View Full Version : madVR - high quality video renderer (GPU assisted)
Manni
11th March 2016, 00:24
Has anyone tried the latest Crimson driver 16.3 on Win10x64, MadVR?
and does MadVR work properly on Win10x64? or is it still Win8.1 that MadVR works perfectly.
I have tested Crimson 16.3 on 8.1 x64 tonight quickly and it seems to be working fine (see my sig for detailed config).
The white crush in 3D seems to be improves, now it does resolve up to 231-232, so it's not too bad, but still not as good as with 14.12.
My only issue is that I get a crash of MPC-BE almost everytime the first time I launch a movie. Second time works fine. It appeared relatively recently but I haven't been able to trace which version of MadVR or MPC-BE introduced this.
It could be a Crimson issue (but not 16.32 specific, I had it earlier), I still have to try to revert to 14.12 to see if it goes away, but as it means I'd lose 3D support with LAV/MadVR, which I really like, I'd rather try to find a solution. There is no error message or crash dump, so debugging isn't easy.
Otherwise all seems fine.
aufkrawall
11th March 2016, 02:39
and that powerstate is the highes possible so it can help here.
Unfortunately, you can't do anything if the driver refuses to cange the p-state due to a bug that makes it stuck.
RyuzakiL
11th March 2016, 02:46
I have tested Crimson 16.3 on 8.1 x64 tonight quickly and it seems to be working fine (see my sig for detailed config).
The white crush in 3D seems to be improves, now it does resolve up to 231-232, so it's not too bad, but still not as good as with 14.12.
My only issue is that I get a crash of MPC-BE almost everytime the first time I launch a movie. Second time works fine. It appeared relatively recently but I haven't been able to trace which version of MadVR or MPC-BE introduced this.
It could be a Crimson issue (but not 16.32 specific, I had it earlier), I still have to try to revert to 14.12 to see if it goes away, but as it means I'd lose 3D support with LAV/MadVR, which I really like, I'd rather try to find a solution. There is no error message or crash dump, so debugging isn't easy.
Otherwise all seems fine.
Yep I also updated my drivers to the latest AMD Crimson 16.3 coming from the last AMD CCC under Win 8.1 and surprised with the performance it brings with my games (Shadow of Mordor).
BUT AMD Crimson enables my Crossfire when it detects MadVR (Wasting power on the 2nd GPU) unlike the old AMD CCC which it continues to use Single GPU only.
So that being said it's another round of compromises but this time, I might stick on using the latest AMD Crimson since I get higher frame rates on my games.
I wonder how it handles under Win.10, I heard scary stories on that Spying OS and the Last time I used it with MadVR It doesn't like Fullscreen Exclusive Mode and leads to dropped frames and CTD's
RyuzakiL
11th March 2016, 04:30
So I stumble about these 2 settings in AMD Crimson and I read about disabling ITC processing and enabling GPU Scaling bypasses the display processor entirely so ergo eliminating redundant conversions by the HDTV's processor since it was bypassed.
and also by doing this you're letting the GPU itself to do all the video processing and eliminating the HDTV to do redundant conversions, also it gives the fastest refresh and best fps in games when bypassing the display processor.
leeperry
11th March 2016, 06:48
disabling ITC processing and enabling GPU Scaling bypasses the display processor entirely so ergo eliminating redundant conversions by the HDTV's processor
never seen any diff with mVR using 13.12 in 8bit
Uoppi
11th March 2016, 08:48
and nvidia deint was kind of "broken" the last time i check and had issue with telecine/hybrid content.
Do you mean cadence detection? I see occasional cadence breaks with animated 3:2 content (Ren & Stimpy) but with movie content (Cronenberg's Crash), IVTC seems to work just fine, i.e. looks like cadence is consistently detected.
DragonQ
11th March 2016, 09:02
amd crimson deint is simply broken it is using a low quality bob. i haven't tested the newest beta yet.
and nvidia deint was kind of "broken" the last time i check and had issue with telecine/hybrid content. the high quality CUVID deint is not usable at all under windows 10.
