View Full Version : madVR - high quality video renderer (GPU assisted)
CruNcher
6th January 2017, 09:16
or you are not using madVR...:
https://abload.de/img/madvrldp0v.png
this is without debanding. i don't even want to know how you get so much banding in picture.
8 bit output chain because that's what we talked about debanding 8 bit and how efficient MadVR does it :)
10 Bit p010 is real 10 bit and most probably never needs debanding with those inputs and then it's entirely up to your display panel efficiency what you'll see :)
10 bit->8 bit->MadVR (Debanding)->Panel
you now did
10 bit->MadVR (no debanding/no dithering)->Panel
Lets sumarize it :)
10 bit->MadVR (Pure)
http://i1.sendpic.org/t/fn/fnUsFlHUlkwjMxIPaQngAmkdOCI.jpg (http://sendpic.org/view/1/i/3ZQDpARxxnzodT1w5vIVD8zbMvT.png)
10 bit->8 bit->MadVR (Pure)
http://i1.sendpic.org/t/1r/1r07kpEfYj0blHW6ovPt5PDHEkl.jpg (http://sendpic.org/view/1/i/jAjmuzQcT6YW1XSAFAYFv93xc8d.png)
10 bit->8 bit->MadVR Debanding High
http://i1.sendpic.org/t/oi/oiv6UBKxXNMUKfkf6AXq2ppr49d.jpg (http://sendpic.org/view/1/i/yOXETivVuChMjwNCacLVMFH3vjb.png)
10 bit->8 bit->Ordered Dithering->MadVR (Pure)
http://i1.sendpic.org/t/cd/cdfgUmK7Ba5IIhzRBv4h5xbgLrD.jpg (http://sendpic.org/view/1/i/jpZbjLD5mfSbtGKoCYrvINFRpQ3.png)
10 bit->8 bit->Random Dithering->MadVR (Pure)
http://i1.sendpic.org/t/zz/zzmPxtUwZcLyyPlJ7MHWtTv4XAi.jpg (http://sendpic.org/view/1/i/jLL6l8rLUkbRmsdySpdgl80DQsr.png)
Wow here you really see the Power of Ordered/Random Dithering that is Psy wise almost a 100% hit of the input quality without MadVR touching anything.
Now it shouldn't surprise anyone anymore why there is no (No Dithering) Option in Lav Video ;)
ryrynz
6th January 2017, 13:40
Jesus, would you look at that detail loss on the transparent skirt in 8 bit... Crazy.
Looks like there's some debanding required even in 10 bit :D
Clemens
6th January 2017, 15:36
I described my problem here: https://forum.doom9.org/showthread.php?p=1792014#post1792014
Here the ctrl-j output: http://puu.sh/td7rh/19227caded.jpg
Here a sample: https://www.dropbox.com/s/cbbp9073q8b0702/01-06_15-30-50_PEARL%20TV%204K%20UHD_.ts?dl=0
Kind regards,
Clemens
CruNcher
6th January 2017, 15:48
That combination seems to cry after problems 8bit p10 DXVA2 :D
Try Fullscreen Exclusive to get into 10 bit
Clemens
6th January 2017, 15:55
The same problems on a channel with p1010 10 bit 4:2:0 (dxva2)
http://puu.sh/td8tH/3ca44b49e7.jpg
(If sample needed please ask)
I will try your suggestion
EDIT
How do I do this? (sorry I'm a noob :) )
Which shortcut I have to press or which setting I have to change for this?
CruNcher
6th January 2017, 16:07
These 2 options at least
http://i1.sendpic.org/i/Jc/JcN9aftsarvYByHtnmGjzPCtx9.png
Then go into Fullscreen
YxP
6th January 2017, 16:13
That is due to the fade detection kicking in. If you keep debanding at the same strength fade detection doesn't run at all for a small increase in performance, and it never rerenders 5 frames which causes that spike and can cause dropped frames.
