madVR - high quality video renderer (GPU assisted)

[huhn]

Luma quadrupling yes. But chroma doubling/quadrupling? I'd like to see a comparison screenshot.

Does the test build fix the refresh rate problem for you?

i wasn't on the right PC haven't even installed that.

i go install 87.21 do some screens (found the video i did these chroma doubling/quadrupling test month ago) and than test your new 88 build.

nedi totally fails for pixel art by the way.

[madshi]

How'd I use chroma doubling with 0.87? Testing with 4:2:2 content? If yes, this is way too much effort.
The source format doesn't matter. Any source will do, 4:2:0, 4:2:2, 4:4:4 or even RGB. We're not talking about 4:2:0 -> 4:4:4 chroma upsampling here, but about "doubling" the chroma channels when the luma channel is doubled with NNEDI3 at the same time. With v0.87.21 you can double chroma in the "image doubling" settings page. Or you can switch "image upscaling" to Jinc3 AR, then that will be used for doubling chroma, as well, if NNEDI3 chroma doubling is disabled.

Yes, but afaik, C-R is hardly better in any situation compared to Jinc3.
Jinc3 AR does cost a lot more performance than Catmull-Rom AR. The difference should be noticeable even on a fast GPU. It might not be a night-and-day difference, but even saving 2-3ms can help if you want to use other algorithms, as well.

[madshi]

My question is, although I know my gpu is literally at its absolute limits, how come some of the queues seem full, while others are nearly empty all the time, yet my gpu isn't under enough pressure to drop frames. Absolutely very low priority question, just thought it was worth mentioning as I'm not sure if the render and present queues not being full in this scenario may be something madVR is controlling and could potentially be improved to have them more full seeing as frames aren't being dropped.
I don't know. If your GPU is fast enough, the queues should be full. If it's not fast enough, you should be dropping frames. I don't know why your queues would be near empty, but playback is still smooth. Make sure in your NVidia control panel you have the max number of pre-rendered frames set to application controlled.

i go install 87.21 do some screens (found the video i did these chroma doubling/quadrupling test month ago) and than test your new 88 build.

nedi totally fails for pixel art by the way.
Thanks.

NEDI works ok for pixel art, when combined with SuperRes. Without SuperRes it totally fails, that's true.

[mark0077]

I don't know. If your GPU is fast enough, the queues should be full. If it's not fast enough, you should be dropping frames. I don't know why your queues would be near empty, but playback is still smooth. Make sure in your NVidia control panel you have the max number of pre-rendered frames set to application controlled.


OK cool, when I go down to 16 neurons all queues fill up again, just thought it might be an algorithm in madVR perhaps to balance the queues based on some priority, but I guess not.

[XMonarchY]

1) With v0.87.21, what happens if you disable Chroma doubling and Chroma quadrupling (right side of the image doubling page)? Do you then get similar results to v0.88.5?


With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
- Rendering times are about 20-22ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (BOTH LUMA AND CHROMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (ONLY LUMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.
- I do not get the same results with build 88.5. Please see below.



2) What are your rendering times with: a) v0.87.21 with your original settings. b) v0.87.21 with disabled Chroma doubling/quadrupling. c) v0.88.5?


A) With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
Original Settings
- Rendering times are about 20-22ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (BOTH LUMA AND CHROMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.

B) With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
Chroma Doubling + Quadrupling disabled
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (ONLY LUMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.

C) With build 88.5 (ALL new features turned off, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image Doubling + Quadrupling to NNEDI3 32 neurons. No frame drop this time!

I even went back and set Image Upscaling to Jinc 3 tapm, enabled the Shiandow's Deband Algorithm, and enabled SuperRes filter for Chroma Upscaling and it produced no frame drops! Quadrupling issue is SOLVED with 88.5! . Using SuperRes fot Image Refinment in addition to settings above resulted in frame drops, but as you said - those settings are only for testing.

[AngelGraves13]

Will you be adding 60 frames to render ahead for Windowed and Fullscreen now that D3D11 can sync to the VSYNC rate? Or am I misunderstanding the whole D3D11 mode that was recently added?

I've found setting Windowed and Fullscreen to 16 and GPU que to 16 with CPU que at 20 provides very low ms (10-20 ms) playback.

[detmek]

Ouch. I'm sorry that you spent so much time with different OSs just to find out it didn't help at all. Unfortunately I don't know what to do. I can't reproduce the problem you have on my PC. And it seems nobody else has the same problem which you have. I've no clue why.

If it would be very easy for you, maybe you could create a short digicam video, showing the problem you're seeing? But don't bother if it would be hard for you to do.



No problem. Like I said, I wanted to try Windows 10. I guess I should tried fresh install as Win 8.1 > Win 10 didn't work for me.

Anyway, I created 2 clips, one with Bandicam and one with my compact digicam.
http://dc180.4shared.com/img/-aBBKUPvce/14d4a449088/madVR_stutter (http://www.4shared.com/video/-aBBKUPvce/madVR_stutter.html)
P.S. That half a second black screen is because I restarted playback with Ctrl + E.
http://dc180.4shared.com/img/bzvo2yg6ba/14d4a51c758/madVR_stutter_1 (http://www.4shared.com/video/bzvo2yg6ba/madVR_stutter_1.html)

I also noticed that Render Queue is not full in D3D11 mode although rendering times are low but I can't remember if I saw that under Windows 8.1.

