View Full Version : madVR - high quality video renderer (GPU assisted)
tp4tissue
21st July 2019, 04:38
I think the best target is 4K. videos a@ 1080@24fps, 4k@24fps and 4K at 60fps and HDR tone mapping for 4K@60fps possibly as well.
Will wait a couple of weeks for you to create the thread and we'll go from there.
Problem is variable performance within the same video. It's hard to pick the exact read spot for the video because it's unstable.
There also isn't enough resolution in the timer to tease out differences between cards that are close.
huhn
21st July 2019, 04:49
i have an idea for a rough test.
there is nothing worse then over doing the settings and making it to complicated and it doesn't have to be very accurate.
ryrynz
21st July 2019, 04:50
Yeah. That's why we'll just use screenshots of graphs from the log data and videos that are long enough to give good averages. I'm thinking about 2 mins. If you want to discuss feel free to msg me.
Actually let's just start a discussion thread first then once we've defined parameters then we'll make a new one for results.
mclingo
22nd July 2019, 14:37
I was trying to think of a dirty way of creating a tool or at least something that was portable, something we could all use, something that forces us all to use the same settings at least, other than just using stock settings. This idea might be possible but its dirty as hell but could be refined if someone knows how to create scripts and compile exe's.
someone could create an extracting exe to do something like the following
Creates a folder c:\MADVR TEST
Runs a script which queries the current location of MADVR install and renames is MADVR.old
extracts a proconfigured MADVR install to c:\MADVR TEST\MADVR
extracts a proconfigured portable MPC-HC program to c:\MADV TEST\MPC
extracts a video file to c:\MADVR TEST
Runs a script which installs MADVR
Runs a scripts which runs the video file so user can take a screen shot
When movies closes a script renames MADVR.old back to MADVR and installs it again.
Script delete c:\MADVR TEST
Would this work though, would you be able to run MPC portable without setting the decoder or configure anything in MADVR?
ryrynz
23rd July 2019, 00:07
I wouldn't put too much effort into it right now as there's a chance madshi might re-release his benchmarking version sometime in the near future. The goal imo is to simply determine average performance, not to make a quick n dirty tool providing more accurate and comparable results. We've made do for this long.. Just need to decide on content, settings, tools are reporting method.
HillieSan
23rd July 2019, 09:20
👍 sounds good.
mclingo
23rd July 2019, 13:23
I hope madshi does re-release the tool, I would say its more essential now than it has ever been due to the sheer number of cards and driver version available, it would be good to compare results from different driver versions as well as cards ! :)
huhn
23rd July 2019, 14:41
benchmark by doing as many frames as possible is not reliable test too each shader has an overhead of about 0.30 ms so a high end card would loose more performance to that then a try level card.
el Filou
23rd July 2019, 17:45
If you log with a monitoring tool (I've read madVR doesn't like GPU-Z and the like but I've personally never witnessed any difference in performance when it's running) and multiply the GPU clock by the load percentage, then divide that number by the max boost clock (which can usually be found in reviews, or just max out your madVR settings to find it), you can get an estimation of the relative work that is needed from the GPU for a certain madVR config.
It's not a 100% accurate evaluation but it's closer than "this GPU can probably do X scaling algo at Y source and Z output resolution"
HillieSan
23rd July 2019, 17:49
benchmark by doing as many frames as possible is not reliable test too each shader has an overhead of about 0.30 ms so a high end card would loose more performance to that then a try level card.
I don’t understand your logic. The performance of the NGU algorithm translates to a benchmark of many frames as possible for certain “ideal” settings. I can imagine that an expensive graphics card may not be the winner. Let the tool determine the best cards for the NGU algorithm (=madvr) and settings. Just my 2 cents.
huhn
23rd July 2019, 18:44
the 2 finalsteps currently take about 2 ms.
to make math simple if you manage to reach 250 FPS with the giving setting you are currently loosing 50% of the performance to the finalsteps this is not an accurate representation of the card.
that a problem in game benchmarking too but it is far far less of an issue there we have way more steps then the finalsteps.
If you log with a monitoring tool (I've read madVR doesn't like GPU-Z and the like but I've personally never witnessed any difference in performance when it's running) and multiply the GPU clock by the load percentage, then divide that number by the max boost clock (which can usually be found in reviews, or just max out your madVR settings to find it), you can get an estimation of the relative work that is needed from the GPU for a certain madVR config.
