlaced and madVR sees it as that, but MediaInfo says it is progressive.
Does that mean that no software can tell you for sure what it is?

[huhn]

What sort of GPU do I need for smooth motion?

I have an Intel Core i7-3770 in my HTPC, but no dedicated graphics card (just the Intel HD4000 on chip graphics). If I enable "don't use linear light for smooth motion frame blending" the video becomes very choppy during pans.

I want to get a fanless GPU to boost the performance, but the fastest passively cooled GPU seems to be the GeForce GT 730. Would that be fast enough to enable the above setting with high bit rate Hi10P 1080p content?

If not, would a GTX 750 do? I've seen articles about a fanless Zotac GeForce GTX 750 Zone Edition, but it's not currently available anywhere. One German retailer seems to say the end of August.

I'd be grateful for any advice on what I should get.
madVR only cares about FPS and how much it has to scale. the rest doesn't matter.

the HD 4000 can do smooth motion without a problem just check the rendertimes and make sure you use smooth motion with the highest possible refresh rate.

there are a lot of "powerful" passive cooled GPUs

these need a case fan:
r9 270 (I got one of these)
gtx 750 ti
gtx 750

these may work without a case fan:
r7 250
r7 240
gt 640

everything below this is normally slow for it's money but should work totally fine with basic scaling and SM.

the gtx 750 ti and 750 are all easily available in germany but you need an active case cooler to be save. the 3770 can be cooled passive I did this my self a year ago.

http://geizhals.at/de/palit-geforce-gtx-750-kalmx-ne5x75000941h-a1140063.html

http://geizhals.at/de/palit-geforce-gtx-750-ti-kalmx-ne5x75t00941h-a1145866.html

if you really need infos about hardware think about a new thread.

[huhn]

I did reset all the settings and now its working again. Maybe after some years and a view updates that was needed.

But there is another thing i wonder.
I have one file that is progressive but madVR says and handles it as interlaced.
I thought only interlaced files can be treated wrong?
So how to handle that? Turning of automatic interlacing in madVR and set a flac in every interlaced movie file name?
But what happens then to discs?

The other file that is funny is interlaced and madVR sees it as that, but MediaInfo says it is progressive.
Does that mean that no software can tell you for sure what it is?

do you have a ~30 sec sample? sounds very very strange

the decoder is giving this information to madVR. just make sure your deinterlacing settings are "if in doubt, deactivate deinterlacing"

[MrGlasspoole]

do you have a ~30 sec sample? sounds very very strange
I made a test. If i change aggressive to auto in LAV then the video is handled progressive by madVR.
I uploaded it:
http://skulltronics.de/uploads/DivX5-720p-Madagascar.avi

Nowhere i can find an answer what the right setting in LAV is in combination with madVR.

[huhn]

I made a test. If i change aggressive to auto in LAV then the video is handled progressive by madVR.
I uploaded it:
http://skulltronics.de/uploads/DivX5-720p-Madagascar.avi

Nowhere i can find an answer what the right setting in LAV is in combination with madVR.

just leave it at auto. in nearly all cases the default setting of lavfilter are right.

the most changed setting that make sense to change is the choice of a hardware decoder or no hardware decoder.

[clsid]

What aggressive deinterlacing does is:
If any frame is flagged as interlaced, then assume that the rest of the video that follows is also interlaced (even if flagged as progressive).
This option is mainly intended for dealing with incorrectly flagged streams. Better leave if at Auto like huhn said.

Those two videos of yours might have some frames that were flagged as interlaced. That could explain the behavior.

MediaInfo only analyzes a small part of the video, so it can't be accurate on mixed content.

[MrGlasspoole]

the most changed setting that make sense to change is the choice of a hardware decoder or no hardware decoder.Hm, i thought its all about using the GPU for decoding. Cause they are optimized for all kind of codecs.
But well, maybe with todays CPUs that can handle Hi10p or 4K-h264...
I never locked at it this way - then madVR can use more from the GPU?
Last time i locked at this stuff CPUs did not have the power to decode VC1, H.264, 1080pMPEG...

