madVR - high quality video renderer (GPU assisted)

[patul]

I experience noticeable frame drops in full screen mode even with OSD off (which I didn't see when I used version 0.9) when playing MKV (DVD rip) with MPC HC. OSD confirms it too..

CPU: Core 2 Duo E4400 @ 2GHz
GFX: NVidia Geforce 7300 GT
Win XP SP3.

[foxyshadis]

On madVR 0.10, it only gets somewhat smoothy moving effect with 1280x720 output window size. If you are interested in taking a look why madVR 0.10 does not work well with it, here is the bt link of this video. It is somehwat big -- 410MB.

.....

The playback is by MPC-HC build 1081 + CoreAVC 1.9.5.0

I realize you're trying to help, but this is rule 6 material. Illegal copywrited downloads are absolutely not allowed here (that includes anime not distributed here), see the rules link at the top of every page. Everyone gets at least one warning, don't worry.

[AJ73]

You can find non copyright protected HD scenes here: http://www.w6rz.net/
By the way there are also MPEG-2 samples with 4:2:2.

[pie1394]

You can find non copyright protected HD scenes here: http://www.w6rz.net/
By the way there are also MPEG-2 samples with 4:2:2.

Most video clips in the above link seem related to video resolution, quality, and color accuracy.

Is there any non copyright protected contents in 23.976 / 29.97 / 59.94 fps which contain the smoothy scrolling texts or scene motion? Once those contents are checked and found with some issues, they should have higher chance to be checked again by more people or madshi without the violation of forum rules.

[wayland]

maybe try that bigbuckbunny? that has a creativecommons licence and is available in various formats

[cyberbeing]

Non-copyrighted content is not really the point. Doom9 doesn't promote illegal downloads. In other words, not illegally downloading copyrighted content, which you can find on the internet is the problem.

As long as you have bought the content, posting a small sample of a problem area is usually accepted as OK here on Doom9. For example, if pie1394 owns a blu-ray that has nice smooth scrolling text or panning scenes, he could rip it, trim out or encode a small sample from the m2ts of the problem area, and post it here.

[Thunderbolt8]

when I use yadif deinterlacer and checkmark the double framerate option, then the delayed frames number increases constantly at playback (renderer queue is at 5-7/8 most of the time then). after some time it becomes a little slower and stops then for a certan time, then it begins to increase again (and it repeats). is all this intended to happen this way, or is it somewhat buggy? (tested with 23.976fps x264 .mkv file)

[Mark_A_W]

Most video clips in the above link seem related to video resolution, quality, and color accuracy.

Is there any non copyright protected contents in 23.976 / 29.97 / 59.94 fps which contain the smoothy scrolling texts or scene motion? Once those contents are checked and found with some issues, they should have higher chance to be checked again by more people or madshi without the violation of forum rules.

I've just been staring at scrolling credits....for ages on end.

It doesn't really matter what movie you use, as long as it has scrolling credits, and comes from a clean source (like bluray, but NOT a recorded TS file as they are glitchy).

[Neeto]

I've been researching building by new HTPC and came across this thread a while back and having been following along.
I look forward to a stable madVR appearing - great work :thanks:

I'm seeking guidance on the mimimum ATI graphics card that "should" work with the stable madVR. I know there is a bit of "looking in the crystal ball" about this, but would be interested in peoples thoughts before starting out - hopefully others might find it useful.

New HTPC will most likely be a ATI 4650 series card.

There are some comments earlier in the thread about cards not having enough GPU grunt, particularly when doing upscaling from low resolution, interlaced stuff e.g 480i to 1920x1080p.
I'm assuming 4650 will be fine, but still interested in the mimimum for retrofitting madVR into the 4 other old HTPCs I have.

I'd also like to clarify if the plan is to have de-interlacing, EE, de-noise, etc still work on the ATI cards once madVR is complete. I think the answer is yet, but....

Thanks, neeto

[Thunderbolt8]

I had a 7600GT until recently, and it was a bit too slow. so perhaps something like 1 year younger than that or something similar could already work.