This is completely from my memory and don't really want to go into any testing, but try disabling use a separate device for presentation. I did sometime ago and haven't had problems since. I did other changes as well so that's why I'm mostly guessing here.
Clemens
6th January 2017, 16:20
I have done some more tests:
- CPU decoding -> colors are fine (off course no fluent image)
- DXVA2 + custom EVR Renderer -> works fine
- DXVA2 + MadVR -> doesn't work
--> seems to be very similar to the magenta bug in the bug tracker:
http://bugs.madshi.net/view.php?id=456
I guess DXVA2 is doing something wrong...
Have you the same problems with other GPUs? (mine is a GTX 960)
The Exclusive mode doesn't change anything...
Problem is there for all 4k HEVC 10 bit channels (also the HLG teststream channel)
CruNcher
6th January 2017, 17:37
Indeed strange VP7 should do it are you sure the Mixer input output types are 10 bit with EVR Custom rendering and DXVA ?
And are you in 10 bit Display Driver Mode also for MadVR and still in Exclusive Fullscreen you get those color shifting ?
Clemens
6th January 2017, 17:47
I have no idea what the EVR renderer does :) How can I check this?
And I'm using a 8bit UHD display.
If you mean with your question "the native display bitdepth is:" then I can say I tried 10 bit and 8 bit there no difference.
Can you play the sample file without any problems (some posts above)? Would like to know if only I have the problem.
And what do you mean with the VP7 thing? (as said I am a noob :) )
Btw. thanks for looking at it (it's not that important to me I just thougt to report it to make madvr better if the bug is in madvr)
Damien147
6th January 2017, 17:47
I have done some more tests:
- CPU decoding -> colors are fine (off course no fluent image)
- DXVA2 + custom EVR Renderer -> works fine
- DXVA2 + MadVR -> doesn't work
--> seems to be very similar to the magenta bug in the bug tracker:
http://bugs.madshi.net/view.php?id=456
I guess DXVA2 is doing something wrong...
Have you the same problems with other GPUs? (mine is a GTX 960)
The Exclusive mode doesn't change anything...
Problem is there for all 4k HEVC 10 bit channels (also the HLG teststream channel)
Same results with an RX 470 here,dxva2(native) gives black screen.
Damien147
6th January 2017, 17:52
Have a Rx460, don't seem to have much problem with ngu 1080p to 4k.
Really?With rx470 the best I can do is ngu med.Is there something wrong with my system?:scared:
Asmodian
6th January 2017, 17:55
10 bit->8 bit->MadVR (Debanding)->Panel
That isn't how madVR works, what are you doing to get it to 8-bit before madVR?
10 bit->16 bit madVR (debanding)->8-bit->Panel
Really?With rx470 the best I can do is ngu med.Is there something wrong with my system?:scared:
ngu low is consistent with "NGU 1080p to 4k"? ;)
CruNcher
6th January 2017, 18:00
10 bit P010->YV12 8 bit (no dithering)->MadVR Debanding High
Is the heavy banding result above
with 10 bit P010->YV12 8 bit (Ordered/Random Dithering) has the biggest visual Deband effect, you virtualy dont need MadVR Debanding High and it's overhead anymore.
Clemens
6th January 2017, 18:10
Other test done with my problem: DXVA2(copy back) works also fine and with correct colors...DXVA2(native) gives the purple screen...
So:
DXVA2 (native) -> EVR custom -> OK
DXVA2 (native) -> MadVR -> NOT OK, purple colors
no Hardware Decoding -> OK with both (but stuttering bacause of software decoding)
DXVA2 (copy back) -> both OK
CruNcher
6th January 2017, 18:14
Same with the Samsung 10 bit Streams ?
http://demo-uhd3d.com/fiche.php?cat=uhd&id=91
http://demo-uhd3d.com/fiche.php?cat=uhd&id=97
Hmm so DXVA Native wont work, but why ?
