madVR - high quality video renderer (GPU assisted)

[robpdotcom]

According to Madshi:

With Blu-Ray the luma (brightness) information is stored in 1920x1080 pixels, however, the chroma (color) information is only stored in 960x540 pixels. So someone somewhere has to upscale those 960x540 chroma information to 1920x1080.

Also worth noting, as there seems to be some confusion:

If you display the source 1:1 in its original size, all luma scaling is turned off. The whole scaling options in the settings dialog have no effect in that situation. If you want to rescale the original resolution to something else, all the various filters have their own advantages and disadvantages, as I already said. E.g. Lanczos is the sharpest of the bunch, but also adds a lot of ringing to the image. Scaling is always a balance act between sharpness, aliasing and ringing. Just play with the filters and choose which looks best to your eyes...

[dansrfe]

dansrfe - I was under the impression that chroma has to be upsampled when expanding from 4:2:2 to 4:4:4.

Yes, that's true. Mistake on my part.

[JarrettH]

I've done a bit more testing, I still very much enjoy the softest chroma + sharpest luma combo...it gives the most stunning PQ to my eyes :)

upscaling in ffdshow: bicubic 0.0 for chroma, spline for luma

in mVR: softcubic100 for chroma, spline3 for luma up/down

I use CRT and DLP, no LCD..

I usually go between Mitchell (luma upscaling) + SoftCubic70 (Chroma upscaling) for sources that weren't encoded great to Catmull (luma upscaling) for nice sources, sharp sources. How can you even tell a difference changing chroma upscaling? It's much harder to see...

[leeperry]

How can you even tell a difference changing chroma upscaling? It's much harder to see...
I'm colorblind, so we do see colors differently...and I usually use a very colorful cartoon for those tests.

If I don't use the "softest chroma/sharpest luma" combo, I get the feeling that they interfere w/ one another. I'll try to come back w/ some screenshots ^^

PS: ah well, my Bronzés.mkv (http://www.sendspace.com/file/tiv885) red text on black background test pattern(that was used in the OP) clearly shows that softcubic100 is much softer than spline3...breaking news :D

http://thumbnails7.imagebam.com/10988/d8148a109878963.jpg (http://www.imagebam.com/image/d8148a109878963) http://thumbnails18.imagebam.com/10988/346f92109878970.jpg (http://www.imagebam.com/image/346f92109878970)

But what I'm seeing in that cartoon doesn't really seem to appear on random static screenshots...basically it kills all the chroma noise and allows the luma to look much clearer and sharper. Otherwise what I see is the chroma noise interfering w/ the luma, rejoicing into a mushy PQ when put in motion..maybe my color blindness condition amplifies this, but anyway I enjoy having a noise-free chroma together w/ a very sharp and clear luma \o/

[kutjong]

Hm, for some reason madVR drops frames like crazy as soon as I go full screen.
My source is 1080p x264 and my display is also at 1080p so I guess only chroma needs scaling in this situation. Funny thing is it doesn't matter which scaling algorithm I choose, madVR drops frames with every algorithm and it starts doing it as soon as I go fullscreen. In windowed mode playback is smooth.

The problem only seems to be with 1080p material. 720p doesn't drop frames and I can run with spline3 on everything without issues.

Does anybody else have this problem? :confused:
My guess is this has something to do with exclusive mode...

My system:
Intel i7 930
Radeon 5850 (Catalyst 10.11)
6 GB RAM
Windows 7 x64
MPC-HC 1.4.2745

[janos666]

My guess is this has something to do with exclusive mode...


Then try to disable it.
And check your V-sync (should be application controlled or off) and anti-alising and anistropic filtering modes (they should be application controlled or off --> a forced AA mode, like Morfological-AA with ATI cards, can kill madVR!)
And check your GPU and CPU usage during the full screen playback...