It's not a 100% accurate evaluation but it's closer than "this GPU can probably do X scaling algo at Y source and Z output resolution"
if my card jumps between 1000 and 700 mhz the load is going crazy i would get number apart by more then 100 % for the same load making the matrix pretty inaccurate at best.
Bryanhoop
23rd July 2019, 18:44
My GPU can't keep up with NGU doubling on low when playing low-resolution 50fps PAL content (Antiques Roadshow), but is fine playing basically all other content at even NGU high.
Is there a way to modify the upsampling algorithm when playing >=50fps content?
huhn
23rd July 2019, 18:50
check out the render profiles on page one 3rd post.
el Filou
23rd July 2019, 19:28
if my card jumps between 1000 and 700 mhz the load is going crazy i would get number apart by more then 100 % for the same load making the matrix pretty inaccurate at best.It keeps changing that much? That's interesting, on Adaptive mine always stabilizes after less than a minute on some clock/load combination, even with light stuff.
Maybe rendering and queue settings influence this.
Shiroe
24th July 2019, 04:26
Is there a madVR setting that increases playback FPS? When using LAV Filters Megamix mid-tier (NGU medium) preset, PotPlayer's OSD information displays this:
https://i.postimg.cc/Zb7hVpst/zuikaku-S9-YJb-KMTj-M.jpg
One line reads:
"Output: P010, 2048x1080(1.9:1/1.78:1), FPS: 23.976->59.94"
If I use another video renderer, e.g., EVR-CP, the FPS would read:
"23.976->23.976"
Just in case, madVR's OSD information displays this:
https://i.postimg.cc/Bs7mQM1f/zuikaku-c6-Xopp-Q0o2.jpg
If a madVR setting is increasing playback FPS, is that setting affected by the Windows version?
The reason I ask is using LAV Filters Megamix's mid-tier (NGU medium) preset on Windows 10 to watch anime gave me 60 FPS. I switched to Windows 8.1 for reasons, and the same preset gave me 30 FPS although now it is giving me 60 FPS, even though I made no changes.
ryrynz
24th July 2019, 06:13
Your refresh rate is 60Hz (you are asking your display to show 60fps), madVR in your particular case is blending frames AKA Smooth Motion (Seen enabled in your OSD)
If you are not blending frames this way say with another renderer, then you have judder as a result of having a 3:2 pulldown as frames are displayed in a 3-2 pattern.
Keep Smooth Motion on and enjoy, madVR is the best renderer available for your video content so stick with it.
HillieSan
24th July 2019, 08:00
the 2 finalsteps currently take about 2 ms.
to make math simple if you manage to reach 250 FPS with the giving setting you are currently loosing 50% of the performance to the finalsteps this is not an accurate representation of the card.
that a problem in game benchmarking too but it is far far less of an issue there we have way more steps then the finalsteps.
if my card jumps between 1000 and 700 mhz the load is going crazy i would get number apart by more then 100 % for the same load making the matrix pretty inaccurate at best.
I see. I remember that on my RX480 madvr did it’s job fine when I keep the latency below 15 ms (not sure if it was exactly 15 ms). Any swing of the clock didn’t matter as long as it stays around 15 ms. Is this useful for a benchmark? I had to keep the error dithering simple (I think ordered setting) to allow NGU high for most movies. This helped to drop the latence under the 15 ms. This latence was my reference for performance tuning.
ryrynz
24th July 2019, 09:05
Like Nev said the benchmark runs the card at full speed so clock speeds won't be a concern. Mostly we just want to test/compare NGU and HDR algorithm performance.
chros
24th July 2019, 10:27
Your refresh rate is 60Hz (you are asking your display to show 60fps), madVR in your particular case is blending frames AKA Smooth Motion (Seen enabled in your OSD)
Or switch refresh rate of the display to 23p if that's possible and turn smooth motion off.
Shiroe
24th July 2019, 12:08
Your refresh rate is 60Hz (you are asking your display to show 60fps), madVR in your particular case is blending frames AKA Smooth Motion (Seen enabled in your OSD)
If you are not blending frames this way say with another renderer, then you have judder as a result of having a 3:2 pulldown as frames are displayed in a 3-2 pattern.
Keep Smooth Motion on and enjoy, madVR is the best renderer available for your video content so stick with it.
