View Full Version : madVR - high quality video renderer (GPU assisted)
nevcairiel
22nd November 2018, 09:42
you can already use a different GPU to render the on your screen. it's a default feature of windows 10.
you go to settings-> display-> graphics settings -> browse ->add your player
now you go to options and select high performance which should be your AMD card.
This works with switchable graphics in laptops, but not with two dedicated graphics cards, not that I know of anyway.
huhn
22nd November 2018, 09:44
sorry i don't trust anything any tech company writes on there products.
you have to test this.
as an example pure direct modes on Panasonic isn't pure at all and clearly not direct what so ever.
huhn
22nd November 2018, 09:48
This works with switchable graphics in laptops, but not with two dedicated graphics cards, not that I know of anyway.
i just did it on a desktop PC the HDMI connector isn't even shared.
not sure if my board has two pcie "16x" slots and i'm currently not interested in creating one with a knife to test this.
JarrettH
23rd November 2018, 06:10
NGU AA low vs Medium (http://screenshotcomparison.com/comparison/123459) Bear in mind this is enlarged 300%.
The jump from NGU AA low to medium is greater than medium to high, personally I prefer medium as my base quality setting but it's relatively minor at the end of the day. I say change it to low and be happy with the lower noise.
Thanks, by the way.
Do you mind showing me the same thing as super-xbr 100 vs ngu aa low? I'm curious if the low setting is any better.
ryrynz
23rd November 2018, 07:59
NGU AA low vs Super-xbr 100 AR (http://screenshotcomparison.com/comparison/124826)
Super-xbr 100 is a nice light weight option and my previous favorite but at 100 I feel is a tad sharp which you may notice as I did, at 75 however it should look fairly similar to NGU AA low.
voyager6868
23rd November 2018, 09:17
From what I can tell madVR doesn't do 3:2 pulldown properly (see more technical details below). That is, if I play a 24fps video on a 60Hz display, it doesn't seem to show a regular cadence of AAABBCCCDDEEEFF etc.
A few points I'd like to make in favor of fixing this:
Many TVs can automatically do reverse pulldown if the cadence is proper and can show the video without judder. With an unpredictable cadence like madVR appears to provide, this isn't possible.
So, why not just switch the video mode to 24Hz? One, it takes a few seconds for some TVs to switch (blank screen while it does so) between modes, and two (this may be uncommon) some TVs have different input lag at 24Hz versus 60Hz, so the lipsync setting on a receiver needs to be regularly adjusted back and forth (mine has a difference of 110ms).
Further, some users may have a screen that does not support 24Hz and are used to this type of judder (e.g., North American users), so don't want to devote the processing power to enabling smooth motion (or are more bothered by the smooth motion artifacts that this judder). The current implementation is more jarring than typical judder as it appears to be more unpredictable.
My methodology:
1. Download the test here:
https://www.rtings.com/images/test-materials/2015/305_24p.mp4
2. Ensure your screen is set up to reverse the 3:2 pulldown. On many Samsung TVs, for example, this can be done by turning on Auto Motion Plus, Custom, setting both sliders to 0. If your screen can't do this, you can still run the test, it will just have a different ideal result.
3. Play the video in MPCHC with madVR output.
4. While playing the video, take a 1-second exposure of the screen (this can be done, for example, using the Camera app in Pro mode on many recent Samsung phones)
Ideal result: If your screen does the pulldown, you should see 24 equally bright squares. If your screen does not, you should see a regular checkerboard pattern.
Result with current madVR: Squares appear to have random brightness with no discernible pattern.
Result with YouTube in Chrome browser (search rtings judder test), YouTube app on AppleTV, Infuse 5 on AppleTV, Windows 10 stock media player is the ideal result as above.
So, madVR seems to be an outlier with regards to this support.
If I use Handbrake to set the video's FPS to 60, everything looks correct in madVR, so seems clear it is the 3:2 pulldown algorithm that is at fault.
ryrynz
23rd November 2018, 11:23
Hmmmmm.. Yeah, known issue.