[Razoola]

Hm, for some reason madVR drops frames like crazy as soon as I go full screen.
My source is 1080p x264 and my display is also at 1080p so I guess only chroma needs scaling in this situation. Funny thing is it doesn't matter which scaling algorithm I choose, madVR drops frames with every algorithm and it starts doing it as soon as I go fullscreen. In windowed mode playback is smooth.

The problem only seems to be with 1080p material. 720p doesn't drop frames and I can run with spline3 on everything without issues.

Does anybody else have this problem? :confused:
My guess is this has something to do with exclusive mode...

My system:
Intel i7 930
Radeon 5850 (Catalyst 10.11)
6 GB RAM
Windows 7 x64
MPC-HC 1.4.2745

Yes I have had this problem where for some reason the resizers are slower in exclusive mode compared to windowed mode when the times should be identical speeds. You can confirm this by looking at the queue stats in both modes.

I have mentioned this to madshi and he was not sure why it was happening but suggested to wait until his exclusive mode code was complete.

[dansrfe]

Kind of weird but I had a dream that madshi had updated this thread with a new madVR. Only just a dream I guess. :(

[pankov]

Such dreams can be labeled as "wet dreams" ... and sadly after "puberty" they become more rare
;)

[leeperry]

Kind of weird but I had a dream that madshi had updated this thread with a new madVR. Only just a dream I guess. :(
give us 0.35 or the bunny gets it http://forum-images.hardware.fr/images/perso/4/le_phoque_pedoque.gif

[kutjong]

Then try to disable it.
And check your V-sync (should be application controlled or off) and anti-alising and anistropic filtering modes (they should be application controlled or off --> a forced AA mode, like Morfological-AA with ATI cards, can kill madVR!)
And check your GPU and CPU usage during the full screen playback...
Ah, I mistakenly had AA and AF enabled. After setting them to application preference there are no dropped frames! Such a noob mistake... :p
Thanks!

[Gser]

Ah, I mistakenly had AA and AF enabled. After setting them to application preference there are no dropped frames! Such a noob mistake... :p
Thanks!

I use Ati tray tool that initializes a "No AA & AF profile" every time I use MPC-HC and after its use, reverts to former settings. Easy as eating pie.

[leeperry]

How can you even tell a difference changing chroma upscaling? It's much harder to see...
Having more thoughts about it: we're doing a major upscale on chroma, which will result in a major amount of resizing artifacts...I believe it's encoded at half-res for the reason that the brain isn't too sensitive to chroma resolution? If you look at the two comparisons screenshots I've made, you can see that spline increases the chroma noise drastically...nothing more.

http://www.pixelz.fr/d/1/e/2df28c4170e763631644d79b84084.png

Better hide it than amplify it :p

[6233638]

What you suggest blurs the chroma though. It is best to upsample chroma so that it is as good a match to the luma information as possible to avoid it bleeding out of the lines or causing other problems.

If you want to eliminate chroma noise, upsampling chroma correctly and then adding noise reduction to that image should give much better results. (but it is best to leave the source as-is)

SoftCubic100 blurs chroma beyond the edges of luma causing colour to bleed out the edges considerably, resulting in desaturated colour.
Spline is not a good choice either as you say. But spline does not look good with anything to me. It looks too processed/artificial from the way it treats the image like it's made up of curves to try and avoid ringing.

Your test does not seem to be ideal as it looks like you are starting with a poor source where even the best chroma upsampling results in a very blurred image.

On proper chroma tests or with better quality sources, Bicubic75 seems to do the best job of upsampling chroma information so that it is a close match to the luma information. Then choose your luma upsampling algorithm depending on how you want the image to look.

For Luma, SoftCubic50 seems truest to the source without adding ringing/sharpening or removing detail/grain.

[jmone]

FYI there a a group of us using JR Media Center pushing for the abilility to use madVR as the Video Renderer but it looks like it is not going to happen as one of the devs over there have quoted:

http://yabb.jriver.com/interact/index.php?topic=60810.0
Unfortunately madVR is too inflexible - it does not support all video cards, does not support DVD playback, and, more importantly for us, does not work when we using our way of building a DirectShow graph. I am not convinced yet that I should overhaul our video playback engine in order to support the use of madVR. Sorry, maybe some time in the future.

