madVR - high quality video renderer (GPU assisted)

[Sunset1982]

ok, had showrendersteps activated too.

Performance is a bit better now:

can use NGU med with chroma and luma. 30ms

choma reconstruction soft and double luma with NGU med: 23ms
choma reconstruction soft and double luma with NGU high: 59ms


sadly, NGU high is still to much for my rx480.

I will do some more testings tomorrow. To bad, I have to work tonight...

[ryrynz]

Do you have a "groundtruth" to compare to in any of your tests? Maybe an old SD encoding of some anime for which a Blu-Ray is now available?

For one I do, my settings including Avisynth actually produced a fairly similar image. Anyway I have a screenshot that kinda shows what I'm talking about.

Yes, I got this too. Normally madVR causes just some electric buzzing and not this high "coil whine" sound, which is very strange.

It depends on the type of load required by the card, generally the harder it has to work the louder the whine. Best thing you can try and do to minimize it is try overclocking or just replace the card with one that doesn't do it.

Madshi,
http://screenshotcomparison.com/comparison/191145

This is with Lanczos 3 as second upscaler, normally I'd use Jinc but w/e, this kinda helps exacerbate the issues anyway..
This is just clean NNEDI3 64 vs NGU HQ SE1, SE2 is just too blurry to be used for any real comparison.


sadly, NGU high is still to much for my rx480.


That sucks, my 960 breezes through it, doesn't use much more performance than NNEDI3 64 neurons, extra few ms render time that's it =/
I've only tested 480 > 1080 though..

[madshi]

that was NGU only without any other renderstep using showrendertimes because everyone has different chroma scaler and stuff like that. didn't worked out...

FHD -> UHD normal rendertimes
NGU med 25-30 ms
nnedi64 32 ms
NGU high ~80-100ms
nnedi3 256 110ms
nnedi3 128 55ms


NGU med is about 10 ms faster now since i disabled showrenerstep and reset all settings. NGU high doesn't seem to care.
Still far from what you should get. I don't suppose it helps if you switch the first flush setting "after intermediate render steps" to "don't flush", does it? Does turning on/off "use D3D11 for presentation" or "use separate device for presentation" make any noticeable difference at all?

[madshi]

ok, had showrendersteps activated too.

Performance is a bit better now:

can use NGU med with chroma and luma. 30ms

choma reconstruction soft and double luma with NGU med: 23ms
choma reconstruction soft and double luma with NGU high: 59ms
As I said (multiple times now). The performance by itself doesn't matter. I need NNEDI3 comparison, otherwise there's no reference point to compare to.

http://screenshotcomparison.com/comparison/191145

This is with Lanczos 3 as second upscaler, normally I'd use Jinc but w/e, this kinda helps exacerbate the issues anyway..
This is just clean NNEDI3 64 vs NGU HQ SE1, SE2 is just too blurry to be used for any real comparison.
That looks pretty ugly. I don't really see an issue with lines being too thin here, though. What I do see is that NGU hates compression artifacts. But I suppose you'll probably get somewhat similar results when using NNEDI3 + SuperRes 4?

[huhn]

Still far from what you should get. I don't suppose it helps if you switch the first flush setting "after intermediate render steps" to "don't flush", does it? Does turning on/off "use D3D11 for presentation" or "use separate device for presentation" make any noticeable difference at all?



"don't flush" no change.
"use separate device for presentation" no change
using D3D9 present in advance with FSE instead of D3D11 gave me maybe an 1 ms but NGU rendertimes are unstable anyway.

[ryrynz]

That looks pretty ugly. I don't really see an issue with lines being too thin here, though. What I do see is that NGU hates compression artifacts. But I suppose you'll probably get somewhat similar results when using NNEDI3 + SuperRes 4?

There's an O on his helmet, looking at the top line of that right in the middle you can see NGU really thinning that area far more than it should.

SuperRes4 looks even nastier but then it doesn't thin the lines, naturally.