No big deal. I just wanted to try D3D11 for better performance. I don't need 10-bit output or so. :thanks: Madshi

[agustin9]

I've done some digging and hacking and raping Direct3D, I think I have the refresh rate problems sorted out. Give this build a try:

http://madshi.net/madVR885refreshRateFix.zip

Should work for D3D9 and D3D11 now, and also for 64bit. Once again, I've spent a lot of time to fix Microsoft's stupidity. They actually *intentionally* use 59.940Hz (23.976Hz) even if you ask for 60.000Hz (24.000Hz). There's no way to convince Direct3D to use 60.000 (24.000), except by going behind their back and twisting and binding their fingers, like I've done. I wonder if there's any other software besides madVR which can switch FSE mode to 24.000/60.000? I kinda doubt it...
With this build it switches to 59.93, and leave the screen in that refresh rate even after I go back to windowed. Previously when going back to windowed it switched to 59.99. Windows 8.1 here

[huhn]

NEDI works ok for pixel art, when combined with SuperRes. Without SuperRes it totally fails, that's true.

ok quadrupling is not that huge sorry but placebo? not sure.chroma doubling is quiet something!

chroma upscale
jinc3AR: http://abload.de/img/nnedi3jinc3arx2syl.png
nnedix2: http://abload.de/img/nnedi3chromax2posri.png
nnedix4: http://abload.de/img/nnedi3chromax424swl.png
editv2: catmull-rom AR http://abload.de/img/nnedi3catv2misgl.png

and with some more jinc3ar scaling on top:
jinc3AR: http://abload.de/img/nnedi3jinc3arg2sj3.png
nnedix2: http://abload.de/img/nnedi3chromax2rtsql.png
nnedix4: http://abload.de/img/nnedi3chromax4z7ssj.png

i wouldn't be shock if scaling to UHD would show some more potential for quadrupling on this source. but this source is just ...

edit2:added the right frame

[mark0077]

With this build it switches to 59.93, and leave the screen in that refresh rate even after I go back to windowed. Previously when going back to windowed it switched to 59.99. Windows 8.1 here

Maybe you guys already know about this utility, but its what I have used in windows 7 / 8 to stop my PC setting my TV to 59.9X mode.

http://www.monitortests.com/forum/Thread-Custom-Resolution-Utility-CRU

It lets you set what resolutions your display is allowed to use. Once setup correctly, you can make your display only be allowed use 60hz for example by removing the other ones you don't want to allow. This worked for me with a bit of fiddling.

[pirlouy]

Did it ever work for you, in D3D9, I mean?
No. As i said earlier, I already had this problem with 0.87.21 (no D3D11 if I understood correctly).
I noticed the problem with x64 build since the 64 build switches to 59Hz.
I never noticed the problem with x86 builds before because in fact the x86 build auto switches to 60Hz and I was already in 60Hz.

I don't know if I'm clear but:

With x86 build ; desktop to 59Hz; No D3D11; launching file; goes into FullScreen Exclusive --> Switches to 60Hz

With x86 build ; desktop to 60Hz; No D3D11; launching file; goes into FullScreen Exclusive --> No Switches

With x64 build ; desktop to 60Hz; No D3D11; launching file; goes into FullScreen Exclusive --> Switches to 59Hz

With x64 build ; desktop to 59Hz; No D3D11; launching file; goes into FullScreen Exclusive --> No Switches.


But during my extensive tests today, I've most of all noticed Fullscreen Exclusive causes dropped frames and other nasty things. For now, the windows renderer is way better for me. But I don't think it is linked with 0.88 since it is the same if I don't use D3D11.

[huhn]

refreshratefix doesn't fix the issue and yes the issue is there with normal FSE mode too.

only 32 bit FSE works fine interesting.

[Dogway]

Profiles are not supported yet for displays. I'm not totally sure I understand what you need/want. madVR should only switch to 1080p24 for 24fps sources, anyway. So what else do you want?

If I only have 1080p24 (or 1080p23 for that matter) in the setting box, every video that goes to fullscreen will turn the TV to 1080p24 mode regardless of source fps. You say this shouldn't happen right? I use madVR v0.87.21, and a bit older MPC-HC (for a reason) v1.6.9.7418 Lite, not sure this is the culprit.

[agustin9]

Maybe you guys already know about this utility, but its what I have used in windows 7 / 8 to stop my PC setting my TV to 59.9X mode.

http://www.monitortests.com/forum/Thread-Custom-Resolution-Utility-CRU

It lets you set what resolutions your display is allowed to use. Once setup correctly, you can make your display only be allowed use 60hz for example by removing the other ones you don't want to allow. This worked for me with a bit of fiddling.
It's a good workaround, thanks!

[pirlouy]

After some tests, here is what I've noticed with 2-3 minutes from the same movie.
From best config to worst

madVR x86, No Fullscreen exclusive (windowed), No D3D11 (D3D9): 0 frames dropped/glitches
madVR x64, No Fullscreen exclusive (windowed), No D3D11 (D3D9): 2 frames dropped/glitches
madVR x86, Fullscreen exclusive, No D3D11 (D3D9): 5 frames dropped/glitches

After, all configs have problems, did not really count, but D3D11 adds dropped frames/glitches, not as much as FullScreen Exclusive+64 bits though.
And so the worst config is:
madVR 64 bits, Fullscreen exclusive, D3D11: 50 frames dropped/glitches

Some comparison with madVR 0.87.21:
madVR x86, No Fullscreen exclusive (windowed): 0 frames dropped/glitches
madVR x64, No Fullscreen exclusive (windowed): 0 frames dropped/glitches
madVR x86, Fullscreen exclusive: 1 frame dropped/glitche
madVR x64, Fullscreen exclusive: 10 frames dropped/glitches

Sorry I read you wanted feedback on debanding for now, but I had some time to do tests, and I'm not interested at all by upscaling/debanding/dithering, I'm not strong enough to notice image quality.