I don't know if madshi is interested in following up with the JR Folk on this.

Thanks
Nathan

[JarrettH]

thanks for the sample leeperry :cool:

[namaiki]

Anyone tried DVDs with MadVR on Windows Vista or preferably Windows XP?

edit: I copied qdvd.dll from my Windows XP desktop PC over to my laptop's Windows 7 installation, and this seems to allow DVDs to be played with the menu. Only thing is Ctrl+J is reporting 'movie unknown fps'.

edit2: movie unknown fps was being reported because I was using MPC-HC's internal MPEG-2 decoder with deinterlacing.

Update: doing this kills Overlay Mixer....

[dansrfe]

Who cares about Overlay Mixer in the first place?

[THX-UltraII]

Some questions:

Is the madVR project dead? It s been a while since a new version came out so I was just wondering this.

Also, there is a discussion on a forum in my country about standalone BRplayers like the Oppo or Dune player <-> HTPCs

If a HTPC is properly configured with the use of MPC-HC, madVR renderer and FFdshow as codec, will this beat a high-end standalone player?

I get one frame drop every 8-16 ours (calculated by madVR, checked with CTRL+J) with my HTPC and the PQ looks superb to me so how could a standalone player do this better.

One last question:
Am I correct that Fullscreen Autochange refresh rate function in MPC-HC does not work when using madVR as renderer?

[pirlouy]

Who cares about Overlay Mixer in the first place?
Before madVR exclusive mode, it was the better renderer.

Is the madVR project dead? It s been a while since a new version came out so I was just wondering this.
No. Search for madshi's posts.

[THX-UltraII]

any thoughts/input about my other questions?

[dansrfe]

Some questions:

Is the madVR project dead? It s been a while since a new version came out so I was just wondering this.

Also, there is a discussion on a forum in my country about standalone BRplayers like the Oppo or Dune player <-> HTPCs

If a HTPC is properly configured with the use of MPC-HC, madVR renderer and FFdshow as codec, will this beat a high-end standalone player?

I get one frame drop every 8-16 ours (calculated by madVR, checked with CTRL+J) with my HTPC and the PQ looks superb to me so how could a standalone player do this better.

One last question:
Am I correct that Fullscreen Autochange refresh rate function in MPC-HC does not work when using madVR as renderer?

No.
HTPC.
Yes.
No.
Yes.

Standalone players suck, it's only for convenience that people use it. When a BluRay player uses Java to playback a BluRay the argument of which is better ends there. My HTPC is full on correctly configured with bitstreaming all formats to my HD receiver, ffdshow decoding everything and sending madVR 4:2:2, all HQ in madVR and Spline @ 3 taps on all, HDTV calibrated using built-in methods, graphics options, and a calibration system, madVR gives me perfect sync, no teraing etc etc.

[robpdotcom]

Am I correct that Fullscreen Autochange refresh rate function in MPC-HC does not work when using madVR as renderer?

Yes, and no. It works if it is also changing the resolution. I keep my resolution set at 1280x720 (it makes the text larger, which is nice since I'm on my sofa across the room) and it switches to 1920x1080 for watching movies. Forcing the resolution change seems to trigger the refresh rate change.

It also worked in the older madVR versions, before v0.28 I think. I'm pretty sure the main difference between the current version and v0.27 is the madHcCtrl, which allows you to control madVR (and MPC-HC) from a different PC on your LAN.

[webs0r]

A quick workaround for the refresh rate change issue is to use reclock's event notification script and a command line refresh rate changer.

[Thunderbolt8]

a question regarding bitdepth of LCD TVs: I came across some TVs which have 8-bit and some with 10-bit bitdepth. now I wonder is that actually enough to display video put out by madVR without any loss of quality?