So you could see why I'd prefer NNEDI3, after Awarpsharp etc has taken care of it I get a very nice sharp & clean image.

[Sunset1982]

some quick tests:

Chroma & Luma doubling both @ NGU med: 32ms
Chroma & Luma doubling both @ NGU high: 79 ms
Chroma & Luma doubling both @ NNEDI32: 25 ms
Chroma & Luma doubling both @ NNEDI64: 36ms

Chroma @ NGU med & Luma doubling @ NGU High: 60ms

luma & Chroma doubling set to NGU high (luma) and NGU med: 78 ms


So, still NGU high is no option on rx480 yet.

[ryrynz]

some quick tests

I wouldn't even bother with chroma doubling with this in it's current state, but that's me.. not sure if it'll make enough of a change in performance for you to run it. But like madshi said, there isn't a huge visual difference between medium and high. But I guess you're just testing rather than anything else..

[madshi]

There's an O on his helmet, looking at the top line of that right in the middle you can see NGU really thinning that area far more than it should.
I see what you mean. However, why is that part of the O gray in the source and not black, like the rest of the O? If the whole O had the same line thickness and line color in the source, NGU wouldn't "focus" some parts of the O but not others.

In order to properly upscale low-res images to hi-res in a way to get as near to the groundtruth as possible, NGU needs to decide which lines in the source were likely bloated through downscaling and were much thinner in the original hi-res image, and which lines were already thick in the hi-res source. NGU can't just thin all lines by a certain factor (as awarpsharp does). While that might produce pleasing results for very low quality anime sources with high compression artifacts, it would produce results very far from the groundtruth for high quality sources.

SuperRes4 looks even nastier but then it doesn't thin the lines, naturally.
That's "good" to hear. So it's really mostly caused by the bad source quality. I think what you need most of all is a good algorithm to remove compression artifacts.

some quick tests:

Chroma & Luma doubling both @ NGU med: 32ms
Chroma & Luma doubling both @ NGU high: 79 ms
Chroma & Luma doubling both @ NNEDI32: 25 ms
Chroma & Luma doubling both @ NNEDI64: 36ms

So, still NGU high is no option on rx480 yet.
Ok, so no joy with the RX480, for some weird reason.

I'll try my luck with D3D11 shaders, maybe that solves the issue.

[huhn]

well maybe wait for a working RX 480 driver...

i would down grade to a better driver but i can't remember a driver that doesn't have serious issues and is working with the RX 480.

[AndreaMG]

@Madshi Windows1064bit latest and gtx970 here, "install.bat" hangs and won't install, never had such problems before.:thanks:

[ryrynz]

I see what you mean. However, why is that part of the O gray in the source and not black, like the rest of the O? If the whole O had the same line thickness and line color in the source, NGU wouldn't "focus" some parts of the O but not others.


Yeah, I understand. NGU isn't tailored to this sort of content and NNEDI3 does particularly well here. If you ever did another version of this that went a little less towards ground truth then I'd give it a shot, this sort of content needs something a little less aggressive with it's line prediction.


That's "good" to hear. So it's really mostly caused by the bad source quality. I think what you need most of all is a good algorithm to remove compression artifacts.


That would certainly help, I could take care of this in my script but that wouldn't solve my main contention with this.

[madshi]

Here's a test build for those of you who still have problems with keyboard shortcuts and NGU not working:

http://madshi.net/madVR91c.rar

Installation instructions:

1) Make sure you're in release mode.
2) Overwrite the files.
3) Activate debug mode.
4) Create a short log file and upload it somewhere for me to look at (don't attach to this forum).

[Sunset1982]

I'll try my luck with D3D11 shaders, maybe that solves the issue.

I'm ready to do some tests. Let me know if I can do some testing for you.


well maybe wait for a working RX 480 driver...

i would down grade to a better driver but i can't remember a driver that doesn't have serious issues and is working with the RX 480.