[Asmodian]

If I only have 1080p24 (or 1080p23 for that matter) in the setting box, every video that goes to fullscreen will turn the TV to 1080p24 mode regardless of source fps. You say this shouldn't happen right? I use madVR v0.87.21, and a bit older MPC-HC (for a reason) v1.6.9.7418 Lite, not sure this is the culprit.

If you only have one setting madVR will switch to that setting.

But I'm willing to bow to the majority. What does everybody else say? Should I "allow" Jinc3 to be used for chroma doubling? Instead of blocking it and using Catmull-Rom instead?

Personally I am happy to have Catmull-Rom AR used for doubling the chroma. I can see some improvements in huhn's example, but that is Jinc3 v.s. NNEDI3 not Catmull-Rom v.s Jinc3. For this use case maybe a trade quality for performance option that used chroma scaling, not image scaling, so one could still get NNEDI3 chroma doubling without any extra settings? Is Jinc3 enough of an improvement over Catmull-Rom AR with that source to want the trade quality for performance option?

I take this phrasing to mean that chroma is doubled to match luma, the video is converted back to RGB, and then downscaled with image downscaling (assuming a <2 scaling factor)? Sorry if I am reading too much into your choice of words. :o

[huhn]

don't forget my source was lossless RGB a normal 4:2:0 like 99 % of all videos needs to use nnedi3 on chroma first and that means the chroma doubling will lose even more of it's power.

[Warner306]

If you only have one setting madVR will switch to that setting.



Personally I am happy to have Catmull-Rom AR used for doubling the chroma. I can see some improvements in huhn's example, but that is Jinc3 v.s. NNEDI3 not Catmull-Rom v.s Jinc3. For this use case maybe a trade quality for performance option that used chroma scaling, not image scaling, so one could still get NNEDI3 chroma doubling without any extra settings? Is Jinc3 enough of an improvement over Catmull-Rom AR with that source to want the trade quality for performance option?

I take this phrasing to mean that chroma is doubled to match luma, the video is converted back to RGB, and then downscaled with image downscaling (assuming a <2 scaling factor)? Sorry if I am reading too much into your choice of words. :o

The improvement in that image comparison is very minimal.

A trade quality for performance checkbox that forces Catmull-Rom chroma scaling would make sense. NNEDI3 is meant to be the best of the best scaling algorithm. If the opposite logic was used, place a checkbox next to Image Doubling that allows for Jinc3 chroma scaling. Either way, it is only one checkbox.

[Warner306]

I don't find image doubling too complex or confusing in its current iteration. It is Image Enhancements and Upscaling Refinement that require simplification.

[Anime Viewer]

You won't get highest quality by using the most expensive algorithm for every part of the processing pipeline. You will get highest quality by spending your GPU for algorithms which make a difference. Instead of worrying about chroma doubling, have a look at FineSharp and SuperRes.

But I'm willing to bow to the majority. What does everybody else say? Should I "allow" Jinc3 to be used for chroma doubling? Instead of blocking it and using Catmull-Rom instead?



One other option you could consider regarding Chroma is possibly instituting a SuperChromaRes option (similar to what MPDN has done) like you have done with SuperRes. It might be a less resource draining option than chroma doubling.

[JarrettH]

Where does NEDI image doubling fall in terms of quality? As good as NNEDI3? Better than Jinc? It's certainly fast :devil:

[Anime Viewer]

Thanks for the feedback, guys. But I do want to clear up debanding first, before moving to the other new options. FWIW, Shiandow is still working on maybe improving his debanding algorithm further. So let's give him a couple of days, and then retest, and then decide whether we keep his algorithm in madVR or not. And no, it will not stay as an option. It will either replace the current "high" preset, or it will go.



Shiandow made the following post in the MPDN thread, so he may have something new for madVR to integrate soon:
http://forum.doom9.org/showthread.php?p=1721756#post1721756

Settings on the new debanding uses settings of threshold: 0.2 and detail level: 1

[Asmodian]

don't forget my source was lossless RGB a normal 4:2:0 like 99 % of all videos needs to use nnedi3 on chroma first and that means the chroma doubling will lose even more of it's power.

The improvement in that image comparison is very minimal.

A trade quality for performance checkbox that forces Catmull-Rom chroma scaling would make sense. NNEDI3 is meant to be the best of the best scaling algorithm. If the opposite logic was used, place a checkbox next to Image Doubling that allows for Jinc3 chroma scaling. Either way, it is only one checkbox.

Oh I fully agree, this isn't a use case I personally think needs supporting and I like madshi's current method. Catmull-Rom AR is great, no need for anything else. :)

But if users did want the slightly improved quality for similar rare sources madshi would need to allow NNEDI3 or similar, Jinc3 isn't what gave the minor improvement.

I don't find image doubling too complex or confusing in its current iteration. It is Image Enhancements and Upscaling Refinement that require simplification.