Yeah I had to switch back to Catalyst because deinterlacing is completely buggered with Crimson. I couldn't even find a link to Catalyst on AMD's website so had to download an old beta version then run the updater to bring it up to date. I'm amazed a bug with such a simple fix (i.e. change the default deinterlacing method to vector adaptive) hasn't been fixed yet.
huhn
11th March 2016, 09:56
never seen any diff with mVR using 13.12 in 8bit
older driver than 14.something are always outputting 10 bit if the device supports that (and i haven't seen a TV that doesn't support 12 bit input) regardless of the input bit deep. which is hopefully just adding some 0 zeros and nothing else but you got an extra conversation step with older driver.
huhn
11th March 2016, 10:00
Do you mean cadence detection? I see occasional cadence breaks with animated 3:2 content (Ren & Stimpy) but with movie content (Cronenberg's Crash), IVTC seems to work just fine, i.e. looks like cadence is consistently detected.
it was really bad when lavfilter added w3fdif in the nightlies i haven't check it again. could be ok now.
James Freeman
11th March 2016, 17:53
madshi, I don't know what algorithm you used to remap the HDR content to SDR but according to the ST.2084 standard everything under 100nit should be exactly the same as in the SDR.
Do you think it is the right time to discuss light output, gamma curves remapping, color space remapping, etc?
nevcairiel
11th March 2016, 17:55
madshi, I don't know what algorithm you used to remap the HDR content to SDR but according to the ST.2084 standard everything under 100nit should be exactly the same as in the SDR.
I'm sure he has read the appropriate specifications.
leeperry
11th March 2016, 18:01
older driver than 14.something are always outputting 10 bit if the device supports that (and i haven't seen a TV that doesn't support 12 bit input) regardless of the input bit deep. which is hopefully just adding some 0 zeros and nothing else but you got an extra conversation step with older driver.
Fair enough but did anyone run benchmarks with mVR between 13.12 and newer ones? Unkeen on wasting hours to lose 10% performance.
James Freeman
11th March 2016, 18:05
I'm sure he has read the appropriate specifications.
Well, I am sure too, but It does not look like the content below 100nit keeps the same light output level, it changes with the "HDR Peak Luminance" setting in madVR.
I don't doubt madshi's decisions about the remapping algorithm for a second, that is why I asked if it is the right time to talk about the whole HDR to SDR remapping issue so that madshi will be open to suggestions.
huhn
11th March 2016, 18:42
Fair enough but did anyone run benchmarks with mVR between 13.12 and newer ones? Unkeen on wasting hours to lose 10% performance.
i don't know about the performance difference and i don't need to max out my GPU when i'm using madVR but i need the new driver.
nevcairiel
11th March 2016, 19:00
Well, I am sure too, but It does not look like the content below 100nit keeps the same light output level, it changes with the "HDR Peak Luminance" setting in madVR.
I don't doubt madshi's decisions about the remapping algorithm for a second, that is why I asked if it is the right time to talk about the whole HDR to SDR remapping issue so that madshi will be open to suggestions.
None of these HDR standards really define a way to convert them to SDR, so you can't judge any behavior thats supposed to happen on a HDR display when converted to SDR.
Just judge if it looks good - which it certainly does.
PS:
I don't see anything concerning "content below 100nits" in SMPTE ST 2084 in any case. Its literally just an equation to convert non-linear (gamma) to linear.
RenderGuy2
11th March 2016, 21:49
I'm really excited about the recent 3D/MVC developments! Huge thank you to Madshi and Nevcairiel. The only problem I seem to have is that subtitles are not rendered with depth. Source is 3D Blu-ray. Playing the disc with TMT6 results in the subtitles being displayed with varying amounts of (usually negative) parallax. With MadVR subtitles are always at the screen plane. I'm currently using MadVR 0.90.15 and Lav 0.68.0-2.
madshi
11th March 2016, 22:00
Just a quick reply to this. I'll reply to the other posts in detail in the next 2 days.
I'm really excited about the recent 3D/MVC developments! Huge thank you to Madshi and Nevcairiel. The only problem I seem to have is that subtitles are not rendered with depth. Source is 3D Blu-ray. Playing the disc with TMT6 results in the subtitles being displayed with varying amounts of (usually negative) parallax. With MadVR subtitles are always at the screen plane. I'm currently using MadVR 0.90.15 and Lav 0.68.0-2.