They bring new beta drivers every week atm, but sadly just optimized for new games instead of bringing a bug free fro multimedia use.

[wanezhiling]

@Madshi Windows1064bit latest and gtx970 here, "install.bat" hangs and won't install, never had such problems before.:thanks:
same here
win10 14393 64-bit

[chros]

Here's a test build for those of you who still have problems with keyboard shortcuts and NGU not working:

http://madshi.net/madVR91c.rar

Installation instructions:

1) Make sure you're in release mode.
2) Overwrite the files.
3) Activate debug mode.
4) Create a short log file and upload it somewhere for me to look at (don't attach to this forum).
I don't have time right now to try this, but 0.91b doesn't work for me either (keyboard shortcut). 0.90.24 doesn't have this issue. (Optimus system.)

Thanks for the new version!

[MariaX9]

Performance:
R9 GPU
chroma and image doubling NGU High: 20s
chroma and image doubling NNedi 64: 22s

I did a reset of the Settings and the NGU High render time lowered to 18s with the same Settings as before. Dont know why, but the IQ look better now as before IMO.

[madshi]

I don't have time right now to try this, but 0.91b doesn't work for me either (keyboard shortcut).
It's not supposed to work. It's supposed to create a log file with information that might help me figure out why it's not working... ;)

[Thunderbolt8]

3) NGU is currently available for both doubling and chroma upscaling. The chroma upscaling algo does currently not make use of the luma channel. I plan to try adding that to a future NGU version, but that will take serious development time, so don't expect it too soon.what does this mean specifically when watching a (non animated) movie? that the algorithm only makes use of half of the picture or picture information so to say? that it is incomplete atm?

---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

the install and unistall batch file of that new version dont work for me. when I execute it nothing happens e.g. the black cmd window remains opened without anything happening (admin rights activated). I do have an optimus card.

is there any other way to install this manually? I did regsvr32 madvr.ax and same with madvr64.ax, but in mpc-hc madvr remains not installed in fact. its working fine with 0.90.24 though.

[Ver Greeneyes]

what does this mean specifically when watching a (non animated) movie? that the algorithm only makes use of half of the picture or picture information so to say? that it is incomplete atm?Some chroma upscaling algorithms like Bilateral and Reconstruction use the luma channel, which has twice the horizontal and vertical resolution, to guide the way they upscale the chroma channel. This can be very helpful when there's a lot of overlap between the chroma and the luma (e.g. if a change in color is usually accompanied by a change in brightness), but it can also lead to artifacts where they don't match well enough. NGU for chroma upscaling doesn't make use of this technique yet.

[CarlosCaco]

I will try NGU tonight... i don't have time now for tweaking so I want know, you guys suggest don t use any sharpener? And enhance details ?


Enviado do meu iPhone usando Tapatalk

[heiseikiseki]

Here's a test build for those of you who still have problems with keyboard shortcuts and NGU not working:

http://madshi.net/madVR91c.rar

Installation instructions:

1) Make sure you're in release mode.
2) Overwrite the files.
3) Activate debug mode.
4) Create a short log file and upload it somewhere for me to look at (don't attach to this forum).

Thanks.
Here is my smart log download link.
https://www.dropbox.com/s/p7p2611ene8ksrb/heiseikiseki%20madVR%20-%20log.zip?dl=0

In debug mode ,when I press CTRL+J it shows the OSD.
But NGU doesn't work.

[madshi]

Thanks.
Here is my smart log download link.
https://www.dropbox.com/s/p7p2611ene8ksrb/heiseikiseki%20madVR%20-%20log.zip?dl=0
Thanks. Are you an Optimus user? Or are you running any external software like FRAPS, f.lux, GpuZ or Reclock which might access the GPU, too?

[mzso]

madVR v0.91.0 released

2) In this initial version only 2 quality levels (high and medium) are available. A future version will add at least one more level which is faster than medium, maybe there will be even more levels, I don't know yet.