[6233638]

0.10 seems like a great release so far. Perfectly smooth playback and no dropped frames using my onboard 9400 and the default settings. I was ready to upgrade the graphics card but it looks like it may not be necessary. :D

[Jong]

I get really good smooth playback with typically 16/16, 7-8/16, 5-6/16 buffers full and then with no warning (no visible drop in buffers) frames will occasionally be dropped or delayed. I realise this is the first version attempting smooth playback, so no biggie, but I thought it was worth feeding back that the buffer report in the OSD does not seem as helpful as you might expect in predicting if/when frames will be dropped or delayed.

[cyberbeing]

With 0.10, what have seen is that if the decode and render buffers/queues are not completely full, madVR has the potential to drop/delay frames. This is more along the lines that madVR is always pushing the GPU exremely hard to keep the buffers/queues full. If the gpu isn't fast enough to keep the buffers/queues full, playback suffers. This is just an educated guess, but maybe madshi will say this shouldn't be the case?

madshi, if it is the case, you may want to look into dynamically limiting the maximum buffer/queue size to what the GPU can always keep full. I wouldn't be surprised if that fixed the delayed/dropped frame problem I have occasionally run into when the render queue drops to 7/8.

[Thunderbolt8]

could that problem of dropped and delayed frames have something to do with doubled frames of those 59,94fps mpeg2 files? I noticed when I tried to use yadif double framerate with 23.976fps stuff, then even my CPU & new HD 4770 seems to be too weak and I get the same amount of delay piling up. perhaps its the same with those 720p files that madVR would suddenly have double the amount of frames to render in the same time and cannot really handle it.
but then again I dont know why this problem with those files only appears in fullscreen, but not in window mode.

[ajp_anton]

What kind of resize is most demanding, maximum upscale, maximum downscale, or some "odd" non-integer resizes?

[stpdrgstr]

I'm getting instant freezes here with 0.10, whenever I try to open anything, MPCHC freezes.

With 0.9, it doesn't happen.

I have an ATI HD4870 512 mb. Any ideas?

[KoD]

I'm getting instant freezes here with 0.10, whenever I try to open anything, MPCHC freezes.

With 0.9, it doesn't happen.

I have an ATI HD4870 512 mb. Any ideas?

If you wait something like 20 - 30 seconds, you'll see that playback starts. I have the same issue with ZoomPlayer. madshi says that the renderer is waiting to determine the refresh rate, and doesn't start playback until then. I think that's a bad design idea. It could be something else, like threading synchronization. But debugging that is hell since you usually can't easily reproduce it.

[STARSBG]

@madshi

When you add an option for subtitles?

[Mark_A_W]

@madshi

When you add an option for subtitles?

Subtitles work fine with VSFilter/DirectVOBSUB v2.39 installed.



(Oh, I found out why my subs weren't working originally - I'd installed AutoGK, and it includes a really old version of DirectVOBSUB..wrecking my subtitle playback.)

[lunkens]

Hello!

MadVR 0.9 worked for me but 0.10 only works in "windowed mode" because if i try to go to "fullscreen mode" i first get similar green screen as Mark0077 but when i exit MPCHC i get a system wide screen error and it gets worst for every second until i reboot but before windows shuts down it unloads Nvidia drivers and then the graphic errors go away.


MPCHC: 1.2.1111.0 x86
ffdshow rev.2895 x86
Microsoft Windows Vista SP1 X64

anyone else got system wide graphics errors?


Madshi: a direct question to you... will there be a x64 version?

[stpdrgstr]

If you wait something like 20 - 30 seconds, you'll see that playback starts. I have the same issue with ZoomPlayer. madshi says that the renderer is waiting to determine the refresh rate, and doesn't start playback until then. I think that's a bad design idea. It could be something else, like threading synchronization. But debugging that is hell since you usually can't easily reproduce it.

I'm waiting since like 1 minute and the Player keeps freezed. I don't think it's working at all :(

[Blight]

KoD:
I concur, there is no reason to delay playback until the refresh rate is properly determined. At least it should be optional as most users would rather have playback start as soon as possible and take these few seconds to detect the refresh rate while the video is already playing.