But now this: If i set LAV to auto then another file shows the saw tooth edges (deinterlacing not working).
But madVR says "deinterlacing on (says upstream)" :(

[huhn]

Hm, i thought its all about using the GPU for decoding. Cause they are optimized for all kind of codecs.
But well, maybe with todays CPUs that can handle Hi10p or 4K-h264...
I never locked at it this way - then madVR can use more from the GPU?
Last time i locked at this stuff CPUs did not have the power to decode VC1, H.264, 1080pMPEG...

you are misinformed.
no hardware decoder can decode 10 bit no amd GPU can do UHD. UHD is disabled by default in lavfilter.
GPU can do only very little codecs and even if they can do them they only up to a level like h264 4.1 for AMD.

But now this: If i set LAV to auto then another file shows the saw tooth edges (deinterlacing not working).
But madVR says "deinterlacing on (says upstream)" :(

this means madVR is using deinterlacing have you disabled it in the gpu driver?

[MrGlasspoole]

you are misinformed.
no hardware decoder can decode 10 bit no amd GPU can do UHD.
I know that UHD and Hi10p is software decoded.

this means madVR is using deinterlacing have you disabled it in the gpu driver?This is a interlaced file, so if madVR uses deinterlacing then this is how it should be?
GPU driver? You mean Catalyst Control Center? Yes, i tried it with enabled and disabled.

Another thing:
The workload on CPU/GPU is linked to the total pixels of a video - right?
2048×1152 = 2.359.296px
1792×1344 = 2.408.448px
Different width and high but almost the same amount of pixels.
So the workload is the same?

If thats the case it would be easier to make a script based on resolution if it would
be possible to do math in the script.
Something like:
if (srcWidth * srcHeight > 2000000px) "FHD"
Or if there would be "srcPix".

The problem is that there are resolutions that are for example below "1920 width" but have a
higher height. If i have for example a 3D 1920p movie, then the height is 2160...
Also looking for >=1280 and >=720 does not really work cause if the movie is not 16:9 then the height is less.

[GCRaistlin]

in nearly all cases the default setting of lavfilter are right.

Including "RGB Output levels" setting which default value is "PC (0-255)"?

[huhn]

I know that UHD and Hi10p is software decoded.
This is a interlaced file, so if madVR uses deinterlacing then this is how it should be?
GPU driver? You mean Catalyst Control Center? Yes, i tried it with enabled and disabled.
can you try the same file with EVR and check if you still see wave interlacing? if yes think about a bug report.


Another thing:
The workload on CPU/GPU is linked to the total pixels of a video - right?
2048×1152 = 2.359.296px
1792×1344 = 2.408.448px
Different width and high but almost the same amount of pixels.
So the workload is the same?

If thats the case it would be easier to make a script based on resolution if it would
be possible to do math in the script.
Something like:
if (srcWidth * srcHeight > 2000000px) "FHD"
Or if there would be "srcPix".

The problem is that there are resolutions that are for example below "1920 width" but have a
higher height. If i have for example a 3D 1920p movie, then the height is 2160...
Also looking for >=1280 and >=720 does not really work cause if the movie is not 16:9 then the height is less.

for decoding it's the ~kbit even this is simply not 100 % true.

for gpu. a 720p video need a lot more processing on a 1080p screen than a 1080p file on the same screen. it depends on how much scaling is needed and the frame rate.


Including "RGB Output levels" setting which default value is "PC (0-255)"?
useless for madVR. madVR can read all colorspaces lavfilter can output so this not used.

