madVR - high quality video renderer (GPU assisted)

[Warner306]

If I'm going for Mediabrowser it will be WMC + Mediabrowser Classic.. Mediabrowser Theater does not seem like a good idea, I would much rather have MPC-HC as external player.

Though I'm not entirely sure how that would work out towards my Plex server.

I don't know if Media Browser has an addon for Plex, but Kodi does:

PlexBMC: https://forums.plex.tv/index.php/topic/106593-plexbmc-xbmc-add-on-to-connect-to-plex-media-server/

The add-on is very popular. The catch is you need a skin that offers integrated support, however.

Media Browser is also great. And switching to its server system is not a significant downgrade. I also use it on a regular basis. But I prefer Plex for its easy-to-use interface. I've only remained with MediaBrowser for its transcoding to mobile devices due to it reading nfo files and file names created by Kodi.

From use, it is clear that Kodi and Media Browser are close friends due to their cross-compatibility, and you can also use a plugin to harness the power of your Media Browser server with Kodi. So Plex and Media Browser play nice with Kodi, but I don't know if the same can be said about Media Browser and Plex.

[Warner306]

You are not going to be able to use NNEDI3 with an iGPU, not yet. As an aside why are you doubling chroma? Using my Titan, I prefer 128 neuron luma doubling over 16 neuron luma + chroma doubling + quadrupling.



All vendors seem to do better with DXVA2 vs anything else (it uses the same hardware as QuickSync/CUDA but is more robust).



That is not what smooth motion does. It does exact frame blending to perfectly display one frame rate on a display with a different refresh rate. This avoid the artifacts associated with SVP and the judder associated with mismatched frame rates, e.g. 24p @ 60 Hz.

It is more demanding if using an ED dithering technique because dithering is performed on more frames (assuming the display refresh rate is above the source frame rate).

I can use bicubic scaling with smooth motion to 60 Hz and ordered dithering on an HD 3000.

Maybe my MadVR Options Explained (http://forum.doom9.org/showthread.php?p=1709584#post1709584) would help?

Smooth motion still looks like interpolation to me. It is really a matter of taste. What it should be described as is the best interpolation that is available today. But it still alters the video in an unnatural way.

I am using an HD3000 on a laptop and I can see the upgrade in picture quality at:

Chroma: Bicubic100 + AR
Luma: Lanczos3 + AR

I can't turn anything else on, including smooth motion, but I wouldn't watch video using any other renderer. After a while, even the default settings are obvious to a trained eye.

[RamGuy]

You are not going to be able to use NNEDI3 with an iGPU, not yet. As an aside why are you doubling chroma? Using my Titan, I prefer 128 neuron luma doubling over 16 neuron luma + chroma doubling + quadrupling.



All vendors seem to do better with DXVA2 vs anything else (it uses the same hardware as QuickSync/CUDA but is more robust).



That is not what smooth motion does. It does exact frame blending to perfectly display one frame rate on a display with a different refresh rate. This avoid the artifacts associated with SVP and the judder associated with mismatched frame rates, e.g. 24p @ 60 Hz.

It is more demanding if using an ED dithering technique because dithering is performed on more frames (assuming the display refresh rate is above the source frame rate).

I can use bicubic scaling with smooth motion to 60 Hz and ordered dithering on an HD 3000.

Maybe my MadVR Options Explained (http://forum.doom9.org/showthread.php?p=1709584#post1709584) would help?

I don't know if Media Browser has an addon for Plex, but Kodi does:

PlexBMC: https://forums.plex.tv/index.php/topic/106593-plexbmc-xbmc-add-on-to-connect-to-plex-media-server/

The add-on is very popular. The catch is you need a skin that offers integrated support, however.

Media Browser is also great. And switching to its server system is not a significant downgrade. I also use it on a regular basis. But I prefer Plex for its easy-to-use interface. I've only remained with MediaBrowser for its transcoding to mobile devices due to it reading nfo files and file names created by Kodi.

From use, it is clear that Kodi and Media Browser are close friends due to their cross-compatibility, and you can also use a plugin to harness the power of your Media Browser server with Kodi. So Plex and Media Browser play nice with Kodi, but I don't know if the same can be said about Media Browser and Plex.



I was considering to drop the Intel NUC 5i5RYK and instead use some of my spare parts, mainly a Intel Xeon E-1270v2 and a Asus P9B Workstation motherboard and 8GB of DDR3 ECC RAM. Tossing it into a Fractal Design sound dampening case and use a huge arse Nocuta NH-D15 in order to cool it without the need of much fan noise and perhaps try to toss a GTX 960 or 970 with Asus DirectCU II cooler and a Corsair AX power supply.

That should be much more capable in terms of MadVR, and then I could also go with a TV-card and get our satellite television going on the HTPC but sadly it does not seem like my satellite provider is going to support HTPC with these new CI+ CAMs and whatnot so I'm forced to scrap that idea.


That renders the point of going for a larger build less appealing, the only question is whether going GTX 960 / GTX 970 and Intel Xeon E3-1270v2 is worth it compared to the Inel NUC 5i5RYK with Intel Core i5-5250U and Intel HD 6000. Due to me already owning lots of the spare parts the total cost will be about the same, but of course the NUC will be much smaller, more silent and sexy compared to the size of other system. The big question is whether the extra size, heat and noise is worth it or not.

[Asmodian]

Smooth motion still looks like interpolation to me. It is really a matter of taste. What it should be described as is the best interpolation that is available today. But it still alters the video in an unnatural way.

What do you mean? Do you notice the slight softening due to the frame blending?

It is not interpolation, nothing is interpolated in any way, please do not describe it as such. It simply displays each source frame for exactly the amount of time it is supposed to be displayed for. This removes the judder from slightly or very different frame rates v.s. refresh rates but doesn't cause artifacts or generate video that is smoother than the source would be if watched on a display that's refresh rate perfectly matched the frame rate.

The big question is whether the extra size, heat and noise is worth it or not.

Sorry but only you can answer this. :)

Is NNEDI3 and/or Jinc worth it? I am not sure if an HD 6000 could use Jinc. If it could I would go with the NUC as I like Jinc a lot. :p

[Warner306]

What do you mean? Do you notice the slight softening due to the frame blending?

It is not interpolation, nothing is interpolated in any way, please do not describe it as such. It simply displays each source frame for exactly the amount of time it is supposed to be displayed for. This removes the judder from slightly or very different frame rates v.s. refresh rates but doesn't cause artifacts or generate video that is smoother than the source would be if watched on a display that's refresh rate perfectly matched the frame rate.


I find the blurring is not obscured well enough. It is always noticeable with all types of content. The result in terms of appearance of the frame blending is very similar to frame interpolation solutions. This is my opinion, of course. I think the use of smooth motion really comes down to personal taste. I'm not trying to spread misinformation.

True 24p playback, by comparison, is not nearly as smooth. This can be good or bad depending on your tastes.

[Asmodian]

I find the blurring is not obscured well enough.

It is not obscured at all, no processing is done beyond the display time weighted frame blends when two frames need to be displayed during one refresh.

True 24p content, by comparison, is not nearly as smooth. This can be good or bad depending on your tastes.

Is this v.s. 24p at 24/48/72/96/120 Hz or 24p at 60 Hz? Smooth motion looks exactly as smooth as 24p at 24 Hz to me while maybe being a tiny bit softer. I wonder if response times are influencing our perception. Smooth motion is not intended to, and theoretically should not, make the video appear smoother than the source.

Smooth motion looks looks exactly like 30p at 90 Hz (or 25p at 100Hz) when watching 30p (or 25p) at 96 Hz which is the only time I actually it. :p

[Warner306]

It is not obscured at all, no processing is done beyond the display time weighted frame blends when two frames need to be displayed during one refresh.



Is this v.s. 24p at 24/48/72/96/120 Hz or 24p at 60 Hz? Smooth motion looks exactly as smooth as 24p at 24 Hz to me while maybe being a tiny bit softer. I wonder if response times are influencing our perception. Smooth motion is not intended to, and theoretically should not, make the video appear smoother than the source.