I agree and based on my limited early testing I would be happy to have all the sub options in both of those pages removed and simply have check boxes for FineSharp, LumaSharpen, and SuperRes (with its 3 presets). The defaults work pretty well for me and I have not found other settings I would rather use. However, I haven't tested a wide variety of sources yet and I am sure others like tuning for particular videos. ;)

I think it is our job to play with all those options and a wide variety of sources to come up with some simple presets. Maybe we could agree on decent values for "low", "medium", and "high" presets or similar?

[cca]

I've done some digging and hacking and raping Direct3D, I think I have the refresh rate problems sorted out. Give this build a try:

http://madshi.net/madVR885refreshRateFix.zip

Should work for D3D9 and D3D11 now, and also for 64bit. Once again, I've spent a lot of time to fix Microsoft's stupidity. They actually *intentionally* use 59.940Hz (23.976Hz) even if you ask for 60.000Hz (24.000Hz). There's no way to convince Direct3D to use 60.000 (24.000), except by going behind their back and twisting and binding their fingers, like I've done. I wonder if there's any other software besides madVR which can switch FSE mode to 24.000/60.000? I kinda doubt it...

Doesn't work here either, it always switches to 59.92 or 23.97 respectively instead of 60 or 24Hz. My current workaround is to run the desktop in 59 or 23Hz for now.

[edcrfv94]

image quadrupling with SuperRes still not fix

[cvrkuth]

I've done some digging and hacking and raping Direct3D, I think I have the refresh rate problems sorted out. Give this build a try:

http://madshi.net/madVR885refreshRateFix.zip

Should work for D3D9 and D3D11 now, and also for 64bit. Once again, I've spent a lot of time to fix Microsoft's stupidity. They actually *intentionally* use 59.940Hz (23.976Hz) even if you ask for 60.000Hz (24.000Hz). There's no way to convince Direct3D to use 60.000 (24.000), except by going behind their back and twisting and binding their fingers, like I've done. I wonder if there's any other software besides madVR which can switch FSE mode to 24.000/60.000? I kinda doubt it...

This build work for me in D3D9/32bit scenario only. Unfortunately, it does not work in other scenarios (D3D11/32bit, D3D9/64bit, D3D11/64bit)

[madshi]

With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
- Rendering times are about 20-22ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (BOTH LUMA AND CHROMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (ONLY LUMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.
- I do not get the same results with build 88.5. Please see below.

A) With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
Original Settings
- Rendering times are about 20-22ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (BOTH LUMA AND CHROMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.

B) With build 87.21 (new 88.5 features not available in this build, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
Chroma Doubling + Quadrupling disabled
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image (ONLY LUMA) Doubling + Quadrupling to NNEDI3 32 neurons. No frames are dropped.

C) With build 88.5 (ALL new features turned off, Image Upscaling is set to Lanczos 3 tap, and FSE to disabled):
- Rendering times are about 18-20ms when I set Chroma Upscaling to NNEDI3 32 neurons, and Image Doubling + Quadrupling to NNEDI3 32 neurons. No frame drop this time!

I even went back and set Image Upscaling to Jinc 3 tapm, enabled the Shiandow's Deband Algorithm, and enabled SuperRes filter for Chroma Upscaling and it produced no frame drops! Quadrupling issue is SOLVED with 88.5! . Using SuperRes fot Image Refinment in addition to settings above resulted in frame drops, but as you said - those settings are only for testing.
Ok, good to hear. Is this with Jinc3 AR selected for image upscaling?

BTW, there's one difference between v0.87.x and v0.88.x that I forgot to mention: The setting for image quadrupling is now also applied for "octadrupling" and "hexadrupling". Meaning, NNEDI3 doubling is used as long as upscaling is needed. v0.87.x used NNEDI3 max twice, v0.88.x doesn't have this limit. Is it possible that NNEDI3 is used with v0.88.x on your PC more than twice? You can see that in the debug OSD (Ctrl+J). E.g.:

> Nnedi32 > Nnedi32 > Jinc3 AR (NNEDI3 was used twice)
> Nnedi32 > Nnedi32 > Nnedi32 > Jinc3 AR (NNEDI3 was used thrice)

Will you be adding 60 frames to render ahead for Windowed and Fullscreen now that D3D11 can sync to the VSYNC rate? Or am I misunderstanding the whole D3D11 mode that was recently added?

I've found setting Windowed and Fullscreen to 16 and GPU que to 16 with CPU que at 20 provides very low ms (10-20 ms) playback.
I'm not sure I understand what you're asking/saying. What is a "low ms playback"? You mean low rendering times?

Direct3D maxes out at 16 frames presented in advance.

With this build it switches to 59.93, and leave the screen in that refresh rate even after I go back to windowed. Previously when going back to windowed it switched to 59.99. Windows 8.1 here
No. As i said earlier, I already had this problem with 0.87.21 (no D3D11 if I understood correctly).
I noticed the problem with x64 build since the 64 build switches to 59Hz.
I never noticed the problem with x86 builds before because in fact the x86 build auto switches to 60Hz and I was already in 60Hz.
refreshratefix doesn't fix the issue and yes the issue is there with normal FSE mode too.

only 32 bit FSE works fine interesting.
Doesn't work here either, it always switches to 59.92 or 23.97 respectively instead of 60 or 24Hz. My current workaround is to run the desktop in 59 or 23Hz for now.
This build work for me in D3D9/32bit scenario only. Unfortunately, it does not work in other scenarios (D3D11/32bit, D3D9/64bit, D3D11/64bit)
Stupid me. I forgot that the refresh rate fix also needs new madHcNetXx.dlls. You'll find them here:

http://madshi.net/madVR885refreshRateFix2.zip

These files together with the modified madVR(64).ax should make the refresh rate work again.