Native Display Bitdepth (Driver) should be 10 bit The Controll Panel should be set to 10 bit
Though if it only would work with copyback that wouldn't be really that bad the overhead is not that much bigger normally, though for people with laptops that want maximum power saving could be a problem.
VP7 stands for Video Processor (Version 7) the VPU (Video Processing Unit) inside of your GM206 GTX 960 also called marketing wise Nvidia PureVideo.
Damien147
6th January 2017, 18:33
ngu low is consistent with "NGU 1080p to 4k"? ;)
The performance gap between these two seemed bigger.Well...ok.
Clemens
6th January 2017, 18:34
yes same thing with the demo...with copy back it works, with native wrong colors...
huhn
6th January 2017, 18:44
8 bit output chain because that's what we talked about debanding 8 bit and how efficient MadVR does it :)
10 Bit p010 is real 10 bit and most probably never needs debanding with those inputs and then it's entirely up to your display panel efficiency what you'll see :)
sorry you should read up the basic how picture processing is working in a PC.
i will never post you a 10 bit screen because it is technically not doable.
setting the output to 10 bit doesn't mean you get 10 bit too. the windows desktop is working in 8 bit.
madVR is not debanding when outputting 8 bit.
@Clemens
DXVA native is a lowering picture quality with nvidia and it should be avoided using madVR.
your GPU is not by any chance a 2 GB GTX 960?
sneaker_ger
6th January 2017, 18:49
madVR is not debanding when outputting 8 bit.
What exactly are you trying to say?
Madvr is debanding when - unsurprisingly - debanding is ticked in the settings. This is also true for 8 bit output.
huhn
6th January 2017, 18:53
that p010 input with 8 bit output doesn't mean it is debanding.
Clemens
6th January 2017, 18:56
It's a 4GB GTX 960
sneaker_ger
6th January 2017, 18:57
that p010 input with 8 bit output doesn't mean it is debanding.
True but I don't think CruNcher said that anyways.
Like Asmodian said madvr doesn't do 8 bit internally so I assume CruNcher did a conversion from 10 bit to 8 bit to before madvr. One without dithering, that is.
Damien147
6th January 2017, 18:59
My tv has ''pseudo'' hdr.In madvr is it better to leave default value for hdr or passthrough?Tried ''Samsung_HDR_Wonderland'' clip and with passthrough picture seems more dull.:confused:
Asmodian
6th January 2017, 19:14
10 bit P010->YV12 8 bit (no dithering)->MadVR Debanding High
Is the heavy banding result above
with 10 bit P010->YV12 8 bit (Ordered/Random Dithering) has the biggest visual Deband effect, you virtualy dont need MadVR Debanding High and it's overhead anymore.
Of course that has banding! :devil:
My claim that madVR's 8-bit is good expects using 10-bit input (or whatever the source is), going to 8-bit only happens at the very end, right before display.
If you are rounding without dithering before madVR you are adding the banding to the source in the first place. :p
CruNcher
6th January 2017, 19:34
True but I don't think CruNcher said that anyways.
Like Asmodian said madvr doesn't do 8 bit internally so I assume CruNcher did a conversion from 10 bit to 8 bit to before madvr. One without dithering, that is.
Yes someone got it i compared the Impact of nevcairiels Ordered/Random Dithering vs MadVRs Debanding on the Pure 8 bit converted non dithered output for a 8 bit output chain.
I just was curios which has the most impact and which overhead is more worth it and who might introduce negative effects :)
Jesus, would you look at that detail loss on the transparent skirt in 8 bit... Crazy.
Looks like there's some debanding required even in 10 bit :D
Thats caused by the MadVR Debanding on High
Extreme Detail lose
It acts more like a denoiser at that setting.