[MrGlasspoole]

can you try the same file with EVR and check if you still see wave interlacing? if yes think about a bug report.
Ok, the file i was talking about last works with "LAV aggressive" in EVR and madVR. With "auto" i have saw tooth in both renderers even
if madVR says "interlacing on".
But if i use aggressive then my other progressive file is recognized as interlaced by madVR.
And then there is one file that works in EVR with aggressive and auto but not in madVR.
So where is the bug - LAV, madVR or is it me?
Behaviour:
Anixe-HD[1080i] EVRCP+LAV agressive = OK
Anixe-HD[1080i] madVR+LAV agressive = OK
Anixe-HD[1080i] EVRCP+LAV auto = saw toth
Anixe-HD[1080i] madVR+LAV auto = saw toth (madVR says interlacing on)

Koreans[1080i] EVRCP+LAV agressive = OK
Koreans[1080i] madVR+LAV agressive = saw toth (madVR says interlacing on)
Koreans[1080i] EVRCP+LAV auto = OK
Koreans[1080i] madVR+LAV auto = saw toth (madVR says interlacing on)

Madagascar[720p] madVR+LAV agressive = madVR says interlacing on
Madagascar[720p] madVR+LAV auto = madVR says interlacing off
And the files:
http://skulltronics.de/uploads/Anixe-HD[1080i].ts
http://skulltronics.de/uploads/Koreans[1080i].tp
http://skulltronics.de/uploads/Madagascar[720p].avi

for gpu. a 720p video need a lot more processing on a 1080p screen than a 1080p
Sure, thats the reason for profiles cause you can run different settings on different resolutions and FPS.
So the question is how to make the script intelligent enough to differentiate SD, HD, FHD...
Thats the last thing i came up with:
if (srcWidth < 1280) and (deintFps <= 30) "SD 24fps"
else if ((srcWidth >= 1280) and (srcWidth <= 1919)) and (deintFps <= 30) "HD 24fps"
else if ((srcWidth >= 1280) and (srcWidth <= 1919)) and ((deintFps >= 31) and (deintFps <= 60)) "HD 60fps"
else if ((srcWidth >= 1920) and (srcWidth <= 2559)) and (deintFps <= 30) "FHD 24fps"
else if ((srcWidth >= 1920) and (srcWidth <= 2559)) and ((deintFps >= 31) and (deintFps <= 60)) "FHD 60fps"
else if ((srcWidth >= 1920) and (srcWidth <= 2559)) and (deintFps >= 61) "FHD 120fps"
else if ((srcWidth >= 2560) and (srcWidth <= 3839)) and (deintFps <= 30) "QHD 24fps"
else if ((srcWidth >= 2560) and (srcWidth <= 3839)) and ((deintFps >= 31) and (deintFps <= 60)) "QHD 60fps"
else if ((srcWidth >= 2560) and (srcWidth <= 3839)) and (deintFps >= 61) "FHD 120fps"
else if (srcWidth >= 3840) and (deintFps <= 30) "UHD 24fps"
else if (srcWidth >= 3840) and ((deintFps >= 31) and (deintFps <= 60)) "UHD 60fps"
else if (srcWidth >= 3840) and (deintFps >= 61) "UHD 120fps"
But that doesn't take into count for example 3D which is double the resolution (ex. 1920x2160).
I was looking on doing it with scrHeight rules. But there are so many variables...
And sure, it is questionable if it is needed to differentiate on all that - but hey, if it can be done and who does use
madVR if he not wants perfection :D

[nevcairiel]

And then there is one file that works in EVR with aggressive and auto but not in madVR.
So where is the bug - LAV, madVR or is it me?


Your files are broken.
If a file needs even aggressive to deinterlace properly, then it is already broken. Any properly encoded file should deinterlace perfectly in Auto mode.