So that is why I had judder when using video renderers other than madVR. madVR also has a setting to change automatically display modes. Can that setting replace Smooth Motion, and is it better or worse than Smooth Motion, which is to ask does Smooth Motion have downsides?
ryrynz
24th July 2019, 12:20
Or switch refresh rate of the display to 23p if that's possible and turn smooth motion off.
That's If he can, quite a few screens only run at 50 and 60hz.
So that is why I had judder when using video renderers other than madVR.
Mostly, yes. It can replace Smooth Motion as chros mentioned but recommended only if you get good results from it.
Some may prefer Smooth Motion instead, if your screen can change refresh and if you can set it up so you don't get any frame drops or repeats
over the course of a typically watched video then the benefits are in motion quality. Smooth Motion blends frames so you you may notice a slight loss of picture sharpness with it enabled.
el Filou
24th July 2019, 14:47
https://i.postimg.cc/Bs7mQM1f/zuikaku-c6-Xopp-Q0o2.jpg
The reason I ask is using LAV Filters Megamix's mid-tier (NGU medium) preset on Windows 10 to watch anime gave me 60 FPS. I switched to Windows 8.1 for reasons, and the same preset gave me 30 FPS although now it is giving me 60 FPS, even though I made no changes.Is there a reason you're forcing NGU for 1080 source material on a 1080 display? Is the 'preset' you're talking about causing it to be forced on?
Shiroe
24th July 2019, 17:51
Is there a reason you're forcing NGU for 1080 source material on a 1080 display? Is the 'preset' you're talking about causing it to be forced on?
I have always used codec packs for their simplicity to achieve proper playback of and excellent picture quality when watching anime. I first used Kawaii Codec Pack, but I have since moved on to LAV Filters Megamix because the former is no longer maintained. I also prefer PotPlayer, which LAV Filters Megamix includes, to MPC-HC. During the installation process, one can select different presets based on their graphics card and the resolution of their display. I choose the preset closest to mine.
https://postimg.cc/qNYz8y85
https://postimg.cc/KKBvBVc1
I was using a 1600x900 monitor, but I have been using my 1920x1080 Smart TV since I switched to it recently for testing purposes.
I recently added a 1060 3GB to a Dell XPS 410, which initially resulted in problems, hence "testing purposes." I don't get any frame drops during playback, except in rare cases when an English translation replaces Japanese text in a video. I can think of only two instances when this has happened, both in the same series. Otherwise, I get frame drops temporarily only when exiting/entering fullscreen and seeking.
Do you have a suggestion?
Asmodian
24th July 2019, 18:56
Since it includes madVR presets that force NGU without scaling it makes me think that pack should not be used.
My recommendation is to reset madVR to defaults, only change a setting if it looks better to you or at least you know why you are changing it.
QBhd
24th July 2019, 19:33
I've said for nearly 2 decades... codec packs are terrible! Sure, one download and set-it-and-forget-it is appealing, but too many problems arise. I will say it is a good way to dip your toes into the pond, but once you actually take the plunge, a codec pack is never something to go back to.
QB
Shiroe
24th July 2019, 21:48
Since it includes madVR presets that force NGU without scaling it makes me think that pack should not be used.
My recommendation is to reset madVR to defaults, only change a setting if it looks better to you or at least you know why you are changing it.
By "scaling," do you mean using profiles to configure NGU based on the resolution and framerate of content? Since I got a GTX 1060 3GB, when I install LAV Filters Megamix, I create profiles for SD, HD, and FullHD content. I use high settings for SD and HD content, but for FullHD content, I use LAV Filters Megamix's mid-end preset's NGU defaults because I have not needed to change it; a drop in framerate is infinitesimally rare during playback.
I see a difference in picture quality between LAV Filters Megamix's low-end and mid-end presets, at least. The former uses Bicubic50/ Catmull-Rom for chroma upscaling and Lanczos 3 taps for image upscaling without anti-ringing filters while the latter uses NGU Sharp medium for chroma upscaling and image upscaling.
NGU Sharp can make anime look very good, but it sharpens too much sometimes, so I recently began testing NGU Anti-alias, which seems common among people who use madVR to watch anime. Generally, I like sharpened anime because my eyesight is poor; a clear video is less eye-strenuous. I am still experimenting.
Shiroe
24th July 2019, 21:53
I've said for nearly 2 decades... codec packs are terrible! Sure, one download and set-it-and-forget-it is appealing, but too many problems arise. I will say it is a good way to dip your toes into the pond, but once you actually take the plunge, a codec pack is never something to go back to.