Subtitle 3D depth should work just fine with XySubFilter. Or if you're using the MPC-HC/BE internal subtitle renderer, you need to update to the latest MPC-HC/BE nightly build.
leeperry
11th March 2016, 22:57
i don't know about the performance difference and i don't need to max out my GPU when i'm using madVR but i need the new driver.
higher perfs = higher efficiency = lower power consumption = cheaper energy bill :)
aufkrawall
12th March 2016, 00:59
I compared madVR's current super-xbr implementation with Shiandow's latest for MPDN.
Images with 200% super-xbr scaling:
madVR:
http://abload.de/thumb/kodim05madvrf7sfc.png (http://abload.de/image.php?img=kodim05madvrf7sfc.png) http://abload.de/thumb/kodim06madvr2ystg.png (http://abload.de/image.php?img=kodim06madvr2ystg.png) http://abload.de/thumb/bwboatmadvrlns29.png (http://abload.de/image.php?img=bwboatmadvrlns29.png)
Shiandow's:
http://abload.de/thumb/kodim05mpdn3rsdy.png (http://abload.de/image.php?img=kodim05mpdn3rsdy.png) http://abload.de/thumb/kodim06mpdnb1swl.png (http://abload.de/image.php?img=kodim06mpdnb1swl.png) http://abload.de/thumb/bwboatmpdnzxsli.png (http://abload.de/image.php?img=bwboatmpdnzxsli.png)
When watched very closely or with some magnification, madVR shows less aliasing.
But I guess this is due to the tweaked parameters by Hilian, and not because of his improved anti-ringing filter. I think there's not such a case where the anti-ringing filter would remove actual details in a crucial way.
I'd really like to see how the improved deband filter behaves without the other tweaked parameters. Then we'd know for sure.
Hyllian
12th March 2016, 01:16
Yes, the new params slightly add aliasing. Madshi already had spot that when I showed to him and that's the reason he didn't incorporate in madVR.
Other than that, I think the new params improves in fine details, like pixel art, subtitles, etc.
If Shiandow prefer to get rid of all aliasing, he can reverse the changes or add then as an option.
RenderGuy2
12th March 2016, 01:16
Subtitle 3D depth should work just fine with XySubFilter. Or if you're using the MPC-HC/BE internal subtitle renderer, you need to update to the latest MPC-HC/BE nightly build.
Thanks, using XySubFilter worked. For some reason MPC-HC's (1.7.10.101 x64) internal subtitle renderer doesn't render with depth on my system.
aufkrawall
12th March 2016, 01:50
If Shiandow prefer to get rid of all aliasing, he can reverse the changes or add then as an option.
But did I get it right that the changes for pixelart and ringing are two separate things? :)
djfred93
12th March 2016, 01:56
amd crimson deint is simply broken it is using a low quality bob. i haven't tested the newest beta yet.
and nvidia deint was kind of "broken" the last time i check and had issue with telecine/hybrid content. the high quality CUVID deint is not usable at all under windows 10.
Can't you just activate it through registery or DXVA Checker (http://bluesky23.yukishigure.com/en/dxvac/howto/vas2.html) ?
ryrynz
12th March 2016, 02:08
Yes, the new params slightly add aliasing.
Is having an anti-aliasing filter in future a good idea?
Hyllian
12th March 2016, 02:09
But did I get it right that the changes for pixelart and ringing are two separate things? :)
The ringing didn't change with new params. The anti-ringing is independent of these params.
If you're comparing anti-ringing codes, then nervermind my comments.
janos666
12th March 2016, 02:18
I had to increase my CPU and GPU queue depths from 4 (my old setting, the minimum allowed) to 8 and activate the "alternative glitch handling" in order to get smooth 3D playback (otherwise, it stuttered like hell with my i5-2500K + 290X + Win10 system).
But now, it works unsurprisingly well (MPC-HC + madVR changed to 3D a lot faster and smoother than PowerDVD [which ends up in 3D with a 50-50 chance after vibrating the screen for several seconds]).
Thank you for this! (Even if 3D is going out of the TV business, I didn't care to watch any 3D content since late 2012, so I have some movies to attend to for this special occasion :) ).
Separate 3DLUTs would be welcomed though (2D/3D) because my plasma TV behaves extremely differently in 3D mode (but considering how scarce the 3D materials are, it's not that painful to change this manually).
Hyllian
12th March 2016, 02:19
Is having an anti-aliasing filter in future a good idea?