Smooth motion looks looks exactly like 30p at 90 Hz (or 25p at 100Hz) when watching 30p (or 25p) at 96 Hz which is the only time I actually it. :p

My experiences with smooth motion relate to viewing 720p/1080p 24hz content displayed at 60hz. At this frame rate, I don't find it is a good replacement for 3/2 pulldown. In fact, I prefer 3/2 pulldown alone over any type of artificial frame manipulation.

[ryrynz]

MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz (definitely 60Hz) I've tested on my HD4000 and Ti 750 and I wouldn't recommend using 60Hz for your best viewing experience until madshi improves it. EVR renderer, Sync renderer and MPDN deliver far better results, It's quite easy to see this with scrolling text and is extremely noticeable when I enable frame creation on my TV.

No options in MadVR improve this issue unfortunately so if your display supports 24Hz then I'd recommend switching to that, then compare vs 60Hz with Smooth Motion disabled so you can see for yourself.

Unfortunately some monitors like my U2412 don't support 24Hz. Smooth Motion alleviates some of the issues with it, but again with scrolling text you see mirrored copies of the text above and below and it doesn't look very good
and that's just the nature of how it works.

Nothing beats 24Hz with IFC enabled on my TV set to medium. Super smooth, text just glides across or down the screen, no jumps. It's almost like I'm watching true 60FPS content. You could achieve something similar using SVP, but it's not quite as good as most stuff you'll find in TVs and requires quite a bit of processing power. My recommendation would be to own a display that has this technology built in and calibrate your display. Reducing sharpness down from defaults is nearly always a good idea.

[Warner306]

MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz (definitely 60Hz) I've tested on my HD4000 and Ti 750 and I wouldn't recommend using 60Hz for your best viewing experience until madshi improves it. EVR renderer, Sync renderer and MPDN deliver far better results, It's quite easy to see this with scrolling text and is extremely noticeable when I enable frame creation on my TV.

No options in MadVR improve this issue unfortunately so if your display supports 24Hz then I'd recommend switching to that, then compare vs 60Hz with Smooth Motion disabled so you can see for yourself.

Unfortunately some monitors like my U2412 don't support 24Hz. Smooth Motion alleviates some of the issues with it, but again with scrolling text you see mirrored copies of the text above and below and it doesn't look very good
and that's just the nature of how it works.

Nothing beats 24Hz with IFC enabled on my TV set to medium. Super smooth, text just glides across or down the screen, no jumps. It's almost like I'm watching true 60FPS content. You could achieve something similar using SVP, but it's not quite as good as most stuff you'll find in TVs and requires quite a bit of processing power. My recommendation would be to own a display that has this technology built in and calibrate your display. Reducing sharpness down from defaults is nearly always a good idea.

I use 60hz playback with madVR and do not require smooth motion for smooth playback. In fact, I find it quite enjoyable. Items such as choppy text scrolling are more likely due to 3/2 pulldown than any problems with the renderer.

Is it possible you are using aggressive scaling and seeing artifacts instead? madVR can be very aggressive depending on the settings used.

[Asmodian]

My experiences with smooth motion relate to viewing 720p/1080p 24hz content displayed at 60hz. At this frame rate, I don't find it is a good replacement for 3/2 pulldown. In fact, I prefer 3/2 pulldown alone over any type of artificial frame manipulation.

Haha, well smooth motion does look a lot smoother than 3/2 pulldown but that means you would probably prefer 3/2 pulldown to 24p on a 24 Hz display without any type of frame manipulation (like a movie theater). 3/2 is as much a form of artificial frame manipulation as smooth motion is, just a stupid and old one. It is artificial frame manipulation that adds a lot of motion judder that is not in the original source; the blending from smooth motion is much less noticeable to me.

It is funny how we get used to something and end up not liking video that is done correctly. This is similar to how many don't like native 48p/60p instead of native 24p video (but even worse IMO). We are so used to the 3/2 judder we prefer it though I am not sure why as real life tends to not have any. :p

MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz (definitely 60Hz) I've tested on my HD4000 and Ti 750 and I wouldn't recommend using 60Hz for your best viewing experience until madshi improves it. EVR renderer, Sync renderer and MPDN deliver far better results, It's quite easy to see this with scrolling text and is extremely noticeable when I enable frame creation on my TV.

In my experience I get a very smooth and stable frame delivery using 60-144 Hz with madVR. I am not sure what you are talking about.

Nothing beats 24Hz with IFC enabled on my TV set to medium. Super smooth, text just glides across or down the screen, no jumps. It's almost like I'm watching true 60FPS content. You could achieve something similar using SVP, but it's not quite as good as most stuff you'll find in TVs and requires quite a bit of processing power. My recommendation would be to own a display that has this technology built in and calibrate your display. Reducing sharpness down from defaults is nearly always a good idea.

Sadly every form of IFC I have tried fails when there is multiple forms of movement within the same frame. I tend to fixate on the artifacts instead of enjoying the smoothness during the nice uniform motion it is good at. 24p @ 24, 48, 72, 96, or 120 Hz is a more enjoyable experience for me.

[6233638]

As far as I see, there's no technical reason why you should have sync problems without delay, but no sync problems with delay. Actually the delay option is quite "hacky". So if there's a problem, I would have thought it more likely that the delay problem causes a problem. Anyway, after a couple of seconds there should be no difference between delay on/off, so I'm really confused as to where those problems could be coming from. You have the same problems regardless of rendering mode (FSE vs. windowed vs. overlay, "new" presentation path vs. "old")? I think the issue is worth looking into, at least if you can reliably reproduce it.I ended up buying a latency tester and have been doing some testing throughout the week.
I still need to set aside the time to compare things like FSE/Windowed, New/Old presentation, NNEDI3 on/off, but from my testing thus far, it looks like it is probably not madVR at fault here and my issues were specific to JRiver's madVR implementation.

When testing MPC-HC I would see a maximum sync variance in the range of ±8ms from the average when seeking through a video.
In JRiver, that could be as much as ±300ms if madVR's queues were at the maximum size and JRiver's VideoClock feature was disabled.
Enabling their VideoClock feature tightened up sync to a maximum variance of about ±40ms (which is still too much) and it prevented there being any drift once it settled. (without it, sync would change by 1ms every few minutes)

Now that I'm able to get objective data on these audio sync issues, it does not appear that the "delay playback" options have a positive effect on the audio sync; I think it's just been a coincidence that sync appeared to improve whenever I enabled those options, because seeking or pausing/resuming a video is enough to throw it in or out of sync in JRMC (as of 20.0.80) and it just looked like sync was getting better.

Hendrik (Nevcariel) has one of the same latency testers now, so hopefully he will be able to figure out the cause of this variance in JRMC and correct it.

MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz (definitely 60Hz) I've tested on my HD4000 and Ti 750 and I wouldn't recommend using 60Hz for your best viewing experience until madshi improves it. EVR renderer, Sync renderer and MPDN deliver far better results, It's quite easy to see this with scrolling text and is extremely noticeable when I enable frame creation on my TV.That's not been my experience at all. Are you getting dropped/repeated frames during playback?

[TheRyuu]

As per madshi's request[1]:
Correct. If you want this fixed, could you please create a bug tracker report? Personally, I don't find this very important because the 16ms timer is idiotic, anyway. Maybe it made sense 10 years ago. But with today's PCs 16ms timer resolution simply makes no sense, whatsoever. Of course that's only my personal opinion. I do consider it "not nice" that madVR doesn't restore the original timer resolution, though.
I created a bug tracker report[2]. I also created a second one unrelated to the above problem but which was something else I reported several months ago here and still has not been addressed[3].

[1] http://forum.doom9.org/showthread.php?p=1710039#post1710039
[2] http://bugs.madshi.net/view.php?id=255
[3] http://bugs.madshi.net/view.php?id=256

[ryrynz]

That's not been my experience at all. Are you getting dropped/repeated frames during playback?

Nope no issues there. I've actually seen similar issues to a lesser degree with the other renderers. Might have something to do with how the TV deals with the output. MadVR has more of an issue with stuttering and larger jerking.

Can this latency tester tell if the output is exactly like (or better than) the output of a blu-ray player?