[cyberbeing]

My question would be why does it take so long for the refresh rate to be detected on some of the systems you guys have?

On my 4 year old system, madVR takes no more than 1 second longer then other renderers to load (opening 1080p with madVR using a 3dlut 3 seconds, other renderers 2 seconds). I would assume this 1 extra second is normal, and there is something seriously wrong when people have to wait 10 or more seconds. Since I'm sure none of you would be complaining if it was only a 1 second difference like I see, madshi figuring out what is wrong and speeding up refresh rate detection for those people might be the better solution.

For reference my system is as follows:
AMD X2 4800+ @2.64Ghz
2GB DDR400 @DDR433 2-3-3-6
7800GTX 512
Sony GDM-F520 (CRT)
WinXP SP3 x86

[Mark_A_W]

Yep, about one second for me too.


I'm now using MadVR as my primary renderer in the HT, with Reclock running it looks good, and it's SMOOTH.

It's MAD MadCat!

http://recluse.me/uploaded_images/Inspector_Gadget14-786488.jpg

[cyberbeing]

madshi, since it's the weekend now, any chance of getting a new build with at least a fix for luma scaling not being remembered?

Second thing on my priority list is a fix for all the lock-ups that 0.10 is prone to.

Third, would it be possible to add achieved frame rate to the statistics? The source filter you use for determining the media framerate is completely useless if Reclock is being used or the file is VFR.

[nijiko]

Me here. 0.10 can't load settings from setting file.
I try to delete the older 0.9 clear and use again 0.10, problem still.
When I change settings, the settings are already been saving in setting file.
But next loading. madVR 0.10 not to use the settings in file, instead to use default.

[Jong]

On my 4 year old system, madVR takes no more than 1 second longer then other renderers to load (opening 1080p with madVR using a 3dlut 3 seconds, other renderers 2 seconds). I would assume this 1 extra second is normal I was surprised. I was expecting a noticeable delay, but with my system and 3Dlut enabled I have not noticed any change. I am sure it is a little slower and I haven't sat there with a stopwatch, but honestly, it's not big enough to be any bother at all.

[Chumbo]

...I'm now using MadVR ...
It's madVR. Get it right or people will be knocking on your door.

[ikarad]

There is a very big problem (I think it's a bad limitation of beta progress) with MadVR. With VMR9 or EVR color profile (.icm) works but with madvr color profile doesn't work (it's like if overlay is selected all the time).

I could never use madvr if this problem is not corrected because by default my monitors is not calibrated and default colors are bad. Even with monitor calibration wizard it doesn't work. It's like power dvd with blu-ray

I expect that this problem was corrected in a next release

[honai]

It's not supposed to "work". If it were active then Vista's color mixer would apply all sorts of post-processing to the movie image, which we don't want. It's really only useful for still images.

[ikarad]

It's not supposed to "work". If it were active then Vista's color mixer would apply all sorts of post-processing to the movie image, which we don't want. It's really only useful for still images.

No, It works with powerdvd (except with blu-ray), mpc-HC, wmp11 with xp or vista and there is not problem even with hardware acceleration.

use of color profile and use of post processing is not related and has never been related.

[leeperry]

ah, you mean the three 1D graphic card's CLUT is not taken in account?
the only renderer that doesn't care about it is Overlay indeed.

I also hope this will be worked out, as cr3dlut's tables can't embed the CLUT at this point....plus the CLUT actually works in 10-bit!

[honai]

No, It works with powerdvd (except with blu-ray), mpc-HC, wmp11 with xp or vista and there is not problem even with hardware acceleration.

use of color profile and use of post processing is not related and has never been related.

That is 100% untrue.

The color management of Vista (can't speak of XP) compensates for the display, not for the graphics card. In other words, the sole reason for color management in Vista is because some people have displays with primaries so off that Vista needs to adjust the image to compensate for that defect. On a quality display (Eizo TFTs for desktop, Pioneer/Panasonic TV sets) the default settings are Good Enough (TM), and Vista's color management will do more harm than good to the image, especially since it does not operate in 32/64-bit as madVR does. That's the reason why madVR ignores all of Vista's color management and writes directly to the GPU.

