View Full Version : madVR - high quality video renderer (GPU assisted)
jmone
26th October 2018, 20:47
I created some instructions for those who are using current test builds wanting to try madMeasureHDR. Those at AVSForums are already well-versed on the current methods.
Instructions: Using madMeasureHDR to create dynamic HDR10 metadata (https://forum.kodi.tv/showthread.php?tid=223175&pid=2750406#pid2750406)
This is the only place this information is published and I'll update it as the tool changes.
Thanks for the read & I'll give it a go but a couple of Q on how this works. I keep all my UHD BD's as full folder rips and sometimes play them using the BD Menu and sometimes the direct MPLS approach. Some of these discs have seamless branching with various play lengths. Does madMeasureHDR measure then store the HDR10 meta data by a time code per M2TS and the during playback keep track of what M2TS is being used at any particular time?
HillieSan
26th October 2018, 21:14
64-bit processing can be considerably faster and less prone to memory issues.
Can you explain this? I am familiar with the 32-bit and 64-bit programming models, but this is new to me for video processing.
Manni
26th October 2018, 21:22
Thanks for the read & I'll give it a go but a couple of Q on how this works. I keep all my UHD BD's as full folder rips and sometimes play them using the BD Menu and sometimes the direct MPLS approach. Some of these discs have seamless branching with various play lengths. Does madMeasureHDR measure then store the HDR10 meta data by a time code per M2TS and the during playback keep track of what M2TS is being used at any particular time?
AFAIK the tool doesn’t support bd folders, as it can’t process more than one file. You need to rerip as mkv (makemkv can use your existing bd folder as a source, it might be taster than reripping the physical disc).
nevcairiel
26th October 2018, 21:48
AFAIK the tool doesn’t support bd folders, as it can’t process more than one file. You need to rerip as mkv (makemkv can use your existing bd folder as a source, it might be taster than reripping the physical disc).
I would expect it just uses DirectShow to read the files, in which case it should be able to record for anything that LAV Splitter can read for you, no?
SamuriHL
26th October 2018, 22:40
I hope madshi creates a 64-bit build of that tool, because 64-bit processing can be considerably faster and less prone to memory issues. In fact, who doesn't run 64-bit these days, and with such a standalone tool there is no compatibility to consider, so personally I would go exclusively 64-bit.
I asked, he said no plans to do so. :( That he didn't see the value in doing so. I tend to agree with you.
xabregas
26th October 2018, 22:50
Hi,
Just a quick question. For 23,976 and 25 fps movies played in 60HZ monitor what is the safe rendering time on average and max stats?
TIA
Manni
26th October 2018, 22:58
I would expect it just uses DirectShow to read the files, in which case it should be able to record for anything that LAV Splitter can read for you, no?
When I asked Madshi, he didn't reply but I've just done some tests and you are correct, it looks like the tools supports index.bdmv files. :)
EDIT: I spoke too fast, the utility quits after a few minutes.
EDIT2: I tried again and this time it created a measurements file, but it's not taken into account, whether I drop the whole folder in MPC-BE or just the index.bdmv file. Let's wait until we hear from Madshi to know if it's supposed to be supported or not.
Warner306
27th October 2018, 00:43
I followed the guide, the measurement file gets created, but it doesn't show any changes in measured frame and scene in MadVR OSD when i play the movie. Tried both 32 bit and 64 bit MPC-HC/LAV. What am I missing?
Try DXVA2 (copy-back) instead of D3D11.
gfxnow
27th October 2018, 07:52
I am trying to understand how keyboard shortcuts for the different settings work. After pressing the shortcut keys do I also have to hit F2 for the setting to apply or does it apply instantly?
Secondly, I am trying to create shortcut keys for switching between the image-doubling options so I can decide on the fly whether I want NGU anti-alias or NGU sharp. Is this possible?
DragonQ
27th October 2018, 09:49
I'm looking to upgrade my HTPC to add 2160p HDR support. I have two options: either replace my GT 430 with a newer full-height GPU (maybe passively cooled), or swap out the box for my old TV server box, which is much smaller and would require a half-height GPU. So I have 2 questions:
- Does anyone have any experience with the Gigabyte GTX 1050 Ti OC Low Profile 4G? Is it likely to be silent whilst doing madVR duty?
- Is there a chance of a superior madVR HTPC GPU any time soon? Like a GT 2030 or 2050? What about an AMD RX560 would that work as well?
mytbyte
27th October 2018, 09:53
Try DXVA2 (copy-back) instead of D3D11.