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...
Been a few years madshi, maybe you can jump on this?
e-t172
23rd November 2018, 11:41
From what I can tell madVR doesn't do 3:2 pulldown properly (see more technical details below). That is, if I play a 24fps video on a 60Hz display, it doesn't seem to show a regular cadence of AAABBCCCDDEEEFF etc. [...] Many TVs can automatically do reverse pulldown if the cadence is proper and can show the video without judder. With an unpredictable cadence like madVR appears to provide, this isn't possible.
I can confirm, as I reported this a few months ago (http://forum.doom9.org/showthread.php?p=1838742#post1838742). Here's a direct quote from madshi:
Yes, that is still the case. It's been on my to do list for ages to tune madVR to produce a repeatable 3:2 pattern when smooth motion is turned off, but I still haven't found the time to do that yet.
j82k
23rd November 2018, 12:41
Interesting, that would explain why I'm getting weird hitching when playing 24p content over 60Hz even though my TV is supposed to be able to remove 3:2 pulldown judder.
Would be awesome if this could be fixed.
stiffwin
25th November 2018, 11:11
Yes I would love to see this implemented too - no pressure, Madshi :D
Bulldogger
25th November 2018, 19:12
As far as I heard they are both equal. only nvidia can't do 10bit output and the vega needs more power. What does "vega needs more power," exactly mean? Bigger power supply or no limitation that is inherit to the card that can't be solved?
iSeries
25th November 2018, 21:28
The thing with Oleds is, they are terrible at near-blacks. Not only the uniformity but also the near-black gradation so I totally don't mind some dithering noise which hides this ugliness a little bit and also makes the gradation a little smoother. I wish Oleds would just dither the near-blacks by default like plasmas did. This would solve alot of problems and only cause a bit of noise.
edit: madshi should implement some algorithm to madVR for Oled owners that does heavy dithering to near-blacks and less and less on brighter parts of the picture. :D
Madshi, if this is something you could implement, there's a bunch of oled owners who'd be forever grateful! I know you don't have an oled but we could test for you. On my C8, I have to have madVR dither to 5bit to get acceptable near-black gradation.
tony359
25th November 2018, 21:50
Hi there,
I'm using DisplayCAL and MadVR's test pattern generator to profile my display. I've just updated everything to the latest version, including video driver (GTX 1050Ti). MadVR's patterns are constantly flashing on screen, there is a double flash every three seconds. I have never seen this before. Is this intentional/known/am I doing something wrong?
Thanks!
ryrynz
25th November 2018, 22:56
What does "vega needs more power," exactly mean?
Uses more power.
seiyafan
26th November 2018, 03:42
How's performance on Vega 64 compared to GTX 1080?
huhn
26th November 2018, 04:13
there are not a lot of vega cards out there is considered an all over pretty bad card for consumer use.
even the 56 can spike to ~600 watt for an relative long time. and it is unclear if AMD makes any money with them.
so not that easy to find a user with it. it could be much faster for madVR use than a 1080.
JarrettH
26th November 2018, 06:03
NGU AA low vs Super-xbr 100 AR (http://screenshotcomparison.com/comparison/124826)
Super-xbr 100 is a nice light weight option and my previous favorite but at 100 I feel is a tad sharp which you may notice as I did, at 75 however it should look fairly similar to NGU AA low.
Much bigger difference. I used xbr-100 for the longest time too. You can see here it's a bit too juiced up and in certain areas of cloth it feels like the colour is clipping. Thanks!
SirSwede
26th November 2018, 09:49
Does madVR support using DCIP3 as colour gamut instead of Rec709?
tony359
26th November 2018, 11:32
I'd like to update my last post (https://forum.doom9.org/showthread.php?p=1858373#post1858373)
I found that the flashing only happens when FSE is enabled. I've now disabled it and all is good. Which takes me to the following question: is FSE still a good idea these days? I believe I read comments suggesting it's not really needed anymore.