Also, if madVR weren't doing that but instead use Vista's VMR9/EVR interface, in addition to the color management the GPU's "optimizations" for VMR9/EVR would kick in, i.e. colors shifted (especially flesh tones, which most GPU drivers tend to boost), unsharp masking, etc.

So in the end, for the little benefit of ICM color management you'd also get the distortions of GPU-based post-processing. You can't have the one without the other.

[lunkens]

Hello!

MadVR 0.9 worked for me but 0.10 only works in "windowed mode" because if i try to go to "fullscreen mode" i first get similar green screen as Mark0077 but when i exit MPCHC i get a system wide screen error and it gets worst for every second until i reboot but before windows shuts down it unloads Nvidia drivers and then the graphic errors go away.


MPCHC: 1.2.1111.0 x86
ffdshow rev.2895 x86
Microsoft Windows Vista SP1 X64

anyone else got system wide graphics errors?


Madshi: a direct question to you... will there be a x64 version?



follow up:

i tested it on my computer screen and i can with out a problem play the movie at 1600x1200@60Hz in fullscreen and there is no problem at all....

VGA (computer screen)
DVI (projector)
HDMI (nothing)

what could this be?

[honai]

GPU gets too hot. Overclocked? Or PSU can't handle the GPU load. Or power issues on the mainboard.

Definitely sounds hardware-related.

[ikarad]

ah, you mean the three 1D graphic card's CLUT is not taken in account?
the only renderer that doesn't care about it is Overlay indeed.

I also hope this will be worked out, as cr3dlut's tables can't embed the CLUT at this point....plus the CLUT actually works in 10-bit!

I don't know but the problem is that color profile created by color calibrator (for example lacie blu eye pro) is not used with mad vr. It's like if I select overlay under mpc-Hc.

It's a very big problem and this problem is the same under xp or vista.

I expect that Madshi allow to use color profile created by calibrator if not Madvr is unfortunately useless. And like the large majority (90-95 %) of Monitors (lcd and CRT) are not calibrated and often display bad colors

[leeperry]

So in the end, for the little benefit of ICM color management you'd also get the distortions of GPU-based post-processing. You can't have the one without the other.
I believe what the man is saying is that a LUT created w/ the ARGYLLCMS package is not taken in account by mVR ?! I don't use mVR atm so I can't really tell.

but anyway, yeah my CRT is way off...so I use a CLUT in the graphic card to get it to D65/2.2, then I do gamut conversion on top of it(HR does let the CLUT through, only Overlay doesn't).

a simple test is to mess w/ the brightness/contrast in pstrip/graphic drivers, mVR should still reflect these changes..or cr3dlut should allow to import the data from ARGYLLCMS...but cr3dlut works in 8bit, the graphic card's CLUT is 10 bit solid(you can measure the CLUT accurary w/ ARGYLLCMS).

[leeperry]

the problem is that color profile created by color calibrator (for example lacie blu eye pro) is not used with mad vr.
I don't know if you mean strictly .ICM profile or LUT files.

HR doesn't care for .ICM files because it's not color managed..but it does care for LUT files in ARGYLLCMS(actually all the renderers do, except Overlay)

anyway, ARGYLLCMS is far more accurate than any other color calibration package, especially in "HQ" mode :)

[honai]

Madshi already wrote that in the future he'll implement those controls (brightness/contrast/etc) directly into madVR, so no need to rely on PowerStrip or GPU drivers to do it.

And I already wrote, once you allow the GPU driver to mess with the picture (which is always the case with VRM9/EVR) you'll get all of the inherent post-processing that is invisible in the driver settings and impossible to turn off.

[leeperry]

Madshi already wrote that in the future he'll implement those controls (brightness/contrast/etc) directly into madVR, so no need to rely on PowerStrip or GPU drivers to do it.
nah, I think we don't understand each other.