[michkrol]

Sure, thats the reason for profiles cause you can run different settings on different resolutions and FPS.
So the question is how to make the script intelligent enough to differentiate SD, HD, FHD...
(cut)
I was looking on doing it with scrHeight rules. But there are so many variables...
And sure, it is questionable if it is needed to differentiate on all that - but hey, if it can be done and who does use
madVR if he not wants perfection :D

Do you really need all the combinations?
I mean, you use profiles since you want to max out your GPU and image quality, but a lot of the low res combinations will have the same "maximum quality" settings and the high res combinations will have the same "maximum performance" settings with only a few different profiles between them, so it seems pointless to list all possibilities.

Creating profile groups is also something to think over and set different rules for different settings. Chroma upscaling for example IMHO isn't that performance-heavy and I have it set based on resolution only - soft chroma upscaling (to hide artifacts) on SD and sharp chroma on HD and it works, unless you add highly compressed sources, like YouTube into the mix, but then filename/ext rules might come in handy.
My rules are really minimalistic, to give an example with debanding:
if (srcWidth>1920) or (srcHeight>1080) "deband off"
elsif (fileName="*YouTube*") or (srcHeight<600) "deband high"
else "deband low"
And I'm set for debanding. It's off for 4K (low performace with my GPU), high for SD and YouTube (poor source quality) and low for everything else.

As you can see I don't have >= and <= in one rule to make them less complicated and it works the same if the rules are in correct order. I also like to have the profiles named like the settings and not source quality/parameters, so a completely different approach to yours.

So what I recommend is to create 2/3 (or more?) sets of profiles for different settings with the setting on/off or at low/med/high respectively and check the performance with the clips you have at hand - you can always add more rules, but why bother to later realize half of your rules point to exactly the same settings? That's how I've gone about it and I'm quite happy with the results. It did take me a while until I've finally set them and at some point I got so creative that I needed comments to mark what is what :D

[MrGlasspoole]

Your files are broken.
If a file needs even aggressive to deinterlace properly, then it is already broken. Any properly encoded file should deinterlace perfectly in Auto mode.But what about the "Koreans" video that works with EVR but not madVR?

a lot of the low res combinations will have the same "maximum quality" settings and the high res combinations will have the same "maximum performance" settings with only a few different profiles between them, so it seems pointless to list all possibilities.Sure, but there is not much i added. I did 120fps and QHD to play around. But there is definitive a difference between HD < 30fps and HD > 60fps.
I can do "Chroma up Jinc 3 (AR)" below 30fps but only "Lanczos 3 (AR)" above 30fps.

Creating profile groups is also something to think over and set different rules for different settings.What do you mean? Isn't that not what we already talk about?

filename/ext rules might come in handy. My rules are really minimalistic, to give an example with debandingI'm not a big fan of renaming my files.
First you need to go through 100th of movies on your server.
Second i like clean names "Jaws (1975).mkv" and not "Jaws (1975)[some][else].mkv"
Third it does not work with iso's and disc's.

Debanding is the cue. Is it normal that with low the rendering time is higher then with high.
I also thought i use high only on SD and it takes more resources on FHD stuff.
But if i set debanding to high on movies (no matter if SD, HD, FHD) i always have less rending time.

And another question. I see people using Catmull-Rom+AR+LL most of the time for downscaling.
Why not Bicubic 60+AR? Sharper, less aliasing, same ringing...

I also think that SoftCubic 50/60+AR works better for SD (especially my old cartoon DVDs).
But most of the time in tutorials its all about Jinc.

Sorry for maybe stupid questions that are maybe already answered somewhere.
But this thread is so long and often i don't find an answer.

[Novalis]

Hi,

I have installed latest mpc x86 and madvr. Everything seems all right, all filters run. But if i play movie only sound plays, no video.
I have a win7 x64 with internal nvidia gpu and Pci Nvidia GT 610. is that a problem? Can't deactivate the internal gpu, allthough I dont use it. Windows wont start up properly.

Thx,
Novalis

[huhn]

@MrGlasspoole

they all work totally fine. in video mode not film mode.
looks like you force film mode on interlaced content. film mode is for telecine content.