Thanks
huhn
26th November 2018, 11:49
Does madVR support using DCIP3 as colour gamut instead of Rec709?
madVR supports all kinds of gamuts DCI P3 is one of them.
I'd like to update my last post (https://forum.doom9.org/showthread.php?p=1858373#post1858373)
I found that the flashing only happens when FSE is enabled. I've now disabled it and all is good. Which takes me to the following question: is FSE still a good idea these days? I believe I read comments suggesting it's not really needed anymore.
Thanks
it depends on your system but you are usually fine without it.
OnkelB
26th November 2018, 13:42
Hey guys!
I have got an issue with the “automatically detect hard coded black bars” option. It won’t work when using D3D11 automatic (Native) Mode in LAV Filters. I know that it doesn’t work with DXVA2 native mode, but what about D3D11?
The option works when I use copy back mode, but rendering times are getting insane then. Is there any fix for that?
sneaker_ger
26th November 2018, 13:55
No fix. Black bar detection is simply not supported for D3D11 native and DXVA2 native.
OnkelB
26th November 2018, 14:15
That’s what I thought, but I couldn’t find that info anywhere. Thank you for verifying!
tony359
26th November 2018, 16:20
it depends on your system but you are usually fine without it.
Thanks.
How do I find out if I am within the "usual" systems?
cork_OS
26th November 2018, 20:36
Thanks.
How do I find out if I am within the "usual" systems?
If you have no problems with HDR, high bit-depth and custom refresh rates, you don't need FSE.
mclingo
26th November 2018, 21:17
Madshi, if this is something you could implement, there's a bunch of oled owners who'd be forever grateful! I know you don't have an oled but we could test for you. On my C8, I have to have madVR dither to 5bit to get acceptable near-black gradation.
we've discussed this before, it isnt that OLEDS are particularly bad at near black, I keep hearing this but I dont see it. I have a first gen OLED and I can tell you that it you feed it a good image you get one right back if its setup right.
Where they struggle is poorly mastered compressed material where there is macro blocking at near black. An LCD will have exactly the same problems if they could do perfect black, LCD just masks these areas as its all just a grey mush. My OLED cant fix this, Later panels added a black crush and some extra processing to hide some of this but mine has none of this so I use MADVR, I guess you could say i'm seeing the picture as it is, warts and all.
I do agree however some extra dithering patterns near black would be very useful, especially for TV rips. Some of the netflix, amazon and particularly sky stuff has been so bad over the years even though I was paying for it I sometimes had to download the episode so you I can run it through MADVR and apply some image processing.
BBC are the worst offenders IMHO, there compression is horrific, Peaky blinders has been completely unwatchable at times due to streaming compression artifacts.
iSeries
26th November 2018, 21:34
we've discussed this before, it isnt that OLEDS are particularly bad at near black, I keep hearing this but I dont see it. I have a first gen OLED and I can tell you that it you feed it a good image you get one right back if its setup right.
Where they struggle is poorly mastered compressed material where there is macro blocking at near black. An LCD will have exactly the same problems if they could do perfect black, LCD just masks these areas as its all just a grey mush. My OLED cant fix this, Later panels added a black crush and some extra processing to hide some of this but mine has none of this so I use MADVR, I guess you could say i'm seeing the picture as it is, warts and all.
I do agree however some extra dithering patterns near black would be very useful, especially for TV rips. Some of the netflix, amazon and particularly sky stuff has been so bad over the years even though I was paying for it I sometimes had to download the episode so you I can run it through MADVR and apply some image processing.
BBC are the worst offenders IMHO, there compression is horrific, Peaky blinders has been completely unwatchable at times due to streaming compression artifacts.