[6233638]

Is the madVR project dead? It s been a while since a new version came out so I was just wondering this.I don't think so. A while back, Madshi didn't release anything for months and then it started back up again with frequent releases. In its current state there's not much to complain about. Video playback is stable now, performance is good.

Also, there is a discussion on a forum in my country about standalone BRplayers like the Oppo or Dune player <-> HTPCs

If a HTPC is properly configured with the use of MPC-HC, madVR renderer and FFdshow as codec, will this beat a high-end standalone player?madVR vs standalone, the only advantage a high-end standalone has is that madVR is current limited to 8-bit output. Exclusive puts some video cards into 10-bit mode but I think its still 8-bit data from madVR.

Its possible that the high-end Sony players with their 14-bit reconstruction could possibly match madVR. Anything else, madVR wins without a doubt.

HTPC vs Standalone is no contest though. madVR + ReClock beats anything because you can play back any video file at the original 24.0 fps. 23.976 is only used as a compromise so 3:2 can be used for 59.94 compatibility, the original speed is 24.0fps.

ReClock also upsamples audio to 32-bit/192kHz.
25fps content can be slowed down to the original 24fps source. This is mostly a DVD issue but some Blu-ray is 1080i50 from a 24p source still. No standalone can do this.

With HTPC there is no region locking and you can just start up the film .m2ts file off the disc directly instead of forced trailers, menus and other crap.

Am I correct that Fullscreen Autochange refresh rate function in MPC-HC does not work when using madVR as renderer?Doesnt seem to work for me. I think Madshi said that the next madVR release will have custom refresh rate selection.

[dansrfe]

madVR vs standalone, the only advantage a high-end standalone has is that madVR is current limited to 8-bit output. Exclusive puts some video cards into 10-bit mode but I think its still 8-bit data from madVR.

Its possible that the high-end Sony players with their 14-bit reconstruction could possibly match madVR. Anything else, madVR wins without a doubt.


Are you effing crazy??? madshi spent countless MONTHS perfecting this renderer to output in 16-bit all in the name of quality. 16-bit output is default in madVR then you can change that to 10-bit if your computer is too slow to output 16-bit.

[pankov]

dansfre,
I think you are wrong.
All the processing in madVR is done in 16-bit (or 10-bit if change the settings to decrease the processing load) but the actual output is done in dithered 8-bits.
There was a discussion if it's worth even talking about 10-bit vs. dithered 8-bits and madshi's position is it's not ... at least not now when most sources are still in 8-bits and most displays/video cards don't support 10 or more bit's output.

[oddball]

I'd still go with a 10 bit display purely based on it uses less dithering and potentially has less banding issues.

[Razoola]

I don't think so. A while back, Madshi didn't release anything for months and then it started back up again with frequent releases. In its current state there's not much to complain about. Video playback is stable now, performance is good.

Im not so sure, the latest version (034) has problems with video freezing, many of us have had it happen. For stable you need to go back to 032 or use the 034 debug build.

[ryrynz]

madVR vs standalone, the only advantage a high-end standalone has is that madVR is current limited to 8-bit output. Exclusive puts some video cards into 10-bit mode but I think its still 8-bit data from madVR.



features:

- full 16bit processing queue
- final 16bit processing result is dithered down to RGB output bitdepth
- bypasses graphics card's video (damage) algorithms


Fail.

[pankov]

ryrynz,
may be you should learn to read before saying something like this.
As I said a few posts before and as madshi did many times before
madVR outputs at 8-bit RBG (24bits in total)
do a search for "8-bit" in this thread and look for madshi's posts.

[dansrfe]

ryrynz,
may be you should learn to read before saying something like this.
As I said a few posts before and as madshi did many times before
madVR outputs at 8-bit RBG (24bits in total)
do a search for "8-bit" in this thread and look for madshi's posts.

Yes, I second that. Also in correction to my previous post, although madVR doesn't output 16-bit, it surely processes in it which I doubt any standalone player does.

[scoz]

Hey does anybody know where to get old versions of madVR? I tried looking for a link I thought I remember madshi posting but I couldn't find it. Anybody know the link or if theres a zip somewhere with them all?