Georgel
6th January 2017, 21:00
It's a 4GB GTX 960
First please uninstall current driver using free tool DDU
http://www.guru3d.com/files-details/display-driver-uninstaller-download.html
Please install Nvidia driver 373.06 then
Please apply this and see if you still have any problems
http://i.imgur.com/SPvlFIC.png
After this, the problem should be solved :confused:
huhn
6th January 2017, 21:00
https://forum.doom9.org/showpost.php?p=1792149&postcount=41901
would help a lot to know that you round down this image to 8 bit first. and there was never a reason to do that.
aufkrawall
6th January 2017, 21:06
It acts more like a denoiser at that setting.
Which is wonderful for most web-video <4k, which often also have lots of blocking.
cork_OS
7th January 2017, 12:30
Which is wonderful for most web-video <4k, which often also have lots of blocking.
With lots of blocking deband do nothing:
no madVR deband
https://s30.postimg.org/e309fadb1/no_DB.png (https://postimg.org/image/e309fadb1/)
madVR deband high
https://s30.postimg.org/5uybualel/high_DB.png (https://postimg.org/image/5uybualel/)
SmoothD2
https://s30.postimg.org/qiwz916n1/Smooth_D2.png (https://postimg.org/image/qiwz916n1/)
Deblock_QED_MT2Mod
https://s30.postimg.org/ua6jvckbh/Deblock_QED_MT2_Mod.png (https://postimg.org/image/ua6jvckbh/)
MSU Smart Deblocking (VDub plugin with RGB32 conversion)
https://s30.postimg.org/5w89npn8d/MSU_Smart_Deblocking.png (https://postimg.org/image/5w89npn8d/)
Source sample (http://www.mediafire.com/file/9c8ezi3gtojktk4/Blocking_H.264_Eskov.mp4) (frame 361).
EDIT: my apologies for this sample, I've decided that it is irrelevant - file seems corrupted or improperly encoded.
alexnt
7th January 2017, 21:58
Hi!
I have been using madvr with mpc for a long time(very old versions for both software) but not with very complicated settings and I was very happy with it. I used the high settings of the image below:
http://i.imgur.com/HLv1qmj.png%20/
So I installed recently a fresh copy of windows 10, installed again mpc and madvr. In the latest versions of madvr I noticed that there are some changes that make software more complicated for me.
For example, I noticed that there is no catmull-rom for downscaling. Is there any alternative? Do I actually use downscaling if I watch only 720p and 1080p movies(mkv) on a 1080p60 monitor?
I also noticed that jinc cannot make 3 taps anymore, it is greyed out.
I cannot find image doubling neither in the older, nor in latest version of madvr.
So my settings right now are chroma upscaling-> jinc(ar), image upscaling->jinc(ar) and image downscaling->jinc (ar, LL). My rig specs are cpu 4690k, 16gb ram, ssd and r9 290. Are the above settings of madvr ok with the specific hardware. Sould I change anything?
I also had an issue with fullscreen video and a memory leak but I found a solution with direct3d11.
My question here is if this option is better than d3d9 regardless the issue mentioned above.
That's all for now. Thank you in advance and sorry for my english... :p
Asmodian
7th January 2017, 22:12
Catmull-Rom is Cubic -> Bicubic50, but you don't use downscaling when watching 720p or 1080p on a 1080p display.
Jinc is always 3 taps now, the 4+ options have been removed because they only made the quality worse.
Image doubling is set in Image Upscaling now, simply choose a doubling algorithm (super-xbr, NNEDI3, NGU). I recommend you test NGU for yourself, for 720p -> 1080p it is more subtle, but it is madshi's amazing new scaler that is very sharp but without artifacts. It runs quite well on the r9 290.
edit: DX11 is the preferred option for AMD GPUs. AMD just seems to like directX 11 more, there have been many user reports of issues with the various DX9 modes and AMD GPUs.
huhn
8th January 2017, 01:46
you can lower the present in advanced setting to 6 and it will work with D3D9 again.
Bandits
8th January 2017, 04:50
Having issues with chroma at 720x480 with current madVR.
I have completely restored settings to default. Cleared registry and deleted settings.bin Using 1920x1080 display full screen with GTX 1050 and latest drivers.