Trust me, it's different with 2018 OLEDs. For some reason they are pumping brightness through near black and causing near black artifacts to flash brightly. Pause the video and the flashing stops, press play again and the crazy flashing starts again. This does seem to be a bug because sending the same content as bt.2020 instead of bt.709 and this flashing does not occur. That isn't a viable solution but luckily dithering to 5bit also solves the issue (no flashing). This also was not an issue with the two C7s I had. Or any other TV i've had lol.
https://www.avsforum.com/forum/40-oled-technology-flat-panels-general/2962154-2018-lg-c8-e8-owners-thread-no-price-talk-276.html#post57140188
tony359
26th November 2018, 22:30
If you have no problems with HDR, high bit-depth and custom refresh rates, you don't need FSE.
No HDR involved but never managed to get custom refresh rates to work - it's on my to do list to investigate better.
What do you mean with "high bit depth" - I mean, I know what that is but how does it apply here?
Thanks
el Filou
26th November 2018, 23:59
luckily dithering to 5bit also solves the issueIf that is the solution, the panel manufacturer really ought to have a look at the problem tbh. :scared:
iSeries
27th November 2018, 00:18
If that is the solution, the panel manufacturer really ought to have a look at the problem tbh. :scared:
'Solves' is the wrong word. Effectively reduces the effect would be a better way of saying it. And sending bt.709 content as bt.2020 also effectively reduces the effect...I mean, how does that make sense, other than a processing bug in the TV? But yeah, not great for a so-called 10bit panel. Some dark content is downright ugly on this. Thank the lord madshi for madVR though!
ryrynz
27th November 2018, 00:48
What about changing the range a bit at the black end? Does that affect it much?
iSeries
27th November 2018, 01:10
What about changing the range a bit at the black end? Does that affect it much?
Custom levels? A tiny bit...but to truly mitigate it as effectively as 5bit dithering I have to raise it too much. Also raising brightness on the TV to around 56 from 50 mitigates it, but 1 click over 51 and black level is raised.
A dithering method that would dither at 5bit or so at the black end and ramp up to output bit depth would be perfect!
j82k
27th November 2018, 09:43
I made a video about this, otherwise it's hard to imagine what this can look like if you don't have an 2018 LG Oled. There is a ton of noise in the video because of the high ISO recording though.
https://www.youtube.com/watch?v=Sl_t3UQ3gAk
And yes dithering to 5-bits completely gets rid of those flashing artifacts you see in the video. Outputting BT.2020 does too in some cases.
What's funny is that most people completely negate this problem and refuse to even run the test video I made. Though the few that did got very similar results.
iSeries
27th November 2018, 11:29
I made a video about this, otherwise it's hard to imagine what this can look like if you don't have an 2018 LG Oled. There is a ton of noise in the video because of the high ISO recording though.
https://www.youtube.com/watch?v=Sl_t3UQ3gAk
And yes dithering to 5-bits completely gets rid of those flashing artifacts you see in the video. Outputting BT.2020 does too in some cases.
What's funny is that most people completely negate this problem and refuse to even run the test video I made. Though the few that did got very similar results.
Did you post that over at avsforum?
mclingo
27th November 2018, 12:36
Trust me, it's different with 2018 OLEDs. For some reason they are pumping brightness through near black and causing near black artifacts to flash brightly. Pause the video and the flashing stops, press play again and the crazy flashing starts again. This does seem to be a bug because sending the same content as bt.2020 instead of bt.709 and this flashing does not occur. That isn't a viable solution but luckily dithering to 5bit also solves the issue (no flashing). This also was not an issue with the two C7s I had. Or any other TV i've had lol.
https://www.avsforum.com/forum/40-oled-technology-flat-panels-general/2962154-2018-lg-c8-e8-owners-thread-no-price-talk-276.html#post57140188
Sounds like a panel fault to me, there were a number of people who had really bad macro blocking on the 2015 panels who got replacement panels which solved it, have you opened at post of this issue on any forum or contacted LG?
also, post a clip of something that partiularly bad so we can have a look, suggest also moving this dicsussion to the general forum as its not a MADVR issue.
chros
27th November 2018, 13:43
https://www.youtube.com/watch?v=Sl_t3UQ3gAk ... Though the few that did got very similar results.