[leeperry]

where to get old versions of madVR?http://www.videohelp.com/tools/madVR/old-versions#download

[Thunderbolt8]

ryrynz,
may be you should learn to read before saying something like this.
As I said a few posts before and as madshi did many times before
madVR outputs at 8-bit RBG (24bits in total)
do a search for "8-bit" in this thread and look for madshi's posts.how can I find out if my graphic card can output RBG > 8-bit?

[dansrfe]

^ First find out if your screen can output more than 8-bit, then look at the graphics card.

[janos666]

It's an interesting question. I recently bought an used FirePro v5700 to test my display and play with 10 bit/color output.
Photoshop works fine after I enabled the "10-bit pixel format support" in CCC (restricted workstation tab). The 10-bit test gradient looks fine and CMS corrected 8-bit images are relatively "banding-free". It's a really nice feature! :)

But I think this FirePro/Quadro specific feature applies only for some kind of "windowed 10-bit" mode. (PS works in windowed mode!) I think this workstation feature has nothing to do with a D3D fullscreen renderer. I think this "expensive" feature was developed for PhotoShop and similar softwares (but I couldn't find anything else yet). This is the answer for the question: "Why do they advertise their Radeon/Geforce cards with DeepColor support if 10 bit/color is restricted to FirePro/Qudro cards only?" Well..., because this so called workstation 10-bit support is a software specific hack for windowed 10-bit mode! You can use it with softwares where the software developers didn't want to make a complete fullscreen user interface (it would be expensive and I think you don't want to run PhotoShop in Fullscreen mode anyway, it would be annoying! So, you pay for the extra workstation feature and everything is cool... :D).

There a is conversation about the 10-bit mode in the MPC-HC thread. New MPC-HC builds can't produce real 10-bit display mode (it doesn't matter than I have a FirePro and PS works fine). The funny thing is, that old MPC-HC builds are able to produce 10 bit /color output with a Radeon/Graforce card too! (The only problem is that this old MPC-HC build doesn't have the color management feature. So, it's useless for me.)

I think we don't need a professional VGA! But yes, you need a 10 bit/color capable display (and a VGA with DP connection, I think, but it theoretically works with dual-link DVI too).
(You can find some affordable displays nowadays: I think the Dell U2410 is the cheapest one. The next 24" model is a professional NEC display, but I think that one is also "affordable" if you find a refurbished one. I would buy one yesterday but they didn't ship it to Hungary. :D)

[Thunderbolt8]

^ First find out if your screen can output more than 8-bit, then look at the graphics card.
I doubt the current one can, but since my screen might be rather subject to change than my graphic card in the near future, I'd rather want it to know the other way round.

[6233638]

Yes, I second that. Also in correction to my previous post, although madVR doesn't output 16-bit, it surely processes in it which I doubt any standalone player does.You are correct, no standalone processes in 16-bit. The Pioneer Elite players with "16-bit" output process in 10-bit and upsample output to 16-bit. (which is simply input value x 64 = output value)

I think the Sony players are straight 14-bit but do not know for sure. If there is and player that could compete with madVR, these are the only ones. A HTPC setup with madVR + Reclock is better in my opinion though. The only trouble is that HD audio can be tricky if you want instant playback off a disc. I think the only way to get HD Audio is to rip with eac3to?

MadVR's 16-bit processing is amazing, and dithering to 8-bit output is a definite improvement, but outputting true 10-bit would be even better, especially because its being processed in 16-bit.

[hdboy]

You can test with the files from this post (http://forum.doom9.org/showthread.php?p=1230259#post1230259).

On a TV (Sony LCD), i'm NOT supposed to see the black bars below 16, right?

[fairchild]

On a TV (Sony LCD), i'm NOT supposed to see the black bars below 16, right?

Correct, if you see black bars lower than 16, you have to lower your brightness. It's alright to see white bars past 235 though. That just means your setup is passing WTW and BTB.