ok quadrupling is not that huge sorry but placebo? not sure.chroma doubling is quiet something!

chroma upscale
jinc3AR: http://abload.de/img/nnedi3jinc3arx2syl.png
nnedix2: http://abload.de/img/nnedi3chromax2posri.png
nnedix4: http://abload.de/img/nnedi3chromax424swl.png

and with some more jinc3ar scaling on top:
jinc3AR: http://abload.de/img/nnedi3jinc3arg2sj3.png
nnedix2: http://abload.de/img/nnedi3chromax2rtsql.png
nnedix4: http://abload.de/img/nnedi3chromax4z7ssj.png

i wouldn't be shock if scaling to UHD would show some more potential for quadrupling on this source. but this source is just ...
Thanks! Could you add one screenshot with Catmull-Rom AR instead of Jinc3AR? That was the original question: Whether Catmull-Rom AR looked worse than Jinc3AR for chroma doubling.

After some tests, here is what I've noticed with 2-3 minutes from the same movie.
From best config to worst

madVR x86, No Fullscreen exclusive (windowed), No D3D11 (D3D9): 0 frames dropped/glitches
madVR x64, No Fullscreen exclusive (windowed), No D3D11 (D3D9): 2 frames dropped/glitches
madVR x86, Fullscreen exclusive, No D3D11 (D3D9): 5 frames dropped/glitches

After, all configs have problems, did not really count, but D3D11 adds dropped frames/glitches, not as much as FullScreen Exclusive+64 bits though.
And so the worst config is:
madVR 64 bits, Fullscreen exclusive, D3D11: 50 frames dropped/glitches

Some comparison with madVR 0.87.21:
madVR x86, No Fullscreen exclusive (windowed): 0 frames dropped/glitches
madVR x64, No Fullscreen exclusive (windowed): 0 frames dropped/glitches
madVR x86, Fullscreen exclusive: 1 frame dropped/glitche
madVR x64, Fullscreen exclusive: 10 frames dropped/glitches
Do these frame drops occur just when switching modes or seeking or something? Or do these drops increase during playback? If so, when those frame drops during playback occur, are the queues full or empty? Obviously, if the refresh rate fix fails to work, and your movie frame rate is really 60fps, then there *MUST* be framedrops in FSE mode, in order to keep audio and video in sync. This problem should go away once the refresh rate is correctly set to 60.000Hz.

Where does NEDI image doubling fall in terms of quality? As good as NNEDI3? Better than Jinc? It's certainly fast :devil:
It's too early to talk about that. Feedback is concentrated around debanding for now.

image quadrupling with SuperRes still not fix
NVidia, right?

No problem. Like I said, I wanted to try Windows 10. I guess I should tried fresh install as Win 8.1 > Win 10 didn't work for me.

Anyway, I created 2 clips, one with Bandicam and one with my compact digicam.
Euwh, that's ugly! I can see that presentation glitches are increasing like crazy. But this problem only occurs for a few seconds after each windowed <-> FSE switch and afterwards the problem is gone and playback is smooth, without any presentation glitches etc?

Are you using madVR default settings? No custom flush settings, right?

[edcrfv94]

Ok, good to hear. Is this with Jinc3 AR selected for image upscaling?

BTW, there's one difference between v0.87.x and v0.88.x that I forgot to mention: The setting for image quadrupling is now also applied for "octadrupling" and "hexadrupling". Meaning, NNEDI3 doubling is used as long as upscaling is needed. v0.87.x used NNEDI3 max twice, v0.88.x doesn't have this limit. Is it possible that NNEDI3 is used with v0.88.x on your PC more than twice? You can see that in the debug OSD (Ctrl+J). E.g.:

> Nnedi32 > Nnedi32 > Jinc3 AR (NNEDI3 was used twice)
> Nnedi32 > Nnedi32 > Nnedi32 > Jinc3 AR (NNEDI3 was used thrice)


I'm not sure I understand what you're asking/saying. What is a "low ms playback"? You mean low rendering times?

Direct3D maxes out at 16 frames presented in advance.






Stupid me. I forgot that the refresh rate fix also needs new madHcNetXx.dlls. You'll find them here:

http://madshi.net/madVR885refreshRateFix2.zip

These files together with the modified madVR(64).ax should make the refresh rate work again.


Thanks! Could you add one screenshot with Catmull-Rom AR instead of Jinc3AR? That was the original question: Whether Catmull-Rom AR looked worse than Jinc3AR for chroma doubling.


Do these frame drops occur just when switching modes or seeking or something? Or do these drops increase during playback? If so, when those frame drops during playback occur, are the queues full or empty? Obviously, if the refresh rate fix fails to work, and your movie frame rate is really 60fps, then there *MUST* be framedrops in FSE mode, in order to keep audio and video in sync. This problem should go away once the refresh rate is correctly set to 60.000Hz.


It's too early to talk about that. Feedback is concentrated around debanding for now.


NVidia, right?


Euwh, that's ugly! I can see that presentation glitches are increasing like crazy. But this problem only occurs for a few seconds after each windowed <-> FSE switch and afterwards the problem is gone and playback is smooth, without any presentation glitches etc?

Are you using madVR default settings? No custom flush settings, right?

right ,NVidia GTX670 344.75

[Dogway]

If you only have one setting madVR will switch to that setting.

That's not what madshi said:
madVR should only switch to 1080p24 for 24fps sources, anyway. So what else do you want?