It's just ridiculous for a >3000 TV. But if bt.2020 isn't affected then it's not a hardware issue. That means you have to you bt.2020 all the time even for HD/SD if it's possible.
j82k
27th November 2018, 14:17
Did you post that over at avsforum?
Yes, you must have missed it.
Sounds like a panel fault to me, there were a number of people who had really bad macro blocking on the 2015 panels who got replacement panels which solved it, have you opened at post of this issue on any forum or contacted LG?
also, post a clip of something that partiularly bad so we can have a look, suggest also moving this dicsussion to the general forum as its not a MADVR issue.
Here is the clip of the handmaids tale scene and another from pyewacket (blu-ray) which also causes a lot of artifacting.
Both look clean on my monitor and plasma.
https://drive.google.com/open?id=15HqGEnAutHrSup4NvdXrjiAYo9sjxRao
I've already had a panel replacement and the tech who did that knew about this problem and he also knew that a panel wouln't fix this. Though he still replaced my panel but had no explanation for this issue.
nevcairiel
27th November 2018, 14:21
Do remember that we're talking about madVR here, not generic OLED issues. :)
ryrynz
27th November 2018, 21:26
But.. But madVR can surely help!
Every time we go off topic delays the next stable by a week. Plz pray for madshi's free time.
actarusfleed
28th November 2018, 01:18
Hi guys,
in those days I've compared my stand alone UHD player (Oppo UDP 203) and my htpc with MadVr.
The quest is what is the best for viewing UHD HDR10 Films.
My HTPC is windows 10 based (64bit), GTX970 (latest drivers), last stable madvr, MPC-HC, last LAV filters.
My display is a Sony VPL-VW520ES 4K projector.
My reference UHD material are the first mins of REVENANT and some other particular scenes taken from VALERIAN.
Obviously MadVR is better from any point of view (detail, colors, noise, ecc.) less one ... COLOR BANDINGS!
Oppo is better only under this point of view.
Oppo send a HDR signal 422@12bit to my projector.
Initially my GPU was setted in RGB@8bit.
That's why I read (a long time ago) that MadVR natively works in RGB domain.
Into MadVr I've set display=8bit
Conversion to HDR > SDR with bt.2084.
Debanding filter to Max
dithering = ordered
The image is stunning but color banding is unfortunately present.
After that I've decided to put my GPU in 422@10bit output.
Setted MadVr with display=10bit
Conversion to HDR > SDR with bt.2084.
Debanding filter to Max
dithering = ordered
Now MadVr reports me "window exlusive mode" =10bit
In this mode the color banding effect is a little bit lower ... but Oppo is still better under this point of view.
Can someone tell me if I'm missing something?
Someone can help me to put my htpc to perform bandings equal to my oppo player?
Thank you so much
Warner306
28th November 2018, 02:42
The HDR -> SDR conversion could possibly introduce some banding from heavy compression. Your projector might also have issues with RGB inputs compared to YCbCr. The image is being sent from madVR with dithering to eliminate rounding errors that would cause banding, so the problem would likely start at the projector. Does the projector have any options for video levels or color space?
Asmodian
28th November 2018, 03:58
I suspect the HDR -> SDR if you need to target low nits. I think debanding runs before HDR -> SDR?
The projector would process input from the Oppo similar to madVR, wouldn't it?
Warner306
28th November 2018, 05:49
RGB should produce the best result and that might involve extra conversions at the projector, which might create some errors. But I would also expect the two inputs would look more or less the same. Tone mapping does sound like the more likely cause. No pixels are removed, but the luminance steps can be altered enough to cause uneven gradients.
Some things to try:
Set the HDMI video range to Full or whatever Sony projectors call this setting to send RGB Full;
Disable highlight recovery strength in madVR to reduce the amount of image processing;
Find a scene with some banding and try considerably higher or lower target peak nits to see if the banding changes at all.
madjock
28th November 2018, 12:46
I have been working away for a bit, and have lost track of what is happening with madVR in the AVS forum, it just seems to be confusing graphs and the likes for a long time.