[michkrol]

What do you mean? Isn't that not what we already talk about?
What I wanted to say is create separate set of rules for chroma upscaling, separate set of rules for image upscaling, etc. to reduce complexity, since you probably only need like 2 rules for chroma, but much more for image upscaling.
Debanding is the cue. Is it normal that with low the rendering time is higher then with high.
I also thought i use high only on SD and it takes more resources on FHD stuff.
But if i set debanding to high on movies (no matter if SD, HD, FHD) i always have less rending time.
Low setting should affect the image only slightly to avoid detail loss. To achieve it more checks are performed (computed), hence you get higher render times.
Or your GPU might be entering power-saving mode, reducing clock speeds, hence increasing render times.
I also think that SoftCubic 50/60+AR works better for SD (especially my old cartoon DVDs).
But most of the time in tutorials its all about Jinc.
The tutorials aren't always (any) good. Jinc is the algorithm of choice if you don't use profiles.
SoftCubic is nice for "old" sources, but be sure to check out the image doubling (NNEDI3) which is pure magic for low resolution sources, especially cartoons.

Hi,

I have installed latest mpc x86 and madvr. Everything seems all right, all filters run. But if i play movie only sound plays, no video.
I have a win7 x64 with internal nvidia gpu and Pci Nvidia GT 610. is that a problem? Can't deactivate the internal gpu, allthough I dont use it. Windows wont start up properly.

Thx,
Novalis
Do you have all the DirectX updates installed?
Get DirectX End-User Runtime Web Installer from Microsoft here (http://www.microsoft.com/en-US/download/details.aspx?id=35).

By no video do you mean black window or only a small seekbar, like you get when playing an audio file? Is madVR in the filters list in View->Filters menu?
Can you see madVR's OSD (press CTRL+J to show/hide)?
If yes, OSD's screenshot will help.

Does it work with EVR or any other video renderer?
If not, please move to correct/create new thread.

[Novalis]

Do you have all the DirectX updates installed?
Get DirectX End-User Runtime Web Installer from Microsoft here (http://www.microsoft.com/en-US/download/details.aspx?id=35).

By no video do you mean black window or only a small seekbar, like you get when playing an audio file? Is madVR in the filters list in View->Filters menu?
Can you see madVR's OSD (press CTRL+J to show/hide)?
If yes, OSD's screenshot will help.

Does it work with EVR or any other video renderer?
If not, please move to correct/create new thread.

It had something to do with xysubfilter. After uninstalling them i got my video.
But now it's even worse. it runs a couple of seconds and then the complete system crashes an reboots. audio sounds ugly, choppy, thats all I can register before crash. That whole thing seems pretty unstable to me. Never had such a bad experience with other filters/decoders.

[MrGlasspoole]

they all work totally fine. in video mode not film mode.
looks like you force film mode on interlaced content. film mode is for telecine content.
Ok, this drives me crazy.
I did set [deint=video] on the "Koreans" and it worked. Then i removed/disabled video mode and now it still works :angry:
But the beach babes don't work - they only workout :-)
http://forum.doom9.org/attachment.php?attachmentid=14339&stc=1&d=1409344827

What I wanted to say is create separate set of rules for chroma upscaling, separate set of rules for image upscaling, etc. to reduce complexity, since you probably only need like 2 rules for chroma, but much more for image upscaling.Ok, now i know what you mean.
Would be nice if things like artifact removal and dithering where in one place with luma/chroma.
The way it is now you need to make extra profiles in the categories where they are.

Low setting should affect the image only slightly to avoid detail loss. To achieve it more checks are performed (computed), hence you get higher render times.
Ah ok. Makes sense.

What about that Catmull-Rom+AR+LL vs Bicubic 60+AR? Why is Catmull preferred?

Never had such a bad experience with other filters/decoders.Did you had codec packs on your system, did you experiment allot with other filters/codecs before on that system.