[6233638]

Nope no issues there. I've actually seen similar issues to a lesser degree with the other renderers. Might have something to do with how the TV deals with the output. MadVR has more of an issue with stuttering and larger jerking.

Can this latency tester tell if the output is exactly like (or better than) the output of a blu-ray player?The tester (http://www.sync-one2.co.uk/) just measures the delay between a flash on-screen and a tone being played.
It wouldn't be able to test how stable the video presentation is compared to a Blu-ray player, if that's what you mean.

Other than the issues I'm having with sync being offset every time I seek during playback (or pause/resume, or refresh rate changes) which seems to be an issue that is specific to JRMC, with the VideoClock feature enabled (and almost certainly in other players with ReClock or similar) the audio sync remains stable within ±1ms.

Without VideoClock, sync was increasing by about 1ms every couple of minutes, so it would probably be noticeably out of sync if the video is long enough. (or madVR would drop/repeat frames?)
A lot of that probably depends on what your clock deviation is. I'm sure it would have drifted out of sync quicker if I was testing 23FPS content rather than 24FPS for example, as my GPU doesn't seem great at 23Hz.

[Patrik G]

It will be used towards my Pioneer LX 5090H through my Yamaha RX-V3900 AV-reciever.

can you do a test with your 5090 and see what you think.
set refresh rate to 50hz from the PC and select "Drive Mode 2" and Pure Cinema Advanced.
that will give you 100Hz on the Kuro.

just check how you think motion looks like on 24p content or anything else.

i already using this mode on my KRP-500A with madvr

also may i have a suggestion
instead of worrying about graphics cards that do very little for the overall picture quality you should save your money and buy a colormeter instead.
calibrate your Kuro with 9200 Patches 3D LUT for madvr and you will be blown away by the picture quality.
no graphics card can ever give you that reference picture.
with 9200 calibrated colorpatches you will have close to the same reference picture as the Dolby Reference studio monitors has that costs around 40K $
http://www.dolby.com/us/en/professional/cinema/products/prm-4220.html
it takes around 5 hours to calibrate 9200 colorpatches but its worth it.
you will see a picture with a 3D depth like you never seen before on your kuro.

if you have any question on how to set up all programs for calibration 3D LUT for use with madvr just send a PM
i think it will go off topic otherwise :)

one picture from my KRP with madvr and 9200 calibrated patches 3D LUT (Running watercooled GTX 780 SLIs)
http://privat.bahnhof.se/wb192876/madvr1.jpg.jpg

one more in a pitch black room.
http://privat.bahnhof.se/wb192876/Dieanotherday1.jpg

[madshi]

Thanks, to clarify, I'm mostly interested in Blu Ray image quality (most are 1080p24) on a 1080p TV (no 4k) and second priority would be Australian TV which is 576i50, 720p25 and 1080i50. For Blu Ray I would like to be able to match the latest Oppo Blu Ray player. Any recommendations for a GPU and CPU for this? If it has low TDP that's a bonus. Thanks again.
For 1080p playback on a 1080p TV you don't need a lot of GPU power, because no scaling is needed (other than chroma upsampling). Even a budget GPU should be able to do that with good quality when using madVR. Playing back 576i50 will be much more demanding, but it also depends: If you're talking about movie content, you can usually turn deinterlacing off or use madVR's forced film mode, which results in this becoming 576p25. If you're talking about video content (or frame blended encodings), you need to use video mode deinterlacing and you end up with 576p50. Now 576p50 consumes twice as much power as 576p25. So the GPU you need heavily depends on whether your 576i50 content needs video mode deinterlacing or not.

If you "just" want to match the Oppo, Lanczos3AR should do the trick, and a budget GPU should be more than you need. E.g. the NVidia 750 might be a good affordable choice. It might even allow you to use JincAR. Not sure about NNEDI3, which is the best of the scaling algorithms, but also requires the most GPU power. For that the 750 might not be fast enough. Well, maybe it *is* fast enough for 576p25, I'm not sure. But you don't need NNEDI3 to match the Oppo. I'm excluding Darbee processing here, though. Do you want to match Darbee processing, too? Then you need to use some sharpening algorithm in madVR, e.g. LumaSharpen or FineSharp. Those will require some extra GPU power, but I think it's not that much extra GPU power. Not totally sure, though.

In MPC-HC and MadVR neutral values.
Then I don't really know what's happening there. The only explanation that is left seems to be that the test video you're using for madVR playback asks for different color values compared to the calibration software you're using with madTPG. But that doesn't really sound all that likely. What happens if you turn off the 3dlut and then compare madVR vs madTPG in windowed mode? You can use a tool like "colorPicker" to check the exact RGB color values on screen. Are they identical between madVR and madTPG?

Is there a way to enlarge the font of the output of ctrl+j?
Not at the moment.

Just out of curiosity, could someone provide me with the specs and/or graphics card models that can do the following:

Input Resolutions: 720p, 1080p
Output Resolution: 3840 x 2160 (4K)

Desired Settings:

Chroma: Jinc3 + AR
Image: Jinc3 + AR
Luma Doubling: NNEDI364 at 2x scaling factor or greater
Artifact Removal - Debanding: Low/Medium
Dithering: Error Diffusion 1

Smooth Motion is disabled.

What card(s) are capable of these settings (if any)?
Can't tell you, for 2 reasons:

1) You haven't told us which framerates you need support for. 60p requires 2.5 as much GPU power as 24p. So if you need 60p support, your GPU has to be 2.5x as powerful as otherwise.

2) I don't have every GPU that exists here to run tests with. So I've no clue which GPU can run the exact list of algos/settings you need.

You can look for posts by renethx on AVSForum to get an impression of which kind of GPUs can run which algorithms at roughly which fps. Those posts might be a bit older, though, so they might not cover the latest GPUs (not sure).

Does this mean that Nvidia NNEDI3 performance is about to get better for Windows 10 users?
I've no idea, but I doubt it. OpenCL 1.2 is very nice to have, but won't improve performance. It will just improve ease of development, and possible features. They might have improved performance at the same time, but who can know that?

Can someone refresh my memory on why force film mode deinterlacing doesn't work with dxva2 decoding?
Because forced film mode is currently done by the CPU, so the video frames must be delivered to madVR via CPU. So forced film mode works in all modes except native DXVA decoding.

madVR has been showing an error saying it encountered a problem every fifth time or so that I first open and quickly full-screen a video. While the video is usually still playing fine when this pop-up shows up, I usually click "close the application" and relaunch the video just to make sure everything is okay. I attached a crash log from the last time it happened.
That's bad. For some reason the error log doesn't contain enough information. It seems that error reporting is partially broken with the latest build... :( I'll fix that for the next build. You may have to re-report the issue with the next build to get it fixed. Sorry...

madshi, how about providing a boolean variable for smoothmotion that can be used in the profile rules?
Yeah, sure, I can do that.

I find the blurring is not obscured well enough. It is always noticeable with all types of content. The result in terms of appearance of the frame blending is very similar to frame interpolation solutions. This is my opinion, of course. I think the use of smooth motion really comes down to personal taste. I'm not trying to spread misinformation.

True 24p playback, by comparison, is not nearly as smooth. This can be good or bad depending on your tastes.
There is some blurring, that's true. My impression is that the blurring is more visible for some users than others. It might depend on the user, or maybe also on the exact display. Some users with their specific displays don't seem to see much (if any) blurring at all, while others complain about it. Some few users find the blurring too much to handle, but the majority of users doesn't seem to be bothered by it enough to lose the advantage of smoother playback.

The "look" of madVR's smooth motion is *very* different to frame interpolation. Frame interpolation isn't blurry at all (if properly implemented). Frame interpolation is very smooth, but sharp at the same time. Also motion is soo ultra-smooth that it looks like video instead of a movie. The whole "feel" of the movie changes. Some users say it changes for the better, others don't agree, but that's a matter of taste. madVR's smooth motion does not change the look of the movie at all (other than introducing a small amount of blurring/ghosting). The artifacts introduced by madVR's smooth motion algo are totally different compared to what frame interpolation does.