Playing standard Blu-Ray 1920x1080 full screen 23.970.
Chroma > whatever I have selected in settings.
Playing standard DVD 720x480 full screen 59.940
Chroma > DXVA
Luma > whatever I have selected in settings.
Chroma > whatever I have selected in settings.
Is there a reason DXVA is being enabled? Is there a reason 2 chroma settings are applied at the same time? Is there a way to ensure DXVA is never enabled? I have never had this issue before but it has been a while since I played a DVD.
AngelGraves13
8th January 2017, 04:56
Is there a reason DXVA is being enabled?
Don't select DXVA?
ryrynz
8th January 2017, 05:18
Angel Graves, could you try avoiding quoting an entire post when replying directly after the question? Look at the size of your post vs what you wrote. Try and keep it tidy, cheers.
Asmodian
8th January 2017, 05:37
Is there a reason DXVA is being enabled?
In madVR there are two trade quality for performance options addressing DXVA chroma scaling. Maybe deinterlacing activates for the DVD and you have "use DXVA chroma upscaling when doing DXVA deinterlacing" enabled. It is enabled by default.
Do you have DXVA Native decoding configured in LAV Video for DVDs (if so I recommend you disable it)? There is also the option "use DXVA chroma upscaling when doing native DXVA decoding" you could disable.
baii
8th January 2017, 07:08
The performance gap between these two seemed bigger.Well...ok.
If I don't go fancy with other option(essentialy smooth motion on only and nothing else, no 3dlut,) It tcan squeeze ngu med on 1080p23 , maybe 30fps too. 470 is closer to 480, so high should be attainable with some tweaking.
70MM
8th January 2017, 08:52
I still havent gone to the new firmware with NGU as Im so happy with what Im currently using.....
Is there an updated tutorial yet with screenshots setting up using NGU and any other new settings?
JackCY
8th January 2017, 13:00
The way I wrote the profile rules there is nothing wrong with using 1920 or 1080, etc. 1916 would be less than or equal to 1920 but more than 1280, exactly as we want.
How do you start your rules then? You cannot use "OR" instead when using "<=". I see no reason to use slightly off values for 1920, in fact wouldn't that cause problems?
if (srcWidth > 1024) OR (srcHeight > 576) "720p"
else if (srcWidth > 1280) OR (srcHeight > 720) "1080p"
else if (srcWidth > 1920) OR (srcHeight > 1088) "UHD"
else "SD"
Do you mean e.g. keep both ticks the same med/med?
if (deintFps > 31) or (srcWidth > 2100) or (srcHeight > 1200) "2160p"
else if (srcWidth > 1400) or (srcHeight > 800) "1080p"
else if (srcWidth > 800) or (srcHeight > 600) "720p"
else "480p"
But if anyone knows of srcArea or how to get it I would use that. Unfortunately when I made this profile selection years ago there still was no srcArea and sercWidth*srcHeight didn't work.
Did Madshi add srcArea finally? I guess not? Which is truly a shame when making profiles.
Maybe this example convince you more (off/low/medium/high) :
https://s27.postimg.org/6eyf8rugv/0_off.png (https://postimg.org/image/6eyf8rugv/) https://s27.postimg.org/ovstzlaf3/1_low.png (https://postimg.org/image/ovstzlaf3/) https://s27.postimg.org/52gq6vx1b/2_medium.png (https://postimg.org/image/52gq6vx1b/) https://s27.postimg.org/rsfuzvg8v/3_high.png (https://postimg.org/image/rsfuzvg8v/)
Open in original size of you won't see anything.
Banding is still very noticeable on the top-left side with the "low" setting and you can clearly see here the details lost on the bottom of the image in the trees with the "high" setting.
I've had no problem to see the de/banding on the previous examples with back lit woman but this one with trees and clouds I don't really find any beyond some tiny detail removal with high debanding again.
I still havent gone to the new firmware with NGU as Im so happy with what Im currently using.....