[huhn]

It had something to do with xysubfilter. After uninstalling them i got my video.
But now it's even worse. it runs a couple of seconds and then the complete system crashes an reboots. audio sounds ugly, choppy, thats all I can register before crash. That whole thing seems pretty unstable to me. Never had such a bad experience with other filters/decoders.

sound like over heating.

post your hole system a 710 have huge problem with madVR the processing power of this video card is about "nothing".

[Anime Viewer]

It had something to do with xysubfilter. After uninstalling them i got my video.
But now it's even worse. it runs a couple of seconds and then the complete system crashes an reboots. audio sounds ugly, choppy, thats all I can register before crash. That whole thing seems pretty unstable to me. Never had such a bad experience with other filters/decoders.

Have you scanned your system with any codec tweak tools to see if you have broken codec, or mis-registered filters? Since more people haven't reported the issue with xysubfilter your system may have other issues that caused the problem with xysubfilter.

[Novalis]

Have you scanned your system with any codec tweak tools to see if you have broken codec, or mis-registered filters? Since more people haven't reported the issue with xysubfilter your system may have other issues that caused the problem with xysubfilter.

My system is running fine; playing everything. Just this mad thing's running mad. Got rid of it. never mind.

Thanks.

[XMonarchY]

What are the minimum specs for a PC to smoothly run 720pp-1080p films with madVR and the following settings without frame drops:

1 - Jinc3tap chroma upscaling + Jin3tap (A) image upscaling (no doubling/quadrupling) + Catmull-Rom (AA) image downscaling + ordered dithering + mild to moderate artifact removal + (possibly) smooth motion???

2. NNEDI3 32neuron chroma upscaling, NNEDI3 32neuron luma doubling, Jin3Tap (A) image upscaling + Catmull-Rom image downscaling (AA) + OpenCL Error Diffusion + mild to moderate artifact removal + (possibly) smooth motion???

I work at a PC repair shop and some people are interested in HT computers that can provide a good picture quality at a relatively low cost. The 2 sets of settings are for high quality and very high quality image, but I have no clue what kind of minimum specs are needed for either #1 or #2. I have 3770K @ 4.8Ghz and GTX 780 Ti, but that is FAR from minimal AFAIK. Anyone have an approximation?

[clsid]

Several people have reported compatibility problems with Nvidia Optimus systems and the latest version of madVR. It is most likely some driver issue.

Does anyone here have an Optimus system that does work? If so, which Nvidia GPU and which driver do you use?

A Nvidia GT 610 is a very weak card. I isn't powerful enough to run madVR with settings high enough to make it worthwhile.

[*Touche*]

What are the minimum specs for a PC to smoothly run 720pp-1080p films with madVR and the following settings without frame drops:

1 - Jinc3tap chroma upscaling + Jin3tap (A) image upscaling (no doubling/quadrupling) + Catmull-Rom (AA) image downscaling + ordered dithering + mild to moderate artifact removal + (possibly) smooth motion???


I run this with A10-5800K for 720p24->1080.

[vivan]

Does anyone here have an Optimus system that does work? If so, which Nvidia GPU and which driver do you use?540M, 337.88
I don't use it often since it's too slow - but it does work without any issues...

[MrGlasspoole]

What are the minimum specs for a PC to smoothly run 720pp-1080p films with madVRFor #1 i would say Celeron G1840 + R9 270X

But if a R9 270X is enough for NNEDI3 i know in a view weeks after i build the new PC.

[Anime Viewer]

Several people have reported compatibility problems with Nvidia Optimus systems and the latest version of madVR. It is most likely some driver issue.

Does anyone here have an Optimus system that does work? If so, which Nvidia GPU and which driver do you use?