When you say "true 24p playback" is not nearly as smooth than madVR's smooth motion playback, then your "true 24p playback" chain is broken. That is probably caused by your display running at 60Hz, so you get the 3:2 pulldown judder. Please understand that the 3:2 pulldown judder is *not* the way the movie was meant to be seen. It's also *not* what you see in commercial movie theaters. Commercial movie theaters do not have 3:2 pulldown judder. The smoothness achieved by madVR's smooth motion algorithm exactly matches what you get with a perfect 24fps playback chain without any motion interpolation. Basically if you turn madVR's smooth motion algo on, you get exactly the same smoothness you have in a commercial movie theater. However, you do get some blurring added. But the blurring just makes the image softer, or introduces some slight ghost images, but it doesn't make motion smoother than it should be. Making motion smoother than it would be in a commercial movie theater is not within the technical capabilities of my algorithm.

MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz (definitely 60Hz) I've tested on my HD4000 and Ti 750 and I wouldn't recommend using 60Hz for your best viewing experience until madshi improves it.
You need to be careful to be clear about what you're saying. The way you've written I can say with 100% certainty that you're wrong. However, I'm pretty sure that what you really mean is not what you wrote here. So here's what is really true, and I think you'll probably agree with me:

madVR is perfectly smooth when playing back properly encoded movie content with a display which has a refresh rate that either matches the movie framerate, or is an even multiply of the movie framerate. E.g. playing 24fps @ 24Hz is perfectly smooth. Also playing 25fps @ 50Hz is perfectly smooth. Or playing 60fps @ 60Hz.

However, madVR's pulldown algorithm is pretty bad atm. Which means that if you play 24fps content at 60Hz (without using smooth motion), you don't get a reliable 3:2 pulldown pattern. Instead some frames are shown twice, others three times, with no regular pattern. That is not very good and definitely needs to be improved. But this is only a problem if your display refresh rate is not an even multiply of your movie framerate. If you are in that situation, I would usually recommend to activate smooth motion, which should also give you perfect smoothness (but with some blurring/ghosting artifacts, which depending on your display and your eyes/brain may be annoying). From your comment it seems that your TV can take 24p content played with 3:2 playback at 60Hz and turn it into smooth 60p? That sounds kinda weird, but I guess it's possible. In this case madVR is not a good solution for you right now because madVR's pulldown pattern is not regular atm.

I ended up buying a latency tester and have been doing some testing throughout the week.
I still need to set aside the time to compare things like FSE/Windowed, New/Old presentation, NNEDI3 on/off, but from my testing thus far, it looks like it is probably not madVR at fault here and my issues were specific to JRiver's madVR implementation.

When testing MPC-HC I would see a maximum sync variance in the range of ±8ms from the average when seeking through a video.
In JRiver, that could be as much as ±300ms if madVR's queues were at the maximum size and JRiver's VideoClock feature was disabled.
Enabling their VideoClock feature tightened up sync to a maximum variance of about ±40ms (which is still too much) and it prevented there being any drift once it settled. (without it, sync would change by 1ms every few minutes)

Now that I'm able to get objective data on these audio sync issues, it does not appear that the "delay playback" options have a positive effect on the audio sync; I think it's just been a coincidence that sync appeared to improve whenever I enabled those options, because seeking or pausing/resuming a video is enough to throw it in or out of sync in JRMC (as of 20.0.80) and it just looked like sync was getting better.

Hendrik (Nevcariel) has one of the same latency testers now, so hopefully he will be able to figure out the cause of this variance in JRMC and correct it.
That's great - thanks for your efforts!

Without VideoClock, sync was increasing by about 1ms every couple of minutes, so it would probably be noticeably out of sync if the video is long enough. (or madVR would drop/repeat frames?)
What I can say is that every DirectShow graph has a reference/master clock. It's usually provided by the audio renderer. But if the audio renderer doesn't provide one, or if there's no audio track, DirectShow will simply pick a different reference clock. In any case, there will always be one master clock, which must be considered "the law" for every playback component.

madVR carefully presents every video frame so that it's visible on screen exactly at the right time, according to the master clock. This is why I always said that I have a hard time believing that madVR could be at fault here. The whole madVR presentation concept is totally hard wired to obey the master clock. So I don't think madVR can be responsible for any audio/video desync problems - unless such problems only occur in specific modes (e.g. only in new windowed presentation mode or so).

I would guess that the audio renderer used by JRiver doesn't properly sync to the master clock for some reason. Which is weird because the audio renderer is usually supposed to *be* the master clock. Of course when activating VideoClock, things kinda change because VideoClock actually tries to modify the master clock to match the VSync rate reported by madVR. But without VideoClock active, the audio renderer should define the master clock simply by its current location in audio playback. If it did that, there should/would be no desync.

As per madshi's request[1]:

I created a bug tracker report[2]. I also created a second one unrelated to the above problem but which was something else I reported several months ago here and still has not been addressed[3].

[1] http://forum.doom9.org/showthread.php?p=1710039#post1710039
[2] http://bugs.madshi.net/view.php?id=255
[3] http://bugs.madshi.net/view.php?id=256
Thank you, I appreciate #255. However, I don't think I'll fix #256, as discussed/explained earlier here in the forum. Furthermore the cardinal value 256 is outside of 8bit, so I can't fix it, anyway, as long as madVR is limited to 8bit output (just kidding).

[bytheway]

For 1080p playback on a 1080p TV you don't need a lot of GPU power, because no scaling is needed (other than chroma upsampling). Even a budget GPU should be able to do that with good quality when using madVR. Playing back 576i50 will be much more demanding, but it also depends: If you're talking about movie content, you can usually turn deinterlacing off or use madVR's forced film mode, which results in this becoming 576p25. If you're talking about video content (or frame blended encodings), you need to use video mode deinterlacing and you end up with 576p50. Now 576p50 consumes twice as much power as 576p25. So the GPU you need heavily depends on whether your 576i50 content needs video mode deinterlacing or not.

If you "just" want to match the Oppo, Lanczos3AR should do the trick, and a budget GPU should be more than you need. E.g. the NVidia 750 might be a good affordable choice. It might even allow you to use JincAR. Not sure about NNEDI3, which is the best of the scaling algorithms, but also requires the most GPU power. For that the 750 might not be fast enough. Well, maybe it *is* fast enough for 576p25, I'm not sure. But you don't need NNEDI3 to match the Oppo. I'm excluding Darbee processing here, though. Do you want to match Darbee processing, too? Then you need to use some sharpening algorithm in madVR, e.g. LumaSharpen or FineSharp. Those will require some extra GPU power, but I think it's not that much extra GPU power. Not totally sure, though.

Which GPU are you most excited about its potential for madVR that will be released in 2015?

[makz]

Then I don't really know what's happening there. The only explanation that is left seems to be that the test video you're using for madVR playback asks for different color values compared to the calibration software you're using with madTPG. But that doesn't really sound all that likely. What happens if you turn off the 3dlut and then compare madVR vs madTPG in windowed mode? You can use a tool like "colorPicker" to check the exact RGB color values on screen. Are they identical between madVR and madTPG?
I noticed, that with gamma bt1886 2.4 absolut 0% offset in madtpg and mpchc match colors or their very close (result avarage gamma 2.15). When I choose custom 2.4 and offset 50-80, I get oversatturated colors in mpc-hc... maybe this information will help. Test with Ted's LightSpace CMS Calibration Disk and GCD

[ryrynz]

You need to be careful to be clear about what you're saying. The way you've written I can say with 100% certainty that you're wrong. However, I'm pretty sure that what you really mean is not what you wrote here. So here's what is really true, and I think you'll probably agree with me:


I agree with you that what I said wasn't specific, multiples of 24hz of course will be fine, it kinda goes without saying.
Many displays only have 23,24,50,59 & 60 available for selection so if 50,59 and 60 are all having the same issue then how I wrote it appears fine to me.. but you can nail down the specifics where I can't, cheers.


However, madVR's pulldown algorithm is pretty bad atm.


And that would be the nail on the head.


If you are in that situation, I would usually recommend to activate smooth motion, which should also give you perfect smoothness


Smooth motion doesn't allow the FRC to provide perfect smoothness. It happens occasionally for a couple of seconds then breaks out of it. Then there's the shimmering caused by the blending,
I had it enabled until I realized I was better off without it.