Yeah, thanx but it didn't work (and it's veeeeery slow compared to D3D11). It's a GT 1030 (I'm at my weekend location). It won't even start measuring if "measure each frame's peak luminance" is not checked in MadVR (it's only logical?) but I assume I need to deactivate it if I want the measure file to be used?
JNW
27th October 2018, 11:08
Try DXVA2 (copy-back) instead of D3D11.
Just go software
el Filou
27th October 2018, 11:17
For 23,976 and 25 fps movies played in 60HZ monitor what is the safe rendering time on average and max stats?The safe rendering time in average is your frame time (i.e. 41,7 and 40 ms in this case), but using frame time is not the best idea as sometimes you can have dropped frames if the rendering time is too close to frame time even if it doesn't go above it on average, so you should also just play a movie and watch the OSD for dropped frames.- Is there a chance of a superior madVR HTPC GPU any time soon? Like a GT 2030 or 2050? What about an AMD RX560 would that work as well?If we take the Pascal cards as reference, the 1050s were released 6 months after the 1080, and the 1030 a full year later, so it all depends on what you mean by "any time soon" :)
Edit: I would go with a 1050 Ti (4 GB) rather than an RX560 for madVR. The more affordable ones of each model are about the same price.
mytbyte
27th October 2018, 11:37
Just go software
No, it's the same - measured frame 0 nits, scene 0 nits, movie 1008 nits (same as OSD line above, so probably MaxCLL metadata)
xabregas
27th October 2018, 14:33
The safe rendering time in average is your frame time (i.e. 41,7 and 40 ms in this case), but using frame time is not the best idea as sometimes you can have dropped frames if the rendering time is too close to frame time even if it doesn't go above it on average, so you should also just play a movie and watch the OSD for dropped frames.
Mostly less than that. I was playing with my HD7770 and with ~29ms of both average and max stats i can have smooth playback in almost every file. But i only tested for 5 minutes and stop. Then i started watching the last episode of Daredevil and 5 minutes into it bye bye playback. Rendering times went up like crazy. So, for me, safe rendering times is around 20ms. No dropped frames.
But its at a cost. Instead of NNEDi 32 neurons i will need to stick with 16 neurons. And the difference is huge. Believe me. Im talking about chrom upscaling. Luma upscale is fine whatever i put (since its not double). Its the same with downscale except SSIM 2D.
Warner306
27th October 2018, 15:06
I am trying to understand how keyboard shortcuts for the different settings work. After pressing the shortcut keys do I also have to hit F2 for the setting to apply or does it apply instantly?
Secondly, I am trying to create shortcut keys for switching between the image-doubling options so I can decide on the fly whether I want NGU anti-alias or NGU sharp. Is this possible?
The keyboard shortcuts are recognized instantly but can be saved to the active profile with F2.
"image upscaling algorithm - toggle" and "image upscaling parameter - increase/decrease" may be able to do what you want. You would have to otherwise create separate profiles under "scaling" and only use NGU Sharp and Anti-Alias in those profiles and swap between them by assigning each a keyboard shortcut.
Warner306
27th October 2018, 15:09
I'm looking to upgrade my HTPC to add 2160p HDR support. I have two options: either replace my GT 430 with a newer full-height GPU (maybe passively cooled), or swap out the box for my old TV server box, which is much smaller and would require a half-height GPU. So I have 2 questions:
- Does anyone have any experience with the Gigabyte GTX 1050 Ti OC Low Profile 4G? Is it likely to be silent whilst doing madVR duty?
- Is there a chance of a superior madVR HTPC GPU any time soon? Like a GT 2030 or 2050? What about an AMD RX560 would that work as well?
This build guide has some information on GPUs and a link to a noise test for all GTX 1050 Ti: https://www.avsforum.com/forum/26-home-theater-computers/2364113-guide-building-4k-htpc-madvr.html.
Warner306
27th October 2018, 15:10
If I use passthrough HDR to display (instead of tone map HDR using pixel shaders), madVR is still going to use the measurement file or I have to choose 'tone map hdr using pixel shaders' ?
In this case, your TV would be doing the tone mapping, so the metadata couldn't be used because the TV can't receive it or read it.
Warner306
27th October 2018, 15:12
No, it's the same - measured frame 0 nits, scene 0 nits, movie 1008 nits (same as OSD line above, so probably MaxCLL metadata)
If it says "scene" and "movie" the measurements file is working correctly. This information wouldn't otherwise appear.
mytbyte
27th October 2018, 15:39
If it says "scene" and "movie" the measurements file is working correctly. This information wouldn't otherwise appear.