I don't think so.
If you use the old params, you get rid of all aliasing. No need of specific code to get rid of it.
What could upgrade sxbr would be a way to squeeze edge gradients. I've read some papers about it: Source 1 (http://research.microsoft.com/en-us/um/people/jiansun/papers/GradientSR_CVPR08.pdf), Source 2 (http://gr.xjtu.edu.cn/LiferayFCKeditor/UserFiles/File/GradientProfile.pdf), Source 3 (http://www.apsipa.org/proceedings_2012/papers/361.pdf), Source 4 (http://download.springer.com/static/pdf/778/art%253A10.1007%252Fs41095-016-0036-6.pdf?originUrl=http%3A%2F%2Flink.springer.com%2Farticle%2F10.1007%2Fs41095-016-0036-6&token2=exp=1457746508~acl=%2Fstatic%2Fpdf%2F778%2Fart%25253A10.1007%25252Fs41095-016-0036-6.pdf%3ForiginUrl%3Dhttp%253A%252F%252Flink.springer.com%252Farticle%252F10.1007%252Fs41095-016-0036-6*~hmac=943ae0eb229862353f1ca17fd72bd00c59f37c0f81114b418a5143825dae542c), but don't know how to implement a fast way to do it.
ryrynz
12th March 2016, 05:59
What could upgrade sxbr would be a way to squeeze edge gradients. I've read some papers about it, but don't know how to implement a fast way to do it.
Someone else on the forum might be of some help, could be worth creating a new thread.
madshi, NNEDI3 quadrupling is broken with 720 res content.
http://i.imgur.com/72PGUbSl.jpg
Regarding max line thinning, since all upscaling refinement options max at 4 I assume you'll just change the degree of strength per point rather than changing the max value like I suggested.
I think you could increase it per point by 25-50%, users would need to update their settings.. but it keeps it all in line then.
Neo-XP
12th March 2016, 12:17
Hi everyone,
When using MadVR with XySubFilter to display subtitles outside of the video area with the option "move subtitles to bottom of the screen/window" of the "zoom control" menu, when an italic subtitle appears (<i> tag), the italic subtitle and all the following subtitles are displayed into active video area.
Can this bug be fixed by madVR ?
Example with a normal subtitle :
http://img15.hostingpics.net/thumbs/mini_5806351positionok.jpg (http://www.hostingpics.net/viewer.php?id=5806351positionok.jpg)
Here comes the italic subtitle :
http://img15.hostingpics.net/thumbs/mini_6625652italicsubtitle.jpg (http://www.hostingpics.net/viewer.php?id=6625652italicsubtitle.jpg)
All the following subtitles are placed like this :
http://img15.hostingpics.net/thumbs/mini_5690543followingsubtitle.jpg (http://www.hostingpics.net/viewer.php?id=5690543followingsubtitle.jpg)
huhn
12th March 2016, 12:31
madshi, NNEDI3 quadrupling is broken with 720 res content.
http://i.imgur.com/72PGUbSl.jpg
what should be broken on that screen. there is no word of super sampling in quadruple.
ryrynz
12th March 2016, 13:12
There is no word of super sampling in quadruple.
Right.. Didn't expect that limitation.
Guess I'm waiting for support for quadrupling super sampling then.
nevcairiel
12th March 2016, 13:16
Doubling 720 to 1080p already super samples, I doubt there is much of an improvement to be seen further, not to mention that doubling 1440p once more would be extremely expensive.
ryrynz
12th March 2016, 13:21
Yeah higher resolutions would be very expensive no doubt.
Was displaying 960 x 720 content and quadrupled that's not terribly higher than full HD doubled. 3840 x 2880 vs 3840 x 2160.
fedpul
12th March 2016, 13:23
Quoting myself, I have done more tests.
Hi madshi, I kept getting black screens with 90.15, with Chroma NNEDI enabled, Doublind disabled, Chroma Jinc and Doubling NNEDI, and obviously with both using NNEDI. I need to check without using NNEDI at all. I always use NNEDI 32 Neurons for the tests. Black screens are really random, first one occurs after more than 50 minutes of playback, so testing requires lots of time.