How does the speed compare to older methods? Is it slower than anything? Or between some of the old ones?

However, texture detail which is lost during downscaling cannot properly be restored. This can lead to "cartoon" type images when upscaling by large factors with full sharpness, because the edges will be very sharp, but there's no texture detail.

I don't like these add noise "enhancements". Which reminds me:

Most movie sources have a bit of grain in them. The way the deringing works, it totally smoothes areas where ringing was detected. As a result these areas look smoother than the rest of the image, which is a visible artifact. So I'm adding grain to make these areas appear more natural and not overly smoothed over.
So what if the movie is essentially (perceptually) grainless? (Which will be more-and-more common as movie digital cameras get better)
Then this will add a grainy halo. Wouldn't a de-grain filter be more useful? (for actually grainy movies)
It would be nice to optionally disable at least.

[heiseikiseki]

Thanks. Are you an Optimus user? Or are you running any external software like FRAPS, f.lux, GpuZ or Reclock which might access the GPU, too?

yse,I'm an Optimus user.
But none for the others.
Also I didn't met the same problem before 0.91.0.

[Georgel]

GTX 1080 user - NGU doesn't pull more than 90W of GPU, it's almost as if it doesn't access all of GPU power.

How do I explain it... With NNEDI 256 and im doubling I can pull 210W from GTX 1080, but with NGU the render time grows to ~60 ms, and the GPU just doesn't get more usage, tested powers and usage / frequency using HWiNFO

As it looks right now, there's also a problem with routing the usage of GPU clocks. The GPU stays at low power even with NGU set to high in both Chroma and IMG doubling.

[Georgel]

I tried a few different configurations and situations, running Nvidia Driver 373.06, the latest stable Nvidia driver.

Seems that no app, nothing can flag NGU to use the GPU properly. On a similar note, NNEDI 256 will start dropping frames before hitting 100% GPU core, but NGU doesn't reach 40% of GPU core usage.

I don't use FRAP, f.lux, GPUZ, Reclock or other software that might access the GPU, but I have no iGPU installed (Clevo P775 disables iGPU entirely, no Optimus, no tricks).

[madshi]

Are you using the "ShowRenderSteps" empty file hack?

[Razoola]

Im on win7x64, I had the shortcut problem with the official but but the first fix you did fixed the issue for me (madVR91b).

[Georgel]

Are you using the "ShowRenderSteps" empty file hack?

Yes, I was. :o

Totally forgot that it was on and it might affect it.

Cleared it out, tested, 230W max power pull on GPU, max OC clocks achieved.

Nice job!

The visuals are amazing, so much sharpness. At first I almost through why do we have an edge softening algorithm, now I get it :D

Now time to fix my setup and settings, because my actual settings add blur and over sharpening over certain edges... :o

Something I noticed recently, whenever I play a really low resolution video (under 720P) there is an audible feedback coming from the GPU, like a high pitched sound, which is not related to the fan. No idea what it is...

[Mistery 73]

excuse my ignorance as I turn off ShowRenderSteps

[imhh11]

Do you have an Optimus system?

Do you have any 3rd party software running which might also access the GPU, like f.lux, FRAPS, GpuZ or something like that? Or maybe some sort of debugger or something?




No 3rd party software, no flux, fraps or gpu-z. No optimus.
Gtx1070 latest driver
Latest version of lavfilter,mpc-hc
Windows10 64bits V1607

What i don't understand, is that it worked first try on my htpc (gtx960)even without the madVR91b.rar fix but not on my editing pc(gtx1070).
And my htpc has the same windows/software in it.
On my gtx960, i can only use NGu med without image doubling. But quality is as good as Nnedi3 64 + image doubling. It is probably normal for 4k upscaling.

[mrcorbo]

For people reporting that install.bat doesn't work in Windows 10:

madshi knows. It's been reported multiple times already.