OK, I figured as much. So the measurement stage is recording 0-es for some reason?
DragonQ
27th October 2018, 15:43
This build guide has some information on GPUs and a link to a noise test for all GTX 1050 Ti: https://www.avsforum.com/forum/26-home-theater-computers/2364113-guide-building-4k-htpc-madvr.html.
I've seen a comparison between the 3 low profile 1050 Ti models before and the Gigabyte was supposedly the quietest, hence it is currently my preference. However, all of the tests are for gaming, including the one linked to in that thread. Has anyone does noise comparisons during video playback? Obviously without madVR I'd expect all the cards to be silent or quiet but with madVR that could change.
jespermart
27th October 2018, 16:02
I have a problem with my new rtx 2080 ti and madVR that I use with kodi and mpc-be player with lav filters 73.1 When I play primarily 4k movies, I get stuttering when using d3d11 native or copyback but much less stuttering with dxva native or
dxva copyback but still stuttering, have any others with the new 2080 card experienced the same? with my old gtx 1080 i didn't have any stuttering at all no matter if i used d3d11 or dxva and my sttings in madVR are the same, surely the new 2080 schould be able to use the same settings in madVR
Warner306
27th October 2018, 16:08
OK, I figured as much. So the measurement stage is recording 0-es for some reason?
If the scene is a black screen, it is 0 nits. Otherwise, it is not accurate. I just read in another forum that the current measurements may not be 100% accurate for each scene, so stay tuned...
grendelrt
27th October 2018, 16:08
Couple questions on setting up custom modes.
1. I ran the optimizer and got back a list of modes. I tried the first 13 suggestions, all worked on my display but were kicked back by the gpu driver for unknown error when I tried to register them. Is this a driver issue (416.34 on 1080ti)? Can I get around this some how?
http://i.imgur.com/7NCguYU.png (https://imgur.com/7NCguYU)
2. Also when running the optimization, does it matter if the video is maximized or not? The display is set to the 23.97 test mode, the video is 23.97 running, but mpc-hc isnt maximized. I noticed when I maxmized mpc-hc the field composition rate changed to 60hz and assumed that was correct.
Warner306
27th October 2018, 16:09
I've seen a comparison between the 3 low profile 1050 Ti models before and the Gigabyte was supposedly the quietest, hence it is currently my preference. However, all of the tests are for gaming, including the one linked to in that thread. Has anyone does noise comparisons during video playback? Obviously without madVR I'd expect all the cards to be silent or quiet but with madVR that could change.
It would be the same thing. The noise tests are taken when the GPU is under load. Checking a bunch of settings in madVR would also put the GPU under load.
Warner306
27th October 2018, 16:13
Couple questions on setting up custom modes.
1. I ran the optimizer and got back a list of modes. I tried the first 13 suggestions, all worked on my display but were kicked back by the gpu driver for unknown error when I tried to register them. Is this a driver issue (416.34 on 1080ti)? Can I get around this some how?
1. Also when running the optimization, does it matter if the video is maximized or not? The display is set to the 23.97 test mode, the video is 23.97 running, but mpc-hc isnt maximized. I noticed when I maxmized mpc-hc the field composition rate changed to 60hz and assumed that was correct.
Failed custom resolutions can relate to specific driver versions, so it is worth trying another driver. Before rolling back, I would clean the previous driver with DDU.
I think the video should be maximized for the most accurate results, but the frames would be drawn at the same frequency, either way.
mytbyte
27th October 2018, 18:12
If the scene is a black screen, it is 0 nits. Otherwise, it is not accurate. I just read in another forum that the current measurements may not be 100% accurate for each scene, so stay tuned...
Sure, staying tuned...it's 0 nits all the time...also I really don't quite grasp the nit measurement line in OSD, sometimes it says "scene xxx nits" sometimes it's "tone map to xxx nits", while "movie xxx nits" is sometimes there, sometimes not...
el Filou
27th October 2018, 18:35
I was playing with my HD7770 and with ~29ms of both average and max stats i can have smooth playback in almost every file. But i only tested for 5 minutes and stop. Then i started watching the last episode of Daredevil and 5 minutes into it bye bye playback. Rendering times went up like crazy.Maybe your GPU is getting too hot and clock speeds are being throttled? I see no other reason for rendering times to be OK for 5 minutes and then go up.Instead of NNEDi 32 neurons i will need to stick with 16 neurons. And the difference is huge. Believe me. Im talking about chrom upscaling.Just a personal opinion, but chroma upscaling doesn't need a high precision scaler. NNEDI? Are you using a really old build?
yok833
27th October 2018, 19:58
When I run the mesurement software I do not have the percentage progession that appears but just a small flashing line :( ??