From your comment it seems that your TV can take 24p content played with 3:2 playback at 60Hz and turn it into smooth 60p? That sounds kinda weird, but I guess it's possible.


Yup, it's called 24p smooth film. I always preferred to avoid 24Hz because of the screen flicker and I didn't see any advantages as I wasn't testing the right situations.
It's wasn't till I played around with other players/renderers that I saw what was happening.

Instead of using a lower FRC setting on the TV combined with smooth motion, I've switched to 24Hz with a medium FRC option and it's a big difference most noticeably with scrolling where things just glide like true 60fps.


In this case madVR is not a good solution for you right now because madVR's pulldown pattern is not regular atm.


Hopefully you find some time soon to address it, glad to see you're aware of it.. can't remember seeing any posts about it.

[6233638]

What I can say is that every DirectShow graph has a reference/master clock. It's usually provided by the audio renderer. But if the audio renderer doesn't provide one, or if there's no audio track, DirectShow will simply pick a different reference clock. In any case, there will always be one master clock, which must be considered "the law" for every playback component.

madVR carefully presents every video frame so that it's visible on screen exactly at the right time, according to the master clock. This is why I always said that I have a hard time believing that madVR could be at fault here. The whole madVR presentation concept is totally hard wired to obey the master clock. So I don't think madVR can be responsible for any audio/video desync problems - unless such problems only occur in specific modes (e.g. only in new windowed presentation mode or so).

I would guess that the audio renderer used by JRiver doesn't properly sync to the master clock for some reason. Which is weird because the audio renderer is usually supposed to *be* the master clock. Of course when activating VideoClock, things kinda change because VideoClock actually tries to modify the master clock to match the VSync rate reported by madVR. But without VideoClock active, the audio renderer should define the master clock simply by its current location in audio playback. If it did that, there should/would be no desync.I'll have to double-check my results regarding the drift I was seeing as it was only 1ms every few minutes, so I suppose it could have been any number of things.
I'm playing to a TV and a USB DAC, so that may have been a contributing factor for example.
I'd have thought if the drift was caused by them, it's not something that enabling VideoClock would have fixed, but it's difficult to say. That in itself is not really a concern when each frame at 24 FPS is 42ms apart. When you have actual numbers for sync, it's surprising what can be tolerated - or how much someone could be off if they set things up manually.
There are so many potential variables to test here though, that it's a laborious process. (madVR options, JRiver options, DAC/Display options etc.)

Certainly, something is not right in JRiver though, since I am seeing so much sync variance after seeking in a video, while MPC-HC seems to avoid this.
That said, the problem does not seem to be there when using EVR in JRiver.
Without this variance, it would take less than 60 seconds to get sync set up using this tester though, and I've probably used it to set up a dozen or so devices now, so it was still worth getting in my opinion. I have always hated trying to set up lip-sync delays manually.

[nevcairiel]

The big difference in MC is that you probably use a lot of DSP processing, which quickly results in a quite huge audio latency, which needs to be compensated for. A simple DirectSound audio renderer in MPC-HC has practically zero latency, short of the hardware buffer requirements. This means that at the start of playback or after a seek, there is usually going to be a slight timing hicup as this latency is being computed and compensated for, which could conceivably cause a video renderer to hicup as well.

[madshi]

Which GPU are you most excited about its potential for madVR that will be released in 2015?
I've no idea. I like Maxwell for its power efficiency. AMD fixing the D3D9 interop performance problem would be something that I find exciting, but I kinda doubt that's going to happen.

I agree with you that what I said wasn't specific, multiples of 24hz of course will be fine, it kinda goes without saying. Many displays only have 23,24,50,59 & 60 available for selection so if 50,59 and 60 are all having the same issue then how I wrote it appears fine to me..
You wrote "MadVR is actually not very good for frame delivery smoothness on I think anything other than 24Hz". Which suggests that the only way to get smooth playback would be to set the display to 24Hz. But that's just not true. E.g. 60fps @ 60Hz playback is smooth, too. Or 25fps @ 50Hz. Or 24fps @ 48Hz. Or 24fps @72Hz. Or 30fps @ 60Hz. Etc etc. All those combinations are smooth, but your original comment, when taken literally, suggested they were not. So your original comment was clearly not "fine". Anyway, doesn't really matter. I knew what you meant. But I also knew that nobody else (other than you and me) would interpret your comment correctly.

Smooth motion doesn't allow the FRC to provide perfect smoothness. It happens occasionally for a couple of seconds then breaks out of it. Then there's the shimmering caused by the blending,
I had it enabled until I realized I was better off without it.
If you turn on motion interpolation in your TV then madVR's smooth motion is not a good idea, of course. You should only use one or the other.

Yup, it's called 24p smooth film. I always preferred to avoid 24Hz because of the screen flicker and I didn't see any advantages as I wasn't testing the right situations.
It's wasn't till I played around with other players/renderers that I saw what was happening.

Instead of using a lower FRC setting on the TV combined with smooth motion, I've switched to 24Hz with a medium FRC option and it's a big difference most noticeably with scrolling where things just glide like true 60fps.

Hopefully you find some time soon to address it, glad to see you're aware of it.. can't remember seeing any posts about it.
The reason why this problem is not well known is that most people either have matching fps / refresh rate setups (which is normally really the best solution), or alternatively they're usually using smooth motion and turn motion interpolation in their displays off.

It seems that your display takes 60Hz input, internally decimates it back to 24fps, and then uses motion interpolation to get to 60fps once again. That is a highly unusual processing pipeline! I would guess that only one TV model in a hundred does something like that. So you actually need 3:2 pulldowned material. That is *really* rare.

Can't your TV accept 24fps input and then display that in 60Hz, using motion interpolation? That would be what most of today's TVs do, and I would expect the best results with this, if you want to use the TV's motion interpolation algorithm.

There are so many potential variables to test here though, that it's a laborious process. (madVR options, JRiver options, DAC/Display options etc.)

Certainly, something is not right in JRiver though, since I am seeing so much sync variance after seeking in a video, while MPC-HC seems to avoid this.
That said, the problem does not seem to be there when using EVR in JRiver.
Unless you have problems in one specific madVR mode (e.g. new windowed presentation path), you can probably rule out the madVR settings. Just leave them alone, they should make no difference. Or if they make a difference, analyzing that difference won't help to get to the bottom of the real cause. At least that is my current understanding of the situation.

[madshi]

The big difference in MC is that you probably use a lot of DSP processing, which quickly results in a quite huge audio latency, which needs to be compensated for. A simple DirectSound audio renderer in MPC-HC has practically zero latency, short of the hardware buffer requirements. This means that at the start of playback or after a seek, there is usually going to be a slight timing hicup as this latency is being computed and compensated for, which could conceivably cause a video renderer to hicup as well.
If the master clock suddenly stalls or skips ahead, this could cause the video renderer to drop or repeat frames, causing visual stutter. However, the video renderer should not produce audio/video desync in this situation, as long as it properly drops/repeats frames (which madVR does).

Or let's say it in a different way: If both the video renderer and audio renderer deliver the video/audio data to the user at the right time (according to the master clock), then there can't be any audio/video desync. So if there is desync in MC then either the video renderer or the audio renderer must have lost sync with the master clock. I don't think it's madVR, so... :)

[Qaq]

A simple DirectSound audio renderer in MPC-HC has practically zero latency, short of the hardware buffer requirements.For a USB DAC I would say WASAPI Exclusive pull mode. I wonder how that would affect Audio (=Reference) clock. Don't know if jRiver supports pull mode though.

[nevcairiel]

If the master clock suddenly stalls or skips ahead, this could cause the video renderer to drop or repeat frames, causing visual stutter. However, the video renderer should not produce audio/video desync in this situation, as long as it properly drops/repeats frames (which madVR does).

Or let's say it in a different way: If both the video renderer and audio renderer deliver the video/audio data to the user at the right time (according to the master clock), then there can't be any audio/video desync. So if there is desync in MC then either the video renderer or the audio renderer must have lost sync with the master clock. I don't think it's madVR, so... :)

You are free to chime in here:
http://yabb.jriver.com/interact/index.php?topic=95901.msg661912#msg661912

I'm out of ideas, the clock seems to be working perfectly.