To work around this you just need to copy the files from the new archive into the folder of the older version, overwriting the old files. I *always* update madvr this way. Since this update method doesn't change where the madvr files are located, it is not necessary to run install.bat.

[Georgel]

No 3rd party software, no flux, fraps or gpu-z. No optimus.
Gtx1070 latest driver
Latest version of lavfilter,mpc-hc
Windows10 64bits V1607

What i don't understand, is that it worked first try on my htpc (gtx960)even without the madVR91b.rar fix but not on my editing pc(gtx1070).
And my htpc has the same windows/software in it.
On my gtx960, i can only use NGu med without image doubling. But quality is as good as Nnedi3 64 + image doubling. It is probably normal for 4k upscaling.

Absoluetely the latest Nvidia driver is broken, thousands of reports about it.

The latest driver that is stable is 373.06 (heavy testers found this to be the best driver).

IF you have any driver more recent, you might experience some strange things, especially with Pascal GPUs.

[madshi]

No 3rd party software, no flux, fraps or gpu-z. No optimus.
Gtx1070 latest driver
Latest version of lavfilter,mpc-hc
Windows10 64bits V1607
Then please create a log file with the 91b test build I posted earlier.

[khanmein]

Yes, I was. :o

Totally forgot that it was on and it might affect it.

Cleared it out, tested, 230W max power pull on GPU, max OC clocks achieved.

Nice job!

The visuals are amazing, so much sharpness. At first I almost through why do we have an edge softening algorithm, now I get it :D

Now time to fix my setup and settings, because my actual settings add blur and over sharpening over certain edges... :o

Something I noticed recently, whenever I play a really low resolution video (under 720P) there is an audible feedback coming from the GPU, like a high pitched sound, which is not related to the fan. No idea what it is...

i also noticed the pitch noise from my GTX 970 when i play any vids with NGU (tested 360~720p)

[jkauff]

Are you saying that the option "use D3D11 for presentation" makes problems when using NGU, but makes no problems when using other image doublers?
Yes, that's exactly the case.

[khanmein]

For people reporting that install.bat doesn't work in Windows 10:

madshi knows. It's been reported multiple times already.

To work around this you just need to copy the files from the new archive into the folder of the older version, overwriting the old files. I *always* update madvr this way. Since this update method doesn't change where the madvr files are located, it is not necessary to run install.bat.

yeah i noticed the same thing. the solution if u run the install.bat is reboot your pc.

1) uninstall, reboot
2) reboot
3) install, done

[imhh11]

Then please create a log file with the 91b test build I posted earlier.
I'm not very familiar with debug mode. It says '' file "madVR.ax" not found '' but I can see madVR.ax in the folder.

Absoluetely the latest Nvidia driver is broken, thousands of reports about it.

The latest driver that is stable is 373.06 (heavy testers found this to be the best driver).

IF you have any driver more recent, you might experience some strange things, especially with Pascal GPUs.

I tried version 373.06, no luck , still not working.. :(

[HillieSan]

some quick tests:

Chroma & Luma doubling both @ NGU med: 32ms
Chroma & Luma doubling both @ NGU high: 79 ms
Chroma & Luma doubling both @ NNEDI32: 25 ms
Chroma & Luma doubling both @ NNEDI64: 36ms


I have a XFX RX 480 GTR Back Edition and I get worse performance.

Chroma & Luma doubling both @ NGU med: 50 ms
Chroma & Luma doubling both @ NGU high: 120 ms
Chroma & Luma doubling both @ NNEDI32: 40 ms
Chroma & Luma doubling both @ NNEDI64: 58 ms

MadVR 0.91b: With chroma upscaling Lanczos3 and with DX11. DX9 gets even worse.
My system is: A8-6500, 8 GB RAM, Windows 10 and latest AMD drivers

Any idea why this is? I mean this is one of the fastest RX 480 cards and I would expect similar or slightly better results. Sunset1982 are you using some settings that makes it faster? There is a ratio of approx. 1.5. What can explain this?