SirSwede
27th October 2018, 20:16
Hello!
I am looking to buy a Panasonic FZ800 OLED and only have an old DELL Dimension E521 to run Madvr.
I can only afford a new TV and not a new desktop as well. The current graphics card I have in the E521 is GT740. I can, maybe - if I squeeze my purse - get a new graphics card alone. The most I can put in the E521 is a Inno3D GeForce GTX 1050 Ti 1-Slot Edition HDMI DP 4GB.
Most of my material is, still, only 540p and 720p. How well - if at all - would the GTX 1050 (along with a AMD Dual Core CPU Athlon 64X2 6000+ 3.0GHz 89W and 4 GB of RAM) be able to upscale 540p and 720p content to 2160p?
Edit: As far as the RAM goes I am running Windows 7 HP and all I am doing is playing the video with MPC-HC and madVR.
:thanks:
mytbyte
27th October 2018, 20:28
When I run the mesurement software I do not have the percentage progession that appears but just a small flashing line :( ??
Do you have "measure each frame's peak luminance" checked in MadVR's HDR section?
ryrynz
27th October 2018, 21:30
Are you using a really old build?
He said he was and obviously is because as we all know it's been removed for some time.
Also yeah, don't bother with NNEDI3 for chroma.
SirSwede you should be fine with reasonable settings on a 1050.
Warner306
27th October 2018, 23:24
Sure, staying tuned...it's 0 nits all the time...also I really don't quite grasp the nit measurement line in OSD, sometimes it says "scene xxx nits" sometimes it's "tone map to xxx nits", while "movie xxx nits" is sometimes there, sometimes not...
You don't have to check "measure each frame" to create a measurements file. The utility runs on its own.
Based on your description, the measurements file must be incomplete or corrupt. It shouldn't say both "tone map to" and "scene" because they are both meant to represent the same thing.
I would delete the existing measurements file and recreate another fresh one with DXVA2 (copy-back). I would only do this as a test as the measurements files still might not be accurate.
For recent fixes, you also might want to update to the latest test build.
Warner306
27th October 2018, 23:27
Hello!
I am looking to buy a Panasonic FZ800 OLED and only have an old DELL Dimension E521 to run Madvr.
I can only afford a new TV and not a new desktop as well. The current graphics card I have in the E521 is GT740. I can, maybe - if I squeeze my purse - get a new graphics card alone. The most I can put in the E521 is a Inno3D GeForce GTX 1050 Ti 1-Slot Edition HDMI DP 4GB.
Most of my material is, still, only 540p and 720p. How well - if at all - would the GTX 1050 (along with a AMD Dual Core CPU Athlon 64X2 6000+ 3.0GHz 89W and 4 GB of RAM) be able to upscale 540p and 720p content to 2160p?
Edit: As far as the RAM goes I am running Windows 7 HP and all I am doing is playing the video with MPC-HC and madVR.
:thanks:
I would add a little more RAM, but the GTX 1050 Ti can do the job. These are your choices. Don't bother with the cheapest Zotac version: https://us.hardware.info/reviews/7507/18/21-geforce-gtx-1050s-and-gtx-1050-tis-tested-gaming-with-little-ones-noise-levels-and-temperatures
Warner306
27th October 2018, 23:28
When I run the mesurement software I do not have the percentage progession that appears but just a small flashing line :( ??
It is a Command Prompt utility, so it will only flash if you click on it. Follow the instructions here, including installing 32-bit LAV Filters: https://forum.kodi.tv/showthread.php?tid=223175&pid=2750406#pid2750406
The measurements files may not yet be valid until the bugs are worked out, so keep that in mind before creating a bunch.
SamuriHL
27th October 2018, 23:53
If you're going to create measurement files, I HIGHLY recommend going to build 22 or later. Madshi fixed a pretty serious bug and the previous measurement files need to be recreated.
mytbyte
28th October 2018, 01:11
If you're going to create measurement files, I HIGHLY recommend going to build 22 or later. Madshi fixed a pretty serious bug and the previous measurement files need to be recreated.
Yes, with build 23 OSD now shows scene nits change even with the new option "enable dynamic HDR" disabled , but measured frame stays at 0 - this is expected, I suppose?