It looks like madshi has dropped out of the conversation as such, is this getting there, as I said all rather confusing on what is happening apart from everyone measuring their movies for NIT settings and fancy apps to display that info ?
curael
28th November 2018, 14:39
----EDIT
For RTX users with stuttering issues, please read, and try Fullscreen Exclusive mode! It solved my issue that I describe below:
----EDIT
Hi guys,
I'm fairly new on here. :) I've been using MadVR in conjunction with my GTX 1070 from EVGA for some years now and I've been very happy, the video clock on the card/display has been VERY close to 23.976, like.. We're talking .979 instead of .976..
My issue now.. Is that I got an RTX 2080, same TV, same PC, same AVR, same MadVR config.. And I'm getting stuttering every 2-4th second, 4K, 1080, 720.. You name it.
The only difference I can see is that my clock on the new card is up to 23.981.. I have an estimated frame repeat expectancy of 4.4 minutes.. Seems fairly accurate too.. No dropped frames, a render time of 20-31ms on 23 content.. Like.. Everything looks okay.
So this insane stuttering that was introduced with the new card.. Can it be fixed? Is there any way I can get closer to the 23.976? And before that.. Does this little difference actually matter?
I have read about CRU and thinking to try it out to try and get my clock closer, I just want to be sure that it's actually worth it at this point. I'm also getting a new TV really soon and I'd be so sad if I can't actually fix this issue since content is pretty much unwatchable right now cause of the insane amount of juddering introduced from the new card.
For the record, I got a 750W PSU along with the RTX card to make sure that I was supplying enough power, up from a 600W, I don't know if this has anything to say in terms of the clock on the display/card.
I am so sorry if I'm derailing but this is the only thread I really know and I know it holds so much good info on everything display, color, MadVR and overall HTPC related so I'm trying my luck.
Thanks so much.
/CuraeL
madjock
28th November 2018, 14:44
@curael
There have been a few posts about stuttering with the new design cards, so you could maybe see if you could find them.
I think there were not many answers, I think the consensus was it was quite immature drivers for the card as it is, and may improve over time, I guess madshi will look at it in time, but the focus is on HDR at the moment.
I also think due to the fact you are using new card and new drivers you may struggle to get a frame rate match as good as your previous card, and that is why many of us are on older drivers for just the older cards, something you will not be able to do I guess as there will new issues for you due to this.
Heres one similair
https://forum.doom9.org/showthread.php?p=1857345#post1857345
curael
28th November 2018, 15:01
@curael
I also think due to the fact you are using new card and new drivers you may struggle to get a frame rate match as good as your previous card, and that is why many of us are on older drivers for just the older cards, something you will not be able to do I guess as there will new issues for you due to this.
Yeah, I found a few comments on here that sounded like they were about my issue but I wasn't sure if I actually found everything.
I'm wondering the same as you, it could be early drivers.. But it could also just be that the RTX cards are a bit too aggressive when it comes to lower video clocks. :( Who knows..
Would "Smooth motion" fix anything if I set the display to only run 60Hz? I tried to look up what Smooth Motion actually did and found out it's an "image blending" technology, I think someone called it over at JRiver (which was what my Google-fu led me to.) How ever, I'm not sure it will do anything as it seems like Smooth Motion looks like it's designed for very specific frame rates and not a huge fix for juddering.
Anyway, I appreciate your answer and I think it confirms my belief that for now, I'm boned.
Thanks again,
/CuraeL
huhn
28th November 2018, 15:06
can you make a screen of the OSD after this has happend.
austinminton
28th November 2018, 15:08
I have a 2080 ti and get stuttering as well. Using fullscreen exclusive fixed it for me.
curael
28th November 2018, 15:16
I have a 2080 ti and get stuttering as well. Using fullscreen exclusive fixed it for me.
Oh..! I've had some issues with FSE in the past but that's just related to my setup I guess.. I don't recall exactly - but I stopped using it due to crashes but that's long ago. I'll certainly give that a try!
I appreciate that very much! Thanks!
/CuraeL
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.