I have a 60fps video and I don't want madvr to do anything if source framerate is not 24p.
edit: And to answer before you ask, no, I don't want to add 1080p60 since that switches to 60p screen mode, and I want to play the file at "1080 @ 60Hz" screen mode (aka in PC mode).

[detmek]

Euwh, that's ugly! I can see that presentation glitches are increasing like crazy. But this problem only occurs for a few seconds after each windowed <-> FSE switch and afterwards the problem is gone and playback is smooth, without any presentation glitches etc?

Are you using madVR default settings? No custom flush settings, right?

No custom flush settings. I tried with defaults, disabling FSE, changing queues, scaling algorithms... Problem remains. Its worse if I activate Smooth Motion and disable FSE and its a bit worse on Windows 7. Then it stutters the all the time in full screen windowed mode. In FSE it stops after few seconds because presentation queue goes full after few seconds.

Another thing, presetation gliches do not rise always. Sometimes those do not change even if video stutters with out of order frames.

[cvrkuth]

I forgot that the refresh rate fix also needs new madHcNetXx.dlls. You'll find them here:

http://madshi.net/madVR885refreshRateFix2.zip

These files together with the modified madVR(64).ax should make the refresh rate work again.

Yes, madvr display modes work correctly now. Thanks!

[huhn]

Thanks! Could you add one screenshot with Catmull-Rom AR instead of Jinc3AR? That was the original question: Whether Catmull-Rom AR looked worse than Jinc3AR for chroma doubling.
added in the post
i would say catmull-rom is better for this source.
but this isn't a perfect x4 upsacale so it is used for image scaling too. simply not a perfect comparison.

edit: madVR885refreshRateFix2.zip works fine

[pirlouy]

Stupid me. I forgot that the refresh rate fix also needs new madHcNetXx.dlls. [...]
These files together with the modified madVR(64).ax should make the refresh rate work again.Confirmed ! It really works !


Do these frame drops occur just when switching modes or seeking or something? Or do these drops increase during playback?It is during playback. Each time I enter Fullscreen, I wait 2-3 seconds then press Ctrl+R to reset stats.
If so, when those frame drops during playback occur, are the queues full or empty?They are full. Then 1 frame dropped, and they are full just after.
Obviously, if the refresh rate fix fails to work, and your movie frame rate is really 60fps, then there *MUST* be framedrops in FSE mode, in order to keep audio and video in sync. This problem should go away once the refresh rate is correctly set to 60.000Hz.I was testing on a monitor which does not have 59Hz, only 60 (even if it's 59,88 in reality). The video was 25Hz interlaced, with smooth motion. So indeed, 50 does not match 60, but with smooth motion, it should be ok. And I managed to have 0 dropped frame/glitche during these 2-3 minutes.
If I take a classic 23,976 Hz video, I have the same dropped frames/glitches ratio.

[aufkrawall]

added in the post
i would say catmull-rom is better for this source.
but this isn't a perfect x4 upsacale so it is used for image scaling too. simply not a perfect comparison.

Hard to say what is caused by luma or chroma upscaling.
But with C-R, the blue robe of "Mangust" is clearly more aliased than with Jinc.

[huhn]

Hard to say what is caused by luma or chroma upscaling.
But with C-R, the blue robe of "Mangust" is clearly more aliased than with Jinc.

yes agree. but look at the icon on the right jinc3ar is blurring them like crazy. for example look at the person in the red robe with the "sparkles" on top of it catmull-rom is a lot sharper there and shows more "details" for more aliasing. the icon in the top right with the blue dots is clearly better with catmull-rom ar.

this is not even the same frame i try to fix this.

[Werewolfy]

Well in this picture catmull-rom ar is even better than Jinc... Look at the picture on top of the men around the fire. The blue part is blurry with Jinc and Nnedi3.

Edit : Ok I didn't notice you edited your post :D

[kalston]

madshi, this is a question, not specifically related to the newer versions of madVR so absolutely no priority in a reply to this (FYI using the latest 0.88.5 version)

I just got a GTX 980 and have it overclocked as much as I can push it, so I have been experimenting with NNEDI 32 neurons for image doubling / quadrupling. Quadrupling seems to be working fine at 32 neurons on very low resolution files. I'm now testing of a movie which is 1280 x 720, so to get to 1920 x 1080 I can see the following is being used in Ctrl-J stats

chroma > Jinc3 AR
luma > Nnedi32 < Catmull-Rom AR
chroma > Catmull-Rom AR

This seems to be alot more demanding than the quadrupling of the lower resolution files, which is fine, so in gpu-z I can see GPU Load hitting ~90% but my question is.

I don't see any dropped frames, delayed frames or presentation glitches, however the queues are as follows

decoder queue 15-16 / 16
upload queue 7-8 / 8
render queue 1-2 / 8
present queue 0-2 / 8

My question is, although I know my gpu is literally at its absolute limits, how come some of the queues seem full, while others are nearly empty all the time, yet my gpu isn't under enough pressure to drop frames. Absolutely very low priority question, just thought it was worth mentioning as I'm not sure if the render and present queues not being full in this scenario may be something madVR is controlling and could potentially be improved to have them more full seeing as frames aren't being dropped.

Those stats blow, do you have a gsync monitor maybe? I get stats like that (including the GPU usage) only when I go FSE with gsync enabled (it's a bug FYI but I think it's nvidia's and not madvr's fault - by using less demanding settings you can fix the queues and get smooth playback). Otherwise I do NNEDI13 128 or 256 easily for 720p to 1440p on a gtx 970 (assuming 24/25 fps content - might have to go down to 32 for 60fps for example).