QB
I have not had problems using Kawaii Codec Pack or LAV Filters Megamix. I have had problems when installing and configuring PotPlayer, madVR, LAV Filters, and AssFilterMod on my own.
Asmodian
24th July 2019, 22:33
By "scaling," do you mean using profiles to configure NGU based on the resolution and framerate of content?
No, I use many resolution and framerate profiles myself. By scaling I mean changing the resolution, i.e. resizing. Setting the Image Upscaling option "activate doubling / quadrupling" to "always - supersampling" is a bad idea. If you want to do upscaling refinements at a higher resolution it can be useful but as a normal default it is bad. I always use 1.5x/3.0x myself.
I agree, NGU AA is great for anime and NGU Sharp often makes anime look a bit "off"... for lack of a better description. Something is wrong with the edges. I like NGU Sharp with some very high quality anime but most anime blurays look better with NGU AA. :(
This is with a 2080 Ti on a LG C9 OLED from pretty close to the screen, NGU Sharp looks fine if I am much further away.
I link my madVR settings because people do ask for them but I recommend setting up madVR yourself. Any settings pack is going to pick something for a bad reason, at least for your system.
The reason you need average rendering times <15ms is for 60 fps content, to figure out the number for any particular frame rate simply divide 1000 by the framerate and then subtract a couple of ms for overhead. 1000/60 = 16.67 ms. With 24 fps content you can get away with much stronger settings, even 35ms or more should be fine, 1000/24 = 41.67 ms. Profiles for different frame rates are important, this pack's presets do not include any?! :mad:
I have not had problems using Kawaii Codec Pack or LAV Filters Megamix.
Now you have. :p
Limberjork
25th July 2019, 04:21
The "The GPU rejected this mode" is unfortunately a GPU driver bug, not much I can do about it. I've already reported it to Nvidia, a fix might be coming at some point in the future, but probably not very soon.
Does anyone know if Nvidia ever fixed this? I just learned about the custom modes tab and I get the "GPU driver rejected this mode" error with every optimized timing option I try.
70MM
25th July 2019, 06:10
I havent had any instant email notifications when someone posts on this thread for many many months even though I have them turned on. It has worked perfectly for years but not for the last 4 months or so. I have loged out and loged in, and have this thread ticked to recieve instant notifications but still I can no longer get them. Has this happened to anyone else?
Mike5
25th July 2019, 07:17
Yes, I am in the same situation, no notification for months. Previously it worked perfevtly.
nevcairiel
25th July 2019, 07:27
Does anyone know if Nvidia ever fixed this? I just learned about the custom modes tab and I get the "GPU driver rejected this mode" error with every optimized timing option I try.
Just use CRU (https://www.monitortests.com/forum/Thread-Custom-Resolution-Utility-CRU) for custom resolutions, it always works fine.
70MM
25th July 2019, 11:31
Yes, I am in the same situation, no notification for months. Previously it worked perfevtly.
There must be something up, I went to the "Contact Us" page and it doesnt allow you to email Doom..... Oh well:confused:
el Filou
25th July 2019, 13:35
By "scaling," do you mean using profiles to configure NGU based on the resolution and framerate of content?
[...] for FullHD content, I use LAV Filters Megamix's mid-end preset's NGU defaults because I have not needed to change it; a drop in framerate is infinitesimally rare during playback.If you open madVR settings, are there different scaling profiles setup for the different resolutions?
If yes, then go to the profile for Full HD (1080), in image upscaling change 'activate doubling / quadrupling' to something else than 'always'. 'only if scaling factor is 1.5x' is a good option.
If there are no profiles set-up in madVR, it means your codec pack is forcing the same madVR settings, including NGU, for every source material resolution, which in the case of playing 1080p on your 1080 screen, is very inefficient.
thighhighs
25th July 2019, 17:14
No. The author of the pack did not add any profiles. But he mentioned such a possibility in his guide (https://anime.my/tutorials/madvr/).
This man is old madVR user, but some points confused me. Anyway, i always thinking LAV Filters Megamix suitable for anime-fans only.
ScrooW
25th July 2019, 18:29
Guys I have some doubts related to the audio and video sync more precisely the lip sync.
Is madvr's smooth motion immune to audio delay?
I ask this because the problem of synchronization of contents 23.xx / 24p is famous.