I could also make the sync testing device available to you if you want to do some deeper investigation.

[leeperry]

I'm out of ideas, the clock seems to be working perfectly.
And how's the sync in MPC-HC, MPC-BE & PotPlayer if you don't mind me asking? I got tremendous post-processing enabled in ffdshow audio(VST plugins etc) and I would say that it all looks right on time to me with Reclock in WASAPI Exclusive mode & Invariant TSC...of course right after seeking things can go a little OOS but they'll quickly get back on track eventually.

I ended up buying a latency tester and have been doing some testing throughout the week.
Oh wow, £228 ouch....that's what I call love.

Great tool though for perfect A/V sync indeed, I relied on a review of my TV that mentioned 28ms input lag and this figure would appear to do wonders with headphones once set as an audio delay in ffdshow.

[Patrik G]

That is probably caused by your display running at 60Hz, so you get the 3:2 pulldown judder.

you also get pulldown judder with 24p content at 50hz right?
thats why i probably can use 50hz from the pc and still motion is great without judder with 24p content.

smooth motion saves it i guess.

the advantage with the kuro running at 50hz is that you get a calm non flickering 100hz image to look at.
it doubles 50hz to 100hz
usually with my old Pioneer BDP LX58 blu ray player i used 72hz (24x3) for blu ray content.
more flicker in the picture.

Edit: are you the one who made madvr?
you really should get paid for this video renderer that crushes all blu ray players! :)
i sold my LX58 right away after i was done with the 3D LUT calibration with madvr.
not only that the picture is pure reference now when using 3D LUT, the LX58 couldnt come close to the details and textures i get here with the chroma upscaling algorithms.
great job!

[Warner306]

Hopefully you find some time soon to address it, glad to see you're aware of it.. can't remember seeing any posts about it.

With regards to madVR's poor implementation of 3/2 pulldown....I barely notice the judder when playing back content at 60hz. The implementation of 3/2 may be poor, but it shouldn't dissuade anyone from using madVR.

Also, I do have a 24p mode on my TV and the playback is better, but not substantially so. But it looks nothing like smooth motion. The frame blending is artificial - a technically sound solution, perhaps, but still artificial. I only went back to 60hz after experiencing some slight lip sync drifting that seemed worse than playing the video with pulldown enabled. Besides, I don't think anyone who uses 24p playback can admit that panning shots do not contain judder and artifacts are not introduced (one of the most offensive examples of artifacts are found on the windows of the building in the opening pan shot of The Dark Knight). Having a perfect matched frame rate does not guarantee perfect playback. There will always be a compromise until the 1/48 second shutter speed of movie theater projectors make their way to the home market. It will never be perfect even at 24hz.

Granted, if it could be improved further, I would be even more happy. But there is no need to be in fear of 3/2 pulldown if your graphics card can't handle smooth motion.

My experience with native 24p playback is similar to this poster: http://www.projectorcentral.com/judder_24p.htm.

I'm sure some TV's are better and others are worse.

[e-t172]

I ended up buying a latency tester and have been doing some testing throughout the week.

Oh wow, £228 ouch....that's what I call love.

I was looking for a latency tester as well, but I just couldn't bring myself to spending this kind of money on a box that listens to a flash and a beep, seems like a rip-off to me. So I made my own instead, using only three components: a cheap microphone (http://www.amazon.co.uk/gp/product/B000185ALA), a brightness sensor (http://www.conrad-electronic.co.uk/ce/en/product/504926/) with known latency characteristics (http://www.produktinfo.conrad.com/datenblaetter/500000-524999/504926-an-01-en-HELLIGKEITSSENS_NAPICA_THR_HOLE_AMS302.pdf), and a jack plug (www.conrad-electronic.co.uk/ce/en/product/304314/). Total: <£10.

The way it works is, I connect one channel of the jack to the microphone and the other to the brightness sensor (which doesn't even need a current source, because a mic input is actually a 5V DC source - isn't that wonderful). Then I play a test video that beeps at the same time the screen switches from full black to full white, and while this plays, I'm recording the "sound" from the device inside an audio editor. Then it's simply a matter of matching the visible step function from the sensor with the sine wave from the microphone:

http://i.imgur.com/9DE0AMvm.jpg (http://imgur.com/9DE0AMv) http://i.imgur.com/3PSYUztm.png (http://imgur.com/3PSYUzt) http://i.imgur.com/6WUMQo9m.png (http://imgur.com/6WUMQo9)

This works by leveraging the fact that in a stereo audio input, the two L/R channels are always kept in perfect sync. Sure, it's very McGyver-ish, but it works great, and it's 22 times cheaper than the alternative. I wouldn't trust it with <10 ms adjustments (because of the latency of the sensor and the likely phase delays from the audio input filter), but this task doesn't require that level of accuracy.

[Asmodian]

The frame blending is artificial - a technically sound solution, perhaps, but still artificial. I only went back to 60hz after experiencing some slight lip sync drifting that seemed worse than playing the video with pulldown enabled.

I can believe the frame blending bothers you but it also sounds like your TV is doing something odd. There should be no lip sync drifting issues in 24 Hz mode v.s. 60 Hz mode.

Besides, I don't think anyone who uses 24p playback can admit that panning shots do not contain judder and artifacts are not introduced (one of the most offensive examples of artifacts are found on the windows of the building in the opening pan shot of The Dark Knight).

I do not have The Dark Knight so I cannot test what you mean but I use a 24p (really 96, 120, or 144 Hz) playback chain and I do not believe judder or artifacts are introduced. How would playing 24p at 24 Hz introduce artifacts?

24p is not fast enough so panning shots are not smooth but using 3/2 pulldown to play at 60 Hz simply adds judder and does not improve anything. Again, maybe your TV is doing something odd like ryrynz's?

Having a perfect matched frame rate does not guarantee perfect playback. There will always be a compromise until the shutter settings of movie theater projectors make their way to the home market.

A perfectly matched frame rate is perfect playback. There are 24 images in the source per second. You display each of them for 1/24th of a second. No artifacts or motion judder are introduced; what is the compromise? What are shutter settings going to do? All methods besides a matched refresh rate have compromises.

edit:
My experience with native 24p playback is similar to this poster: http://www.projectorcentral.com/judder_24p.htm.

Describing what happens using 24p:
"The people seated at the table come apart at the seams, the tuxes flash and strobe, the Casino Royale logo on the card table blinks like a neon sign."

That post doesn't make any sense, what is wrong with their projector?!

I'm sure some TV's are better and others are worse.

Yes, some TVs are totally broken when in 24p mode. This is not a problem with 24p but with those TVs.

[e-t172]

A perfectly matched frame rate is perfect playback. There are 24 images in the source per second. You display each of them for 1/24th of a second. No artifacts or motion judder are introduced; what is the compromise? What are shutter settings going to do? All methods besides a matched refresh rate have compromises.

edit:


Describing what happens using 24p:
"The people seated at the table come apart at the seams, the tuxes flash and strobe, the Casino Royale logo on the card table blinks like a neon sign."

That post doesn't make any sense, what is wrong with their projector?!

What you're saying is not absolutely true if you consider the fact that modern fixed panel displays are "sample and hold" (i.e. the image is displayed for 1/24th of a second), while traditional cinema projectors and CRTs are more of the flicker type (the image is displayed as a very short impulse, each image often displayed twice or thrice to prevent eyestrain).

It is my understanding that motion with flicker-type displays is typically perceived as much smoother and natural compared to sample-and-hold displays. It is even argued that the rationale behind motion interpolation features of modern TVs is to compensate for that difference.

[Warner306]

What you're saying is not absolutely true if you consider the fact that modern fixed panel displays are "sample and hold" (i.e. the image is displayed for 1/24th of a second), while traditional cinema projectors and CRTs are more of the flicker type (the image is displayed as a very short impulse, each image often displayed twice or thrice to prevent eyestrain).

It is my understanding that motion with flicker-type displays is typically perceived as much smoother and natural compared to sample-and-hold displays. It is even argued that the rationale behind motion interpolation features of modern TVs is to compensate for that difference.