Hi again madshi, I have been doing more tests. Without using NNEDI at all I get black screens too. I also tried to reset all to default settings and configure everything again and it happens anyway. Also I tried not using 10 bits output (using 8 instead) but it also happens. I am a bit clueless now on what it could be. Also last night when I was testing 8 bit, FSE flashed once showing windowed mode when I was watching a movie without touching anything.
I would like to know what alternative glitch hadling does, I have it unckecked.
If you have any questions please feel free to ask me.
huhn
12th March 2016, 13:25
that's a lot more pixel than UHD.
and what do you even want to do at that resolution with a source that is at best worth 720p?
aufkrawall
12th March 2016, 13:57
The ringing didn't change with new params. The anti-ringing is independent of these params.
If you're comparing anti-ringing codes, then nervermind my comments.
I wanted to check if the new anti-ringing has any negative effect in more test images, since it seems to be working extremely fine in one cartoon example I got.
Hyllian
12th March 2016, 14:29
When watched very closely or with some magnification, madVR shows less aliasing.
But I guess this is due to the tweaked parameters by Hilian, and not because of his improved anti-ringing filter. I think there's not such a case where the anti-ringing filter would remove actual details in a crucial way.
You're right, they're unrelated.
James Freeman
12th March 2016, 15:57
madshi, continuing the discussion from AVS, may I suggest some idea about HDR to SDR conversion;
If reference white is still 100 nit then everything below it should look the same for HDR and SDR.
If you will, you can try to tweak your algorithm so that 90 nit and below remains the same (unchanges) and the last 10 nit will compress all the data from the HDR video.
The ratio can be changed for example 70 and 30, but the point is that the lower part should remain unchanged, or mostly unchanged.
Your algo as you mentioned also fixes the colors as it compresses HDR to SDR but I don't know how it will work out with the way I suggested, if at all, thugh it may not matter that much because only the last 10 or 30 nit will have to remap.
From what I see now with the current algorithm even the darkest parts like night sky form Life of Pi HDR demo changes.
The night sky in this demo only look black as they should when I select 1,300 nits for my display which matches the mastering display from the metadata.
KoD
12th March 2016, 16:27
Hi everyone,
When using MadVR with XySubFilter to display subtitles outside of the video area with the option "move subtitles to bottom of the screen/window" of the "zoom control" menu, when an italic subtitle appears (<i> tag), the italic subtitle and all the following subtitles are displayed into active video area.
Can this bug be fixed by madVR ?
Example with a normal subtitle :
http://img15.hostingpics.net/thumbs/mini_5806351positionok.jpg (http://www.hostingpics.net/viewer.php?id=5806351positionok.jpg)
Here comes the italic subtitle :
http://img15.hostingpics.net/thumbs/mini_6625652italicsubtitle.jpg (http://www.hostingpics.net/viewer.php?id=6625652italicsubtitle.jpg)
All the following subtitles are placed like this :
http://img15.hostingpics.net/thumbs/mini_5690543followingsubtitle.jpg (http://www.hostingpics.net/viewer.php?id=5690543followingsubtitle.jpg)
If I remember this right, madshi was looking at some point for someone who has this issue, because he could not reproduce it.
If you could post the subtitle file, and the settings.bin from the madVR folder to some web upload service (I use wikisend as it's fast and free of dangerous and intrusive ads, but anything else can be used instead), and also take screenshots with the madVR OSD active (Ctrl + J) so that your scaling settings are visible, that will likely help.
AngelGraves13
12th March 2016, 18:14
Jinc downscaling looks great, but dear lord is it slow. Had to go back to Catmull Rom.
Hopefully it can be optimized a bit in the future.
James Freeman
12th March 2016, 20:21
Okay I found a very nice setting for the HDR to SDR conversion, all I had to do is to lower some the Brightness (-40) setting in MPC-HC (or madVR) with 400 nit setting.
Sorry to cause some turbulance with my requests. :o
baii
12th March 2016, 20:48
I had used the same Amd GPU from 13.x to 15.x and I would say there is not much performance difference. The only difference is that some may crash madvr or f up deinterlacing or use some stupid power management. The crimson crash 32bit pot player on fury.
Sent from my 306SH
ryrynz
12th March 2016, 21:40
that's a lot more pixel than UHD.
33% more.
and what do you even want to do at that resolution with a source that is at best worth 720p?
Just see what sort of result I get, won't make much difference but whatever. I'll check it out in Avisynth.