From reports the only optimus systems with problems running it are the 800 series. The problem appears to be VMR9/DirectX related. People with the 800 series can't seem to run with the Nvidia GPU selected in Nvidia control panel without a black screen unless they select VMR-7 (Video Mixing Render) along with System Default in MPC Playback -> Output area. People with the 800 series have also been experiencing the same black screen situation in some games. Rumor is that the VMR9 issue has do with how it renders the screen rectangles and may be effected by anti-aliasing settings.As a result there are rumors that turning off anti-aliasing may work as a work around for the issue, but not enough people have confirmed if that is really the case.

For the record I have a GTX 680M (as listed in my signature), and don't have the problem. I don't believe its right to blame optimus for the problem when there is no proof that "optimus" is the cause of the problem as opposed to something with VMR/DirectX. The Nvidia drivers likely aren't the problem as us 600 series users are running the same model drivers as the 800 series people.

[RainyDog]

So it's been a while since I used madVR and I've been experimenting with the newer settings and changes over the past couple of days.

I've been having an issue where MPC-BE is more often than not freezing when leaving FSE and the only way I can seem to fix it is disabling 'present several frames in advance' which also disables DXVA acceleration. In doing this, I've noticed that my PC draws about 80w less power using the same madVR settings which I'm naturally quite happy with as it easily tops 200w when using NNEDI3 at 32/64 neurons in combination with Jinc.

It probably shouldn't be surprising as my GPU is a power hungry Radeon 290 and CPU is a still pretty efficient and beefy i7 3770k. But I just wanted to make sure there's no actual quality difference between using DXVA2 and software for rendering? Nor any quality impact by disabling 'present several frames in advance' in madVR?

Thanks.

[huhn]

So it's been a while since I used madVR and I've been experimenting with the newer settings and changes over the past couple of days.

I've been having an issue where MPC-BE is more often than not freezing when leaving FSE and the only way I can seem to fix it is disabling 'present several frames in advance' which also disables DXVA acceleration. In doing this, I've noticed that my PC draws about 80w less power using the same madVR settings which I'm naturally quite happy with as it easily tops 200w when using NNEDI3 at 32/64 neurons in combination with Jinc.

It probably shouldn't be surprising as my GPU is a power hungry Radeon 290 and CPU is a still pretty efficient and beefy i7 3770k. But I just wanted to make sure there's no actual quality difference between using DXVA2 and software for rendering? Nor any quality impact by disabling 'present several frames in advance' in madVR?

Thanks.

try software and present in advanced first. or disable FSE mode.
software has can better deal with error in the stream but there is no difference in BD playback.

i think you got so high power consumption thx to dxva. dxva forces your gpu in at least mid powerstate and if you are not using nnedi for chroma this is not needed for 1080p on 1080p. so software takes a lot less power then dxva with high performance cards.

[StinDaWg]

And another question. I see people using Catmull-Rom+AR+LL most of the time for downscaling.
Why not Bicubic 60+AR? Sharper, less aliasing, same ringing...
I use Lanczos4 for downloading high quality 720p videos after NNEDI3. If you want a sharper algorithm use Lanczos. Bicubic is a lower quality algorithm.

I don't use AR for downscaling because there is a bug with madVR that causes aliased text with NNEDI3.

http://forum.doom9.org/showthread.php?t=170769

[Anime Viewer]

But I just wanted to make sure there's no actual quality difference between using DXVA2 and software for rendering? Nor any quality impact by disabling 'present several frames in advance' in madVR?

Thanks.

More than image quality those two things are related to speed of processing images. In some cases render times can be lower while using DXVA2, but with some less used forms of video pixelating may appear when using DXVA2 (same for CUVID and other hardware accelerators with Nvidida GPU). The present several frames in advance can some times decrease render times, eliminate dropped frames, or eliminate presentation glitches.

If you're getting decent render times with those disabled, and you aren't seeing any dropped frames or presentation glitches then you might as well leave them both disabled.

[Asmodian]

I use Lanczos4 for downloading high quality 720p videos after NNEDI3. If you want a sharper algorithm use Lanczos. Bicubic is a lower quality algorithm.