Is there an updated tutorial yet with screenshots setting up using NGU and any other new settings?
NGU works OK, disable SuperRes but it asks anyway, higher quality = less vector like and softer edges, it's like a mix of scaling and postprocessing in one, that's how it seems to me, with a noticeable flair of vectorization. Sort of nonglobal scaler, where different scaling is applied to different parts of the image. Where as NNEDI + SuperRes is global and applies the same to everything, problem is SuperRes tends to kill the advantages of using NNEDI and NNEDI alone is quite soft.
With NGU you are fine just with NGU, no postprocessing needed to be added, though I do anyway for personal preference.
It's the same as before, clickity click in a GUI, only now there is less flexibility than before with the updated upscaling selection.
Bandits
8th January 2017, 18:37
Maybe deinterlacing activates for the DVD and you have "use DXVA chroma upscaling when doing DXVA deinterlacing" enabled. It is enabled by default.
Thanks, this was the issue. LAV video decoder is set to deinterlacing mode (auto) and it seems all the DVD's I was trying had a small clip that was indeed interlaced but the actual movie was not. In conjunction with the madVR tradeoff settings caused DXVA to enable on chroma making chroma upscaling appear twice in ctrl+j.
madshi
8th January 2017, 19:40
Hey guys,
just wanted to let you know that I'm a bit busy right now, so I didn't get around answering all your comments & questions yet. Might get to that next weekend.
For now I just want to release a new build, so you have something new to play with.
madshi
8th January 2017, 19:45
madVR v0.91.5 released
http://madshi.net/madVR.zip
* added direct 4x NGU upscalers
This is a first test build with the direct 4x NGU upscalers. Nothing else has changed.
The main benefit of the direct 4x upscalers should be an improved "quality per watt". In this build the direct 4x upscalers become active only if chroma doubling quality is set to "normal". I've done this intentionally so that you can easily compare the 4x algorithm to running 2x twice, which is still done when using "high" or "very high" chroma doubling quality.
For those of you who upscale low quality SD videos, it might be worth checking if the direct 4x upscalers work better than doubling twice. There might be a small quality improvement for you. But ultimately, I'm planning some other improvements for you guys.
(I'm planning to improve the settings dialog in the next build.)
cork_OS
8th January 2017, 21:12
For those of you who upscale low quality SD videos, it might be worth checking if the direct 4x upscalers work better than doubling twice. There might be a small quality improvement for you.
Some low-res videos after NGU doubling twice looks insanely sharp, 4x is better in that case.
But ultimately, I'm planning some other improvements for you guys.
Really glad to hear that! Huge thanks for your work.
Neo-XP
8th January 2017, 21:12
madVR v0.91.5 released
http://madshi.net/madVR.zip
* added direct 4x NGU upscalers
This is a first test build with the direct 4x NGU upscalers. Nothing else has changed.
The main benefit of the direct 4x upscalers should be an improved "quality per watt". In this build the direct 4x upscalers become active only if chroma doubling quality is set to "normal". I've done this intentionally so that you can easily compare the 4x algorithm to running 2x twice, which is still done when using "high" or "very high" chroma doubling quality.
For those of you who upscale low quality SD videos, it might be worth checking if the direct 4x upscalers work better than doubling twice. There might be a small quality improvement for you. But ultimately, I'm planning some other improvements for you guys.
(I'm planning to improve the settings dialog in the next build.)
The 4x algorithm is unfortunately too soft compared to the 2x2x method, but I guess some users will like it because of this (for instance for low quality SD videos like you said)
I can run NGU high 4x with my GTX 970 (I can't run NGU high 2x2x), but I don't like the result, because I like very sharp images.
NGU high (former medium) was already softened in v0.91.2, and what I liked best with NGU originally was its very strong and clean sharpen.