@Warner: with build 19 it indeed was required to check the "measure peak..." for the measurment tool not to get stuck at the start... not anymore
el Filou
28th October 2018, 01:13
HDR looks terrible with that enabled.. Compromise on anything else..Finally took the time to test that, and it turns out it wasn't the new tone mapping that made my rendering times shoot up (and dumb mode indeed looks terrible ;)), it was the new highlight recovery. With it enabled there is just no way I can play 2160p60 HDR, even bilinear downscaling drops frames! (I already have 'scale chroma separately' enabled).
@madshi Would it be technically possible to implement a 'quality/performance compromise' setting for the highlight recovery? When I check with ShowRenderSteps, that option alone takes rendering time from ~12 ms to ~22 ms (downscaling to 1080). :scared:
SirSwede
28th October 2018, 10:38
I would add a little more RAM, but the GTX 1050 Ti can do the job. These are your choices. Don't bother with the cheapest Zotac version: https://us.hardware.info/reviews/7507/18/21-geforce-gtx-1050s-and-gtx-1050-tis-tested-gaming-with-little-ones-noise-levels-and-temperatures
Thanks!
So the Inno3D GeForce GTX 1050 Ti 1-Slot Edition HDMI DP 4GB, which is the only single slot version available here in my country and that I had looked up, wouldn't work? :confused:
nevcairiel
28th October 2018, 10:41
So the Inno3D GeForce GTX 1050 Ti 1-Slot Edition HDMI DP 4GB, which is the only single slot version available here in my country and that I had looked up, wouldn't work? :confused:
Single slot or half-height cards are usually quite a bit worse then full sized cards, because the tiny coolers on them are really inadequate - so either the cards perform badly, or the fans end up being really loud.
declain
28th October 2018, 11:36
I have a strange problem with madVR. If I play avi xvid SD file on the TV that is connected with HDMI, my primary monitor (displayport) goes to standby. For all other files (e.g. MKV H.264), everything is fine. It's definitely madVR because if I change renderer to MPC-HC default, it's ok.
I have nvidia GTX950 with the latest drivers.
Edit: It was a driver bug. New hotfix drivers solved the problem.
arcspin
28th October 2018, 14:53
Hi,
@madshi
Sometimes I bring up the madVR settings when I'm trying to fine tune the picture.
The settings windows is white with black letters that shine really bright when trying to do settings on a dark scene.
Is there a way to invert that to black with white letters?
That would be so much more easier on the eyes when trying to do tweaks with madVR.
huhn
28th October 2018, 14:55
windows 10 high contrast mode does this.
SirSwede
28th October 2018, 17:37
Single slot or half-height cards are usually quite a bit worse then full sized cards, because the tiny coolers on them are really inadequate - so either the cards perform badly, or the fans end up being really loud.
Thanks!
If it does "perform badly" would it be more of a noise nuisance or would the actual picture (that madVR produces) look worse? :confused:
Does a half-height card require the same amount of power/watts as a full/double-height card or less power/watts?
Warner306
28th October 2018, 17:52
The picture wouldn't look much different. You would may have to use slightly lower values. The noise nuisance when the card is pushed hard may be more of an issue. You can get around this somewhat by not using anything higher than NGU low. The quality should still be very good.
I think all GTX 1050 Tis are fixed to 75 watts max.
yok833
28th October 2018, 18:58
Do you have "measure each frame's peak luminance" checked in MadVR's HDR section?
It was not activated indeed !!! Many thanks :)
SirSwede
28th October 2018, 19:35
The picture wouldn't look much different. You would may have to use slightly lower values. The noise nuisance when the card is pushed hard may be more of an issue. You can get around this somewhat by not using anything higher than NGU low. The quality should still be very good.
I think all GTX 1050 Tis are fixed to 75 watts max.
Thanks!
I am looking at the Zotac GeForce GTX 1050 Ti Mini OC HDMI DP 4GB, instead. :)
Ver Greeneyes
28th October 2018, 19:39
Generally speaking if a GPU goes over its power budget, it'll underclock itself to reduce the power draw (reducing performance). Not much you can do about it (beyond modifying the GPU BIOS which I wouldn't advise), but keep it in mind.
mytbyte
28th October 2018, 20:03
It was not activated indeed !!! Many thanks :)
But if you download build 23 you shouldn't need to anyway.
kitame
28th October 2018, 20:52
has anyone seen this?
https://www.nvidia.com/en-us/geforce/news/new-gfe-features-latest-games-get-ansel-and-highlights/
apparently Nvidia RTX cards offers a new feature that can be used as a straight-up upscaling function?
however, its only been mentioned taking in-game screenshots with 1080P source under super-resolution mode, so i wonder if it could work with videos.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.