[madshi]

I'm not very familiar with debug mode. It says '' file "madVR.ax" not found '' but I can see madVR.ax in the folder.
Did you follow the installation instructions? See here:

http://forum.doom9.org/showpost.php?p=1786276&postcount=40282

[imhh11]

Did you follow the installation instructions? See here:

http://forum.doom9.org/showpost.php?p=1786276&postcount=40282

Sorry my english suck, I'm french or I'm too stupid haha,
How do I make sure I'm in release mode or what does it mean?

[huhn]

960 boost clock of 1430 mhz
480 -> FHD 23p
480 -> double -> spline 3 AR
chroma and luma doubling

NGU high: 41 ms 1430 mhz
NGU mid: 25 ms at ~1000 mhz
nnedi3 256: ~100 ms 1265 mhz
nnedi3 128: 48 ms 1265 mhz
nnedi3 64: 39 ms 1265 mhz

NGU high is notable slower than nnedi3 64
so hard to test this card... i guess openCL limited the GPU clock to 1265 mhz.

[madshi]

How do I make sure I'm in release mode or what does it mean?
There's a batch file "activate release mode.bat" or "activate debug mode.bat". Executing this batch switches madVR between release and debug mode. Basically the batch just renames the madVR *.ax files. There are 4 ax files, two for 32bit, two for 64bit. One of these are release build. some debug build. The debug build files create a log file on your desktop.

[madshi]

960 boost clock of 1430 mhz
480 -> FHD 23p
480 -> double -> spline 3 AR
chroma and luma doubling

NGU high: 41 ms 1430 mhz
NGU mid: 25 ms at ~1000 mhz
nnedi3 256: ~100 ms 1265 mhz
nnedi3 128: 48 ms 1265 mhz
nnedi3 64: 39 ms 1265 mhz

NGU high is notable slower than nnedi3 64
so hard to test this card... i guess openCL limited the GPU clock to 1265 mhz.
Ok, much better than with the RX480.

[Sunspark]

That's called coil whine (http://www.overclock.net/t/1094728/possible-fix-for-coil-whine), it's related to the quality of the design/parts in your graphics card. Best to avoid whatever manufacturer made that card in future.. you could try overclocking it, it may help.

The usual manufacturing way to deal with coil whine if nothing else changes is to put a dab of epoxy or something like that on the offending component if you know exactly which one is the one generating the sound. This is part of why you will sometimes see it in some power supplies.

Most people don't know about coil whine. As it can be audibly produced by wifi routers, my suspicion is that it is heard by some and this is part of why people freak out about wifi because they can tell there is "something" the closer they go to the unit emitting gamma rays but do not know it is only sound.

[aufkrawall]

It wasn't about coil whine yes or no, but about the specific tone. Like cnckane noted, NGU leads to quite different sounds than anything else in madVR, and this is not related to the fact that the load would be higher. NNEDI3 64 + SR3 has a higher load than NGU, but the sound frequency is not that high though.
I first thought it wouldn't be worth mentioning, but since cnckane describes exactly the same, it seems more than a just a pure coincidence. Something in the GPU/on the board is going on with NGU that seems to cause similar effects like high fps in games.

[imhh11]

There's a batch file "activate release mode.bat" or "activate debug mode.bat". Executing this batch switches madVR between release and debug mode. Basically the batch just renames the madVR *.ax files. There are 4 ax files, two for 32bit, two for 64bit. One of these are release build. some debug build. The debug build files create a log file on your desktop.

here we go. thank you for your time.

https://www.dropbox.com/s/n4ysa0zouo8i872/madVR%20-%20log.txt?dl=0

[madshi]

There's somebody hooking D3D9 on your PC. I'm not sure who or why. Do you have any funny NVidia extensions installed, like virtual desktop stuff, or maybe some remote desktop software running or something?

[aufkrawall]

If GeForce Experience is installed, I'd try uninstalling it or turning off its capture capabilities.