While testing, I have found the absolute best result for me to be NGU veryHigh (former High) 2x2x in v0.91.1, but I need to change my graphic card to run it :D
I still hope I will be able to use Lanczos3 AR for chroma doubling in next versions (instead of Bicubic60 AR), disable image quadrupling (but still be able to activate 2x supersampling) and that NGU high (former medium) will retrieve its original sharpness (or at least have an option to choose the sharpness, like other algorithms).
:thanks:
thighhighs
8th January 2017, 21:47
720x400@23 -> 1920x1080 NGU quadrupling:
v0.91.4 very-high + high ~30ms
v0.91.5 very-high direct 4x ~23ms
Nice improvement! :thanks:
JackCY
8th January 2017, 22:12
Well here is my experience with the 4x, image quality wise high4x looks as if running high2x + high2x instead of high2x + med2x. Power wise similar barely noticeable, would have to run some long averages to give you a precise number of watts saved reported by the GPU.
So I would say 4x looks "better" and could consume less power.
On the other hand I have to agree that soft DVD sources do need a bit of edge presharpen so that NGU can do a clean edge instead of trying to "vectorize" the thin soft edge gradient with a "single color" vector area, this happens more with the higher quality settings on some samples. With a sharp 4k recording downscaled to DVD resolution it wouldn't be an issue but these older DVDs mastered from film are often soft, if not then they are ruined to hell with a global sharpener and halos by who ever was attempting to convert the film to DVD.
It would be nice to have a variable in the GUI that would change how sensitive NGU is to these soft edges so that one can dial it in.
Will see later when the other improvements are released if it's needed.
Overall I do postedge sharpen and thin edges as before without NGU, looks better to me. Sure NGU is sharper but I guess as other have experienced previous versions, it was even more sharp before and these postprocess filters wouldn't be needed.
I don't mind the conservative approach so far in madVR to default settings, it works reasonably well but sometimes it could be tweakable.
Especially considering NGU, NNEDI, XBR are often used on low quality low resolution sources as doing 2-4x on an HD-FHD source there is really not much visual difference at that level of source detail and the size of target display.
70MM
8th January 2017, 23:08
For those of you who upscale low quality SD videos, ultimately, I'm planning some other improvements for you guys.
Thank you madshi for work in this area as I use alot of low quality material along with BD 1080. Already I see my 480/576 material look way better than using my Oppos, if we can get more out of it I will be over the moon!
Georgel
8th January 2017, 23:11
madVR v0.91.5 released
http://madshi.net/madVR.zip
* added direct 4x NGU upscalers
This is a first test build with the direct 4x NGU upscalers. Nothing else has changed.
The main benefit of the direct 4x upscalers should be an improved "quality per watt". In this build the direct 4x upscalers become active only if chroma doubling quality is set to "normal". I've done this intentionally so that you can easily compare the 4x algorithm to running 2x twice, which is still done when using "high" or "very high" chroma doubling quality.
For those of you who upscale low quality SD videos, it might be worth checking if the direct 4x upscalers work better than doubling twice. There might be a small quality improvement for you. But ultimately, I'm planning some other improvements for you guys.
(I'm planning to improve the settings dialog in the next build.)
Looking forward for the improvements when upscaling from SD!
Sounds like something that might really help us out!
huhn
8th January 2017, 23:26
The main benefit of the direct 4x upscalers should be an improved "quality per watt". In this build the direct 4x upscalers become active only if chroma doubling quality is set to "normal". I've done this intentionally so that you can easily compare the 4x algorithm to running 2x twice, which is still done when using "high" or "very high" chroma doubling quality.
ok i can do 720p -> 2160p with these ms
downscaler is bicubic 60 AR.
ngu very high x4 32 ms
chroma is bicubic 60 AR
ngu very high x2 -> ngu high x2 takes 90-110 ms
chroma is NGU low -> bicubic 60 AR
and i'm still on polaris.
the first thing is why am i able to use NGU very high x4 on a RX 480? and how should i compare them when chroma high is using NGU low?
i have to switch madVR versions to do a proper comparison.
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.