Yes, that is all I was saying. Film has always been shot at 24fps, but, when projected, it appears to be moving faster to due to flicker of the shutter. I've read most films are exposed for 1/48 of a second (showing you 24 unique images but 48 flashes of light). This shutter speed is fast enough to be unnoticeable to most human eyes but has the effect of masking the low frame rate of the movie.

When brought to the digital flat panel market, the low frame rate is no longer masked by changing the exposure and panning scenes are revealed as being slow and choppy. The artifacts I've seen could likely be due to the TV's 24p mode more than anything. I don't know.

So native 24hz playback is not quite the exact same experience as the original projection.

[Asmodian]

What you're saying is not absolutely true if you consider the fact that modern fixed panel displays are "sample and hold" (i.e. the image is displayed for 1/24th of a second), while traditional cinema projectors and CRTs are more of the flicker type (the image is displayed as a very short impulse, each image often displayed twice or thrice to prevent eyestrain).

It is my understanding that motion with flicker-type displays is typically perceived as much smoother and natural compared to sample-and-hold displays. It is even argued that the rationale behind motion interpolation features of modern TVs is to compensate for that difference.

Sample and Hold v.s. Flicker is an independent issue. You still want a matched refresh rate.

Movie theaters display each frame 2 or 3 times to avoid nasty flickering. My display also has a flickering mode (5 times per frame viewing 24p) but this doesn't improve the apparent smoothness, it reduces the apparent motion blur. It does work better on high frame rate content so motion interpolation has a benefit there. With multiple flickers you get apparent ghosting as your eyes track moving objects while the object is actually still for multiple refreshes. I do think the bright line refresh of CRTs and shutter movement of a theater projector is better with this than a backlight flickering sample and hold display but again this doesn't change the fact that matched refresh rates are ideal.

[Warner306]

I can believe the frame blending bothers you but it also sounds like your TV is doing something odd. There should be no lip sync drifting issues in 24 Hz mode v.s. 60 Hz mode.


I have fought with having perfect lip sync at 1080p24 for months. I found an entire thread dedicated to this issue in the Kodi forums. In fact, it is how I wound up using MPC in the first place. I thought ReClock would take care of it. I think the blame lies with my 60F7500 Samsung due to the fact my laptop displays the exact same problem using the same software.

The tightest lip sync was achieved with build 1.4.4 of MPC-BE using the improved MPC Audio Renderer. It is near perfect, but my TV still seems to exhibit more drift than in 60hz mode.

Given a ho-hum experience using 24p mode, I don't feel like I am missing a lot by sticking with 60hz playback. I am happy with it as is.

[bytheway]

I've no idea. I like Maxwell for its power efficiency. AMD fixing the D3D9 interop performance problem would be something that I find exciting, but I kinda doubt that's going to happen.


The last time I was looking into GPUs 3-4 years ago nvidia had the big advantage because of CUVID. Is that still the case or are both AMD and nvidia on a level playing field?

[Warner306]

The last time I was looking into GPUs 3-4 years ago nvidia had the big advantage because of CUVID. Is that still the case or are both AMD and nvidia on a level playing field?

If you are using LAV Filters with madVR, DXVA2 copy-back is known to offer better performance than CUVID. The latest LAV Filters (0.64) likely widened this gap even more. Both AMD and Nvidia work well with DXVA2 copy-back.

This thread offers a lot of information about this topic: http://yabb.jriver.com/interact/index.php?topic=80258.0.

[rack04]

On my system with a GTX 770, this driver is bugged with madVR "Error Diffusion - Option 1" dithering. As far as I can tell, this issue does not occur with any of the other dithering modes.

Thin vertical dotted line columns are displayed over the video, evenly spaced exactly 16px apart. (http://i.imgbox.com/BOi3anKj.png) Same issue with 347.12 which was bundled with the Cuda 7 RC, which was the first public driver to have this problem. The previous 347.09 WHQL geforce driver and prior have no issues.

Can anyone else reproduce this?

Can anyone confirm if this has been fixed by Nvidia yet?

[6233638]

I was looking for a latency tester as well, but I just couldn't bring myself to spending this kind of money on a box that listens to a flash and a beep, seems like a rip-off to me. So I made my own instead, using only three components <snip>Very nice work.
That is along the lines of what I had originally hoped to do, but I was unable to find details on how this might be achieved (people doing similar things were using oscilloscopes rather than a sound card) and I actually do occasional AV calibration work so it suits me a lot more to have a stand-alone device like this which can just go with the rest of my kit.
If you're just setting up one display, one source, and you have a laptop with a suitable mic input, that definitely seems like a good option.
I realize that it is expensive for a "basic" device, but it's entirely self-contained, you get immediate results, it incorporates data logging, and anyone can use it. While expensive for what it is, it's probably the least-expensive part of my calibration gear.

Great tool though for perfect A/V sync indeed, I relied on a review of my TV that mentioned 28ms input lag and this figure would appear to do wonders with headphones once set as an audio delay in ffdshow.There are a lot more things which affect latency than the display. Even with the display, many settings may influence the latency. (refresh rate, picture mode etc.)
My display is about 20ms when measured at 60Hz in game mode--the number you are likely to see published in a review--but I need a total correction of 156ms for AV sync when everything is taken into account. (display at 24Hz, in film mode rather than game, separate audio device latency, video player latency, VST plug-in latency etc.)

[TheRyuu]

However, I don't think I'll fix #256, as discussed/explained earlier here in the forum. Furthermore the cardinal value 256 is outside of 8bit, so I can't fix it, anyway, as long as madVR is limited to 8bit output (just kidding).

I'm afraid I don't follow. Nothing you're saying is related to that bug report... are we looking at the same one?

[resides]

If you are using LAV Filters with madVR, DXVA2 copy-back is known to offer better performance than CUVID. The latest LAV Filters (0.64) likely widened this gap even more. Both AMD and Nvidia work well with DXVA2 copy-back.

How does Intel QuickSync compare? I'm using LAV and having an i7 with Intel HD Graphics. My reasoning is to use the CPU/GPU (Intel HD GFX) for video decoding and to use the NVIDIA GPU for madVR.

I do notice the most gains with madVR, but not too much in using the various hardware acceleration modes like NVIDIA CUDA, Intel QuickSync, DXVA2 (both native and copyback). All look the same to me.

[ryrynz]

The reason why this problem is not well known is that most people either have matching fps / refresh rate setups (which is normally really the best solution), or alternatively they're usually using smooth motion and turn motion interpolation in their displays off.


I figured there would be way more people playing on monitors at 60Hz than anything else TBH and most of them playing 24fps content.


Can't your TV accept 24fps input and then display that in 60Hz, using motion interpolation? That would be what most of today's TVs do, and I would expect the best results with this, if you want to use the TV's motion interpolation algorithm.


I'm fairly certain it does just that.

Using 60Hz with frame creation creates a few issues for me. First one of course is MadVR's 3:2 display breaking it, next switching panning directions frame creation stutters for a second or two and lastly when seeking on occasion there will be massive stuttering that only a new seek fixes, frame creation is not very forgiving it seems. Running at 24Hz fixes all of these.


24p is not fast enough so panning shots are not smooth but using 3/2 pulldown to play at 60 Hz simply adds judder and does not improve anything. Again, maybe your TV is doing something odd like ryrynz's?


Mine isn't doing anything weird, it's just not tolerant of MadVR's 3:2 pulldown pattern.


I don't think anyone who uses 24p playback can admit that panning shots do not contain judder and artifacts are not introduced (one of the most offensive examples of artifacts are found on the windows of the building in the opening pan shot of The Dark Knight)


If I use 24Hz and disable frame creation everything looks fine to me, although honestly I prefer 60Hz with Smooth Motion instead of just 24Hz without frame creation.
I can't imagine how "artifacts are introduced" when you're playing it in it's unadulterated form. Really there is no way around it, if you want "smooth motion" you need to pay the price
with artifacts to do it, whether it's intelligent frame creation on the display, SVP or MadVR's Smooth Motion or something similar. Personally I can't stand juddering pans and scrolls.. so
frame creation it has to be for me. I see the odd artifact (a sort of accidental estimation) that even the latest TVs still do but considering what it gives me in smoothness I can't complain.
If you don't have something similar run SVP.. it's quality isn't that far behind what you'll find in today's 100Hz or similar TV's. It's demanding though and It will crank your fan speed up.