[cca]

Stupid me. I forgot that the refresh rate fix also needs new madHcNetXx.dlls. You'll find them here:

http://madshi.net/madVR885refreshRateFix2.zip

These files together with the modified madVR(64).ax should make the refresh rate work again.

Yay, it works perfectly now!

[madshi]

ok quadrupling is not that huge sorry but placebo? not sure.chroma doubling is quiet something!

chroma upscale
jinc3AR: http://abload.de/img/nnedi3jinc3arx2syl.png
nnedix2: http://abload.de/img/nnedi3chromax2posri.png
nnedix4: http://abload.de/img/nnedi3chromax424swl.png
editv2: catmull-rom AR http://abload.de/img/nnedi3catv2misgl.png

and with some more jinc3ar scaling on top:
jinc3AR: http://abload.de/img/nnedi3jinc3arg2sj3.png
nnedix2: http://abload.de/img/nnedi3chromax2rtsql.png
nnedix4: http://abload.de/img/nnedi3chromax4z7ssj.png

i wouldn't be shock if scaling to UHD would show some more potential for quadrupling on this source. but this source is just ...

edit2:added the right frame
Ok, Jinc has less aliasing but is also softer than Catmull-Rom. At least we can see that there *is* a difference. However, there's zero difference I can see between enabling and disabling chroma quadrupling. Which may suggest that only the first chroma upsampling operation is important. For 4:2:0 that would be 4:2:0 -> 4:4:4 upsampling. After that, probably chroma doubling would show no visible differences.

So I'm somewhat torn now: With this PC game, chroma doubling helps, and also Jinc looks "different" to Catmull-Rom. So maybe I should put both chroma doubling and Jinc back in? But with a 4:2:0 source maybe there's no difference at all. However, "upscaling refinement" could result in sharpening the chroma channel so much that doubling after refinement might make a difference, once again. So maybe that's a good argument for putting it all back in? Opinions?

I have a 60fps video and I don't want madvr to do anything if source framerate is not 24p.
edit: And to answer before you ask, no, I don't want to add 1080p60 since that switches to 60p screen mode, and I want to play the file at "1080 @ 60Hz" screen mode (aka in PC mode).
What is the difference between 60p screen mode and "1080 @ 60Hz PC mode"?

No custom flush settings. I tried with defaults, disabling FSE, changing queues, scaling algorithms... Problem remains. Its worse if I activate Smooth Motion and disable FSE and its a bit worse on Windows 7. Then it stutters the all the time in full screen windowed mode. In FSE it stops after few seconds because presentation queue goes full after few seconds.

Another thing, presetation gliches do not rise always. Sometimes those do not change even if video stutters with out of order frames.
Ok, I have a guess at what might be wrong. I'll create a test build for you or two to test my theory. I really wish I could reproduce it myself, but for some reason I can't...

It is during playback. Each time I enter Fullscreen, I wait 2-3 seconds then press Ctrl+R to reset stats.
They are full. Then 1 frame dropped, and they are full just after.
I was testing on a monitor which does not have 59Hz, only 60 (even if it's 59,88 in reality). The video was 25Hz interlaced, with smooth motion. So indeed, 50 does not match 60, but with smooth motion, it should be ok. And I managed to have 0 dropped frame/glitche during these 2-3 minutes.
If I take a classic 23,976 Hz video, I have the same dropped frames/glitches ratio.
At the moment when the frame drop occurs, are the queues near empty or are they still full right at that moment? In any case, a log file might clear this up. With smooth motion enabled, there really should not be any frame drops, as long as your GPU is fast enough.

[Shiandow]

So I'm somewhat torn now: With this PC game, chroma doubling helps, and also Jinc looks "different" to Catmull-Rom. So maybe I should put both chroma doubling and Jinc back in? But with a 4:2:0 source maybe there's no difference at all. However, "upscaling refinement" could result in sharpening the chroma channel so much that doubling after refinement might make a difference, once again. So maybe that's a good argument for putting it all back in? Opinions?.

Not sure if it helps, but have you tried using SuperChromaRes for the extra chroma scaling? It should be quite good at making the chroma match the luma, and however many passes you need it's probably still going to be faster than NNEDI3. Of course there's not much point if the quality is worse, but I think it could work quite well.

[Werewolfy]

Ok, Jinc has less aliasing but is also softer than Catmull-Rom. At least we can see that there *is* a difference. However, there's zero difference I can see between enabling and disabling chroma quadrupling. Which may suggest that only the first chroma upsampling operation is important. For 4:2:0 that would be 4:2:0 -> 4:4:4 upsampling. After that, probably chroma doubling would show no visible differences.

So I'm somewhat torn now: With this PC game, chroma doubling helps, and also Jinc looks "different" to Catmull-Rom. So maybe I should put both chroma doubling and Jinc back in? But with a 4:2:0 source maybe there's no difference at all. However, "upscaling refinement" could result in sharpening the chroma channel so much that doubling after refinement might make a difference, once again. So maybe that's a good argument for putting it all back in? Opinions?


I think that if we have found a scenario where we can see the difference, there are possibly other scenarios. You should put back these options at least to judge if it's useful for upscaling refinement or anything else. To test your new version, it's better to have all the cards ;)

[detmek]

Ok, I have a guess at what might be wrong. I'll create a test build for you or two to test my theory. I really wish I could reproduce it myself, but for some reason I can't...