I'm almost freaking out because I can't not notice the lack of sync and even searching a lot on the internet and putting the "universal" delays of 175 / 200ms, in some content comes back all over again.
Limberjork
25th July 2019, 21:34
Guys I have some doubts related to the audio and video sync more precisely the lip sync.
Is madvr's smooth motion immune to audio delay?
I ask this because the problem of synchronization of contents 23.xx / 24p is famous.
I'm almost freaking out because I can't not notice the lack of sync and even searching a lot on the internet and putting the "universal" delays of 175 / 200ms, in some content comes back all over again.
I can't see any reason that smooth motion would cause any sync issues. I think you'll need to provide more information about your setup and what types of files you're having issues with if anyone is going to be able to help you. Also I think your issue is almost certainly not with madVR so maybe this thread isn't the best place to get help.
ScrooW
25th July 2019, 22:15
I can't see any reason that smooth motion would cause any sync issues. I think you'll need to provide more information about your setup and what types of files you're having issues with if anyone is going to be able to help you. Also I think your issue is almost certainly not with madVR so maybe this thread isn't the best place to get help.
The first part of my post was to exemplify and the final was the question.
I will exemplify and in the end I will ask the question, otherwise things will be nonsense.
As we know with 23.xx / 24fps running on tvs with the tv set to 23 / 24p there is an audio delay, I think nowadays with modern tvs this is small or over, but if you google it "24hz lip sync "will find many articles, posts in current and former forums.
So my question and I believe you have already answered, activating smooth motion in madvr will not have this problem?
Asmodian
25th July 2019, 22:22
No, madVR times the frame blending to exactly match the audio. madVR always uses the audio clock to figure out how many frames to show per second, dropping or repeating frames to maintain sync when not using smooth motion.
ScrooW
25th July 2019, 23:04
No, madVR times the frame blending to exactly match the audio. madVR always uses the audio clock to figure out how many frames to show per second, dropping or repeating frames to maintain sync when not using smooth motion.
About the last part of your post, I activating smooth motion won't have problems with audio sync either, right?
Sorry for so many questions, I just want to come to a "conclusion".
Thanks for the reply and you got where it was really my question.
Asmodian
25th July 2019, 23:14
Sorry, smooth motion is the frame blending I mentioned. Smooth motion syncs to the audio even more perfectly than with it off, the difference is unimportant but smooth motion is the most accurate.
You will not have any sync issues when using smooth motion.
ScrooW
25th July 2019, 23:19
Sorry, smooth motion is the frame blending I mentioned. Smooth motion syncs to the audio even more perfectly than with it off, the difference is unimportant but smooth motion is the most accurate.
You will not have any sync issues when using smooth motion.
Thank you so much for your attention and patience.
XTrojan
27th July 2019, 23:21
Hi! Been reading around here in the forums about OS HDR vs Nvidia HDR. Is it still recommended to use NV HDR over OS HDR or are they the same thing by now? I've red that OS HDR handles everything by itself and sends no metadata but other posters have disputed this.
Asmodian
28th July 2019, 00:04
As far as I am aware the situation has not changed in anyway. Use NV HDR because Windows' is a one size fits all solution that will mess with the video while madVR can control Nvidia's HDR.
-Michael-
28th July 2019, 17:36
Is it possible to save a movie with the settings of madvr? For example, if I have an SD video, can I convert it to HD with the madvr upscaler?
huhn
28th July 2019, 18:07
this is practical possible but madVR is not design for that and there are multiply problems with recording.
-Michael-
28th July 2019, 18:15
How can this be done?
-Michael-
28th July 2019, 18:24
It would be wonderful madvr could be used with premiere adobe or other similar software. Until now, I haven't found any software that makes the upscaler like Madvr. Precisely for this reason I asked if there is a way to save a movie using the madvr upscaler.
el Filou
28th July 2019, 18:52
If you use a screen recorder that outputs AVC or HEVC using the hardware encoders available with modern graphics cards, or something like FRAPS if you want lossless output (other better apps can be available, I haven't followed recent development), you should be able to grab the result of madVR rendering that way. I've never tested it so can't guarantee the results, try it.
madVR outputs with Direct3D so it should work like when recording a game. (Edit: the only exception is if you use overlay rendering, I think)
Annisman
29th July 2019, 00:43
Hey guys, when I play 4K HDR content MadVR (cntrl+j) reads 'BT2020 -> DCI-P3.... Can someone tell me is this what I want ? What does that mean ?
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.