Quick OT. Can anyone care to enlighten me why there's no judder during panning at the cinemas with what would be standard 24fps films?

[kalston]

Quick OT. Can anyone care to enlighten me why there's no judder during panning at the cinemas with what would be standard 24fps films?

I always see the 24fps judder at the cinemas actually. I didn't really notice back in the days (like 7 years ago or so), but once I got a 72hz display at home and started using madVR, I suddenly started paying attention to that kind of thing.

24fps is just too low a framerate.

[ice25]

The tester (http://www.sync-one2.co.uk/) just measures the delay between a flash on-screen and a tone being played.
It wouldn't be able to test how stable the video presentation is compared to a Blu-ray player, if that's what you mean.

Been using this same device for about 4 months now, sure it's expensive but i couldn't live without it at this point. Just having perfect A/V sync and more importantly knowing that you are absolutely in sync with what you see on the screen from the distance and position you are sitting is very satisfying.
Before i had a way to check the audio sync i was never really sure if everything was perfectly synced. It's crazy how far you can be out of sync but still think everything is fine.
What i noticed is that there are definately some settings in various software (reclock, mpc, madvr, etc) that have an influence on the A/V sync, and without any ability to check what the result is on the sync it's impossible to say if you are in front of behind with the audio.

[madshi]

you also get pulldown judder with 24p content at 50hz right?
thats why i probably can use 50hz from the pc and still motion is great without judder with 24p content.

smooth motion saves it i guess.
Yes, you would get pulldown judder with 24p @ 50Hz, it just wouldn't be 3:2, but something completely different. And thanks for the feedback!

Also, I do have a 24p mode on my TV and the playback is better, but not substantially so. But it looks nothing like smooth motion.
I would guess that your display then internally reapplies 3:2 pulldown judder and internally refreshes at 60Hz. Just a guess, of course...

there is no need to be in fear of 3/2 pulldown
Well, for me pulldown judder is extremely annoying. I absolutely hate it. That might have to do with me living in PAL land, where movies are sped up from 24p to 25p by increasing the playback rate, and then they're shown in 50Hz. So there's never been pulldown here in PAL land.

I'm afraid I don't follow. Nothing you're saying is related to that bug report... are we looking at the same one?
All I've said was meant to be a joke, because your bug report happened to have the ID 0x100. But I was serious about not "fixing" your bug report 256. You had reported this earlier in this thread, and I had explained why it's the way it is, and why I don't plan on doing anything about it. My opinion hasn't changed about that.

I figured there would be way more people playing on monitors at 60Hz than anything else TBH and most of them playing 24fps content.
There are many people playing on computer monitors at 60Hz, that's absolutely correct. But most computer monitors don't have any sort of motion interpolation. That's usually a feature offered by TVs, not by computer monitors. And those monitors/TVs which do offer motion interpolation, don't usually take 60fps input, decimate it back to 24fps and then motion interpolate it up to 60fps. It's really rare and weird what your TV is doing there.

Using 60Hz with frame creation creates a few issues for me. First one of course is MadVR's 3:2 display breaking it, next switching panning directions frame creation stutters for a second or two and lastly when seeking on occasion there will be massive stuttering that only a new seek fixes, frame creation is not very forgiving it seems. Running at 24Hz fixes all of these.

If I use 24Hz and disable frame creation everything looks fine to me, although honestly I prefer 60Hz with Smooth Motion instead of just 24Hz without frame creation.
Why don't you use 24Hz and enable frame creation? Shouldn't that produce identical results to sending 60Hz with enabled frame creation?

[ryrynz]

There are many people playing on computer monitors at 60Hz, that's absolutely correct. But most computer monitors don't have any sort of motion interpolation. That's usually a feature offered by TVs, not by computer monitors. And those monitors/TVs which do offer motion interpolation, don't usually take 60fps input, decimate it back to 24fps and then motion interpolate it up to 60fps. It's really rare and weird what your TV is doing there.


I'm not aware of any monitors that do, it doesn't make much sense to add it considering the application, it's almost a shame really.

What is it about my results that makes you think it does 60->24->60?


Why don't you use 24Hz and enable frame creation? Shouldn't that produce identical results to sending 60Hz with enabled frame creation?

That's what I'm doing, identical results without those three issues I mentioned, two if I use something other than MadVR which is almost as good..
But I just hate anything breaking the smoothness of frame creation so 60Hz is out during video playback. Never thought 3:2 and 60Hz would be such a downer for video playback.. live and learn.

[madshi]

What is it about my results that makes you think it does 60->24->60?
Well, you said that if you feed your display 60Hz with a proper 3:2 pulldown cadence, with enabled frame interpolation, then you get perfectly smooth movie credits. That's only possible if the TV does 60 -> 24 -> 60. You can't get smooth scrolling credits if the pulldown judder is still there. So if you get smooth results, your TV must be removing the pulldown judder.

That's what I'm doing, identical results without those three issues I mentioned, two if I use something other than MadVR which is almost as good..
But I just hate anything breaking the smoothness of frame creation so 60Hz is out during video playback. Never thought 3:2 and 60Hz would be such a downer for video playback.. live and learn.
I'm still not 100% sure if I understand you correctly. Why do you wish to use 60Hz at all? Using 24Hz output of your GPU with enabled frame interpolation in your TV sounds like the perfect setup, if you like frame interpolation. Why do you have any desire to output 60Hz from your PC? Doing so only makes life harder for your TV. I don't see which benefit you're hoping to get from that...

[ryrynz]

Well, you said that if you feed your display 60Hz with a proper 3:2 pulldown cadence, with enabled frame interpolation, then you get perfectly smooth movie credits. That's only possible if the TV does 60 -> 24 -> 60. You can't get smooth scrolling credits if the pulldown judder is still there. So if you get smooth results, your TV must be removing the pulldown judder.


Oh yeah, that's pretty cool now that I think about it.


I'm still not 100% sure if I understand you correctly. Why do you wish to use 60Hz at all?


I don't, I've only stated how it doesn't work as well, but in some cases does a fairly good job. It's just not good enough that I'd want to continue using it.


Using 24Hz output of your GPU with enabled frame interpolation in your TV sounds like the perfect setup

It works nicely, I only wish the TV would work with 0-255 via HDMI rather than just DVI, doesn't seem to be any way around it other than removing the audio block (not good when you have a receiver in the chain)
I tried a HDMI to DVI adapter but that didn't change anything, ah well It's not perfect but at least it's smooth. :o

[madshi]

Ah, I see. So the imperfect 3:2 pulldown used by the current madVR build doesn't really bother you at all. You just noticed it, correct? I think since the introduction of smooth motion FRC most users don't use madVR's pulldown, anymore. Instead most users either have a matching framerate / refresh rate setup, or they use smooth motion FRC. So although I'm aware of that the pulldown needs an improvement, I don't think it's very urgent...

[ryrynz]

Ah, I see. So the imperfect 3:2 pulldown used by the current madVR build doesn't really bother you at all. You just noticed it, correct? I think since the introduction of smooth motion FRC most users don't use madVR's pulldown, anymore. Instead most users either have a matching framerate / refresh rate setup, or they use smooth motion FRC. So although I'm aware of that the pulldown needs an improvement, I don't think it's very urgent...

Doesn't really bother me any when using the PC if Smooth Motion is enabled. On the TV with FRC on medium the pulldown issue breaks the FRC quite badly. I suspect anyone using 59/60Hz with any frame creation on is likely to run into the same issue, but how many are there out there doing that? Nobody's really mentioned anything up until this point so.. no big deal.

Not sure if the pulldown problem is related at all, but sometimes I don't have consistent smooth frame delivery on the PC. It's just the odd small jump whilst replaying panning scenes. It could just be that it occurs after seeking (clock related?) Will maybe do a comparison once 0.88 is out.

[madshi]

I suspect anyone using 59/60Hz with any frame creation on is likely to run into the same issue, but how many are there out there doing that?
Anyone using frame creation is probably using 24Hz output, which is really the best solution.