Thanks madshi. I will test those builds when you post it.

[huhn]

Ok, Jinc has less aliasing but is also softer than Catmull-Rom. At least we can see that there *is* a difference. However, there's zero difference I can see between enabling and disabling chroma quadrupling. Which may suggest that only the first chroma upsampling operation is important. For 4:2:0 that would be 4:2:0 -> 4:4:4 upsampling. After that, probably chroma doubling would show no visible differences.

So I'm somewhat torn now: With this PC game, chroma doubling helps, and also Jinc looks "different" to Catmull-Rom. So maybe I should put both chroma doubling and Jinc back in? But with a 4:2:0 source maybe there's no difference at all. However, "upscaling refinement" could result in sharpening the chroma channel so much that doubling after refinement might make a difference, once again. So maybe that's a good argument for putting it all back in? Opinions?

i would say let it as it is and wait for more UHD screen.
upscaling SD to UHD should make the picture very soft with a default scaler and things like chroma doubling may help there a lot.

[madshi]

Not sure if it helps, but have you tried using SuperChromaRes for the extra chroma scaling? It should be quite good at making the chroma match the luma, and however many passes you need it's probably still going to be faster than NNEDI3. Of course there's not much point if the quality is worse, but I think it could work quite well.
You're right, that would be a good option. However, the problem is how to make all this configurable for the user in a good way. That's the hard part. I'll have to think about it.

[madshi]

So here are 2 test builds.

@detmek:
http://madshi.net/madVR885Haswell.zip
Please try them all and let me know which fix the problem, if any. Thanks.

@everyone with the D3D11 minimize / Aero Peak problem:
http://madshi.net/madVR885minimize.zip
Does this one fix the problem?

[clsid]

Yes, that fixes D3D11 minimize issues.

Suggestion: open the settings when double-clicking the tray icon.

[Dogway]

What is the difference between 60p screen mode and "1080 @ 60Hz PC mode"?

On a SmartTV "1080 @ 60Hz" (PC mode) is the screen mode you get when plugging your HTPC to the TV with HDMI interface and set the input as PC, this mode has some image/tv settings disabled and those that aren't are saved within this mode. "1080 24p" mode is the mode you get when you set 1080p24 in madvr, you have more TV image settings enabled and modifications are stored within this mode and (here comes the oddity) any other "p" mode, namely "1080 60p", "1080 50p", etc. What this means is that when I enable "LED Clear Motion" (a marketing word for black frame insertion) for "1080 24p" in order to deal with the implicit 24p motion blur issues, I will also get it for "1080 60p" mode which doesn't need it. That setting also dims screen a bit so I thought on playing non 24p content as "PC Mode", problem is I can't seem to do it once I write 1080p24 in madvr, it wants to turn my PC mode from where I'm launching my videos into some of the "p" modes.

I don't know if you have a smartTV so sorry if this sounds confusing to you.



PD: btw, on your chroma discussion IHMO you should use nnedi3 chroma where you use nnedi3 luma, otherwise you get horrible chroma mismatches on saturated borders.

[pirlouy]

At the moment when the frame drop occurs, are the queues near empty or are they still full right at that moment? In any case, a log file might clear this up. With smooth motion enabled, there really should not be any frame drops, as long as your GPU is fast enough.
They seem full for me: (7-8), then frame dropped, then (0-8), and in 1ms (7-8) again, until next dropped frame.

Here is a log. Hope it will help. I double click the file which opens the player, go into fullscreen, reset stats (Ctrl+R), let the problems occurs (for 1 minute i guess), then pause, exit fullscreen, closes player.
http://dl.free.fr/jHUVQcG4y

[madshi]

Suggestion: open the settings when double-clicking the tray icon.
Funny, I was thinking exactly the same a couple of days ago... :)

On a SmartTV "1080 @ 60Hz" (PC mode) is the screen mode you get when plugging your HTPC to the TV with HDMI interface and set the input as PC, this mode has some image/tv settings disabled and those that aren't are saved within this mode. "1080 24p" mode is the mode you get when you set 1080p24 in madvr, you have more TV image settings enabled and modifications are stored within this mode and (here comes the oddity) any other "p" mode, namely "1080 60p", "1080 50p", etc. What this means is that when I enable "LED Clear Motion" (a marketing word for black frame insertion) for "1080 24p" in order to deal with the implicit 24p motion blur issues, I will also get it for "1080 60p" mode which doesn't need it. That setting also dims screen a bit so I thought on playing non 24p content as "PC Mode", problem is I can't seem to do it once I write 1080p24 in madvr, it wants to turn my PC mode from where I'm launching my videos into some of the "p" modes.
Ok, but PC modes vs. non-PC mode sounds like a setting in your TV, not on your HTPC. So why does your TV switch into non-PC mode if you let madVR switch to 1080p60? And how do *you* manually switch your HTPC into a mode which turns on PC mode on your TV? Are there two different 1080p60 modes on your HTPC, one for TV PC mode and one not? And madVR switches into the wrong 1080p60 mode? That's the part I don't understand. In theory you should get exactly what you want by adding both "1080p24" and "1080p60" into the madVR display mode edit box.

[James Freeman]

So here are 2 test builds.

@everyone with the D3D11 minimize / Aero Peak problem:
http://madshi.net/madVR885minimize.zip
Does this one fix the problem?

YES it does!
Can we please try a 64 build?