I don't use AR for downscaling because there is a bug with madVR that causes aliased text with NNEDI3.

http://forum.doom9.org/showthread.php?t=170769

That thread isn't about a bug with AR for downscaling causing aliased text with NNEDI3.

That is an issue caused by using NNEDI3 doubling for luma but not for chroma which causes chroma misalignment artifacts which are then exaggerated by AR and/or LL downscaling with the sharper downscaling options. Use chroma doubling too and they go away completely.

Catmul-Rom+AR+LL is good because it has the lest artifacts and you get the benefits of LL downscaling.

Edit: Chroma misalignment probably isn't the right way to describe it, if the luma is sharper than the chroma that means the chroma has been blurred more so it bleeds further into the surrounding pixels but it is still perfectly aligned with the luma.

[RainyDog]

Well, I figured out why I was getting the large power usage discrepancy between using DXVA and software - NNEDI3 wasn't working without DXVA. I did think that using 64 neurons for chroma upscaling and luma doubling was coming a bit too cheaply, actually :p Render times and power draw were the same with any NNEDI3 settings enabled or disabled.

So went back to DXVA2 and the image quality increase was immediately apparent using NNEDI3, as was the power draw and render time differences. Using the latest dev build of MPC-BE seems to have fixed the leaving FSE mode freezing from my brief testing last night too so fingers crossed.

On another note, I updated to the latest catalyst 14.8 WHQL drivers yesterday and power draw, heat and noise went up considerably on my R9 290. PC was pulling 250w using the same settings that only required about 180w on the 14.4 drivers! Think I'm gonna revert right back to 13.12 though as I've read further back through this thread that most folk still recommend 13.12 as the most efficient for madVR.

[octal9]

Well, I figured out why I was getting the large power usage discrepancy between using DXVA and software - NNEDI3 wasn't working without DXVA. I did think that using 64 neurons for chroma upscaling and luma doubling was coming a bit too cheaply, actually :p Render times and power draw were the same with any NNEDI3 settings enabled or disabled.

So went back to DXVA2 and the image quality increase was immediately apparent using NNEDI3, as was the power draw and render time differences. Using the latest dev build of MPC-BE seems to have fixed the leaving FSE mode freezing from my brief testing last night too so fingers crossed.

On another note, I updated to the latest catalyst 14.8 WHQL drivers yesterday and power draw, heat and noise went up considerably on my R9 290. PC was pulling 250w using the same settings that only required about 180w on the 14.4 drivers! Think I'm gonna revert right back to 13.12 though as I've read further back through this thread that most folk still recommend 13.12 as the most efficient for madVR.
i wouldn't advise reverting to 13.12 - when i reverted back to those drivers, nnedi 13 was not working at all for me - the best drivers i've found so far are the 14.7 beta drivers - my 6870 can do very high settings on almost any video and nnedi is working nicely (albeit on 32 neurons luma only).......

[QBhd]

Well, I have tried every driver since 13.12 and each time there has been a performance hit and I rolled back. I push my 270x to the max, so it's very easy to see this hit in performance, since I will immediately get dropped frames. I know every system is different, therefore it's always best to test for yourself, I'm just sharing my experience :)

QB

[RainyDog]

Well, I have tried every driver since 13.12 and each time there has been a performance hit and I rolled back. I push my 270x to the max, so it's very easy to see this hit in performance, since I will immediately get dropped frames. I know every system is different, therefore it's always best to test for yourself, I'm just sharing my experience :)

QB

I decided to roll back to 13.12 since I watch films more than game these days and I'm sticking with them as well. Render times are less in madVR, and GPU power draw and heat are greatly reduced from 14.8. AMD have definitely messed something up with the last few driver releases in that regard.

[RainyDog]

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?

[Megalith]

Wait, so the effects of madVR are only active with certain hardware acceleration options like DXVA2?