[dansrfe]

If you are passing RGB 4:4:4 PC Levels then it should show black at 0 and white at 255 like it is supposed to be. LCD is meant to show the full range. Use it.

[kostik]

I have an interesting question.
I have a pioneer 428XD plasma. When I check GPU SCALING in Catalyst panel, my tv shows that it outputs 36bit!!! When I dont check the GPU scaling option I get 30bit output.
Does it mean that my tv and my radeon 5750 are both capable of outputting 12bit and 10bit color depth?

Thanks!

[nlnl]

^ First find out if your screen can output more than 8-bit, then look at the graphics card.


My card is 5470 + W7 and my Pio Kuro 5090 always reports 30bit deepcolor input even for desktop!.

[kostik]

My card is 5470 + W7 and my Pio Kuro 5090 always reports 30bit deepcolor input even for desktop!.

If you check the gpu scaling option you also get 36 bit like I have reported in my previous reply?

[nlnl]

If you check the gpu scaling option you also get 36 bit like I have reported in my previous reply?

I can try only on Sunday

[janos666]

My card is 5470 + W7 and my Pio Kuro 5090 always reports 30bit deepcolor input even for desktop!.

Because the post-Rv770 Radeons and FirePros (like the Evergreen based products) always output 10 bit/color from the VGA LUT (or dithered 8 bit when the display can't receive 10 bit). But this is only the calibration LUT. If it's linear, then there is no any benefit. If you calibrate your display through this LUT, then it's a good thing, but not enough for everything (other color management things).
But 10-bit from the framebuffers... that's another story (and it looks enough for me). I am absolutely sure that Aero (MS Windows desktop) never works in 10 bit/color mode. The only softwares I know: PhotoShop (restricted to Qudro/FirePro cards) and MPC-HC 1.3.1824 (theoretically works with "everything" [consumer level VGAs, dl-DVI connection, etc.], but you need compatible drivers: 8.8(system driver version)<YOURS<10.5a(catalyst package version)

[Peekstra]

I read that the Dell U2410 accepts 10 bits and uses it to dither the output to 8 bits so reporting this capability doesn't ensure you actually get the 10bits output.

So in the case of a monitor only accepting and dithering the video to 8 bits you may actually degrade picture quality if you leave the dithering to your monitor compared to madVR's high quality dithering.

[janos666]

The panel itself dithers the 10-bit input from the dual-lvds interface. Yes, but given the fact that it has a factory calibrated internal LUT, it always happens. I can't turn off the internal dithering, I have this noise anyway, so "I want to use it well".
This internal dithering causes weaker noise than madVR (BUT I can see madVR's noise on test patterns only, so... just for the record...). -> And this let me think that these noise factors exaggerates each other (different types of dithering modes in the same line... but this is only a theory of mine).
This internal dithering alone is week enough while it feels enough. So, I want to try to turn the software dithering off and let the internal dithering to work with the 10-bit input.

My only real-life experience with this internally dithered 10 bit/color is the high quality color management in PS.
I made a very big XYZ cLUT from ~2100 spectral measurements from the display in native mode (OSD RGB Gains at 255-255-255, linear VGA LUT, native gamut mode, etc) and used it in PS with different rendering intents.
I achieved a gray gradient which I could never see on any displays which doesn't work with 10+ bit rewritable hardware LUTs (like an EIZO CG). For the fist look, it feels almost perfect. (Of course, it's not as good as a CG if you take numerous closer looks but it's good enough for any non-professional usage.) And the dithering noise is relatively weak.
To see the contrast: This gradient really looks awful with 8 bit/color output. So, at the first time, it feels like you found the holly grail of displays! :) (And you forget it later, when you can't use it in real life because you don't work with PS. :D)

By the way, I would like to upgrade to a native 10-bit p-IPS panel based product when I will be able to benefit from it (and somebody in my small country will try to sell an used NEC PA231w for "an affordable price". - May be in the next half year. :D)