Neo-XP
12th March 2016, 23:44
If I remember this right, madshi was looking at some point for someone who has this issue, because he could not reproduce it.
If you could post the subtitle file, and the settings.bin from the madVR folder to some web upload service (I use wikisend as it's fast and free of dangerous and intrusive ads, but anything else can be used instead), and also take screenshots with the madVR OSD active (Ctrl + J) so that your scaling settings are visible, that will likely help.
The settings : settings.reg (http://wikisend.com/download/195188/settings.reg)
The subtitles : subtitles.srt (http://wikisend.com/download/477534/subtitles.srt)
Subtitle n°60 :
http://img4.hostingpics.net/thumbs/mini_839990subtitle60.jpg (http://www.hostingpics.net/viewer.php?id=839990subtitle60.jpg)
Subtitle n°61 :
http://img4.hostingpics.net/thumbs/mini_291372subtitle61.jpg (http://www.hostingpics.net/viewer.php?id=291372subtitle61.jpg)
I'm using MPC-HC internal subtitles renderer for now because it doesn't have this bug, but I hope it can be fixed, because the subtitles look better with XySubFilter.
kiwijunglist
13th March 2016, 05:32
Hi, I'm currently using Win7 (x64) with 32bit MPC-HC, LAV, and MadVR
I want to use mpc-hc for 3d video instead of powerdvd, this is because it's annoying to have 2 different players.
Q: In Jan madshi posted that Win8.1 was preferential to Win10. Is this still the case?
Q: Should I install Win8.1 or Win10, and should I go with 32 or 64 bit? Or am I better sticking with my current setup and PowerDVD for 3d playback for now, as there are still issues to be sorted?
My current specs are:
Intel i3-2100 / 12GB RAM / HD7770 GPU / Samsung HU6400 (1080P 3D) / Use 60hz PC mode for all non 3d video.
Thanks
Kiwi
x7007
13th March 2016, 06:21
I'm using 100 hz laptop monitor and Potplayer + madvr . how come I had Display 100.00000Hz steady. it didn't move at all. I don't know what affected it or what did it, but now it always moves between 100.00392-410. anyone has idea what could affect it and how to make it stuck on 100.00000Hz steady ?
huhn
13th March 2016, 08:29
steady 100.00000 sounds more like a bug for me it looks like the refresh rate was incorrectly read in the past.
James Freeman
13th March 2016, 11:55
Fellows, I want to start a movie with these parameters:
levels=tripleExp
blacklevel=+26
whitelevel=+50
gamma=2.6
then switch back to default and compare between the two, can it be done?
Putting a text file named: "levels=tripleExp, blacklevel=+26, whitelevel=+50, gamma=2.6.txt" doesn't work
Changing the folder name to this does but gamma value does not change (no gamma tag?).
I need a quick one button switch for this test.
EDIT:
Okay, Profiles with a key shortcut works to change the Gamma, but what about the other commands?
"Command line to execute when this profile is activated" should do the trick along with a "keyboard shortcut to this profile" the but what are the command line?
filename tags don't work as "command lines".
6233638
13th March 2016, 12:12
Fellows, I want to start a movie with these parameters:
levels=tripleExp
blacklevel=+26
whitelevel=+50
gamma=2.6
then switch back to default and compare between the two, can it be done?
Putting a text file named: "levels=tripleExp, blacklevel=+26, whitelevel=+50, gamma=2.6.txt" doesn't work
Changing the folder name to this does but gamma value does not change (no gamma tag?).
I need a quick one button switch for this test.There's no way to quickly toggle between the adjustments and the original. That would be useful though.
I requested gamma tags a long time ago, but madshi seemed opposed to the idea for some reason.
That said, even in its current state, the tags are so useful. Though it's not too common, I have a handful of Blu-rays which would have a completely washed-out picture if it were not for these controls.
James Freeman
13th March 2016, 12:21
Yes my profile changes the gamma in the "color & gamma" section but I want to add the rest in the "Command line to execute when this profile is activated".
But it doesn't seem to work.
Also when I try to rename a Profile, madVR setting window skips around so I can't type anything.
Maybe it's a bug? or maybe not.
madshi, executing tags from a profile will be very useful indeed.
Changing many setting with a click of a button, very useful for comparison.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.