I'm not talking about simple .ICM files, I'm talking about calibration LUT files...that are loaded in the graphic card's CLUT.

like the .cal files in ARGYLLCMS :

NUMBER_OF_SETS 256
BEGIN_DATA
0.0000 0.084634 0.0000 0.014302
3.9216e-003 0.087533 0.0000 0.018032
7.8431e-003 0.090470 4.8878e-004 0.021810
0.011765 0.093444 7.9362e-003 0.025639
0.015686 0.096457 0.015668 0.029520
0.019608 0.099509 0.023704 0.033453
0.023529 0.10260 0.032161 0.037440
0.027451 0.10573 0.041201 0.041482
0.031373 0.10891 0.050834 0.045580
0.035294 0.11213 0.060912 0.049735
0.039216 0.11539 0.071064 0.053957
0.043137 0.11870 0.080893 0.058254
0.047059 0.12207 0.090109 0.062628
0.050980 0.12548 0.098656 0.067077
0.054902 0.12894 0.10662 0.071598
0.058824 0.13245 0.11408 0.076191
0.062745 0.13601 0.12113 0.080850
0.066667 0.13962 0.12782 0.085563
0.070588 0.14326 0.13422 0.090308
0.074510 0.14693 0.14032 0.095068

these are processed in 10bit by the graphic card, and allow a display that has no colorimetry option whatsoever(CRT/LCD typically) to meet D65/2.2

all the PC video renderers let it affect the picture(except Overlay), this would be most welcome that mVR does it too(if that's not already the case)

when you play w/ the contrast/brightness in pstrip, it also modifies that graphic card CLUT.

[honai]

I see the point. But that is totally unrelated to the VMR/EVR and ICM/color management discussion.

If I recall correctly, madshi will implement that in a future version.

[leeperry]

well it's very much related actually...most ppl need both a 3D LUT to convert gamuts, together w/ a LUT in the 10bit graphic card's CLUT to calibrate their display to D65/2.2. one cannot go w/o the other.

ok well, if it's on the TODO list...perfect :)

[honai]

Yes, but the LUT is not related to the color correction of Vista via ICM files, and is not related to color correction in VMR9/EVR. If the graphics card has a LUT it's active independently of the OS's color management. What ikarad wrote, namely that the color profile usage (from ICM files) should be enabled, has nothing to do with the graphics card LUT. Color correction via ICM files is done in software in Windows, not on the GPU. And I suspect that the color correction algorithms of Windows are as bad as their previous generation of sound mixing algorithms. ;)

[Grmpf]

nah, I think we don't understand each other.

I'm not talking about simple .ICM files, I'm talking about calibration LUT files...that are loaded in the graphic card's CLUT.

like the .cal files in ARGYLLCMS :

NUMBER_OF_SETS 256
BEGIN_DATA
0.0000 0.084634 0.0000 0.014302
3.9216e-003 0.087533 0.0000 0.018032
7.8431e-003 0.090470 4.8878e-004 0.021810
0.011765 0.093444 7.9362e-003 0.025639
0.015686 0.096457 0.015668 0.029520
0.019608 0.099509 0.023704 0.033453
0.023529 0.10260 0.032161 0.037440
0.027451 0.10573 0.041201 0.041482
0.031373 0.10891 0.050834 0.045580
0.035294 0.11213 0.060912 0.049735
0.039216 0.11539 0.071064 0.053957
0.043137 0.11870 0.080893 0.058254
0.047059 0.12207 0.090109 0.062628
0.050980 0.12548 0.098656 0.067077
0.054902 0.12894 0.10662 0.071598
0.058824 0.13245 0.11408 0.076191
0.062745 0.13601 0.12113 0.080850
0.066667 0.13962 0.12782 0.085563
0.070588 0.14326 0.13422 0.090308
0.074510 0.14693 0.14032 0.095068

these are processed in 10bit by the graphic card, and allow a display that has no colorimetry option whatsoever(CRT/LCD typically) to meet D65/2.2

all the PC video renderers let it affect the picture(except Overlay), this would be most welcome that mVR does it too(if that's not already the case)

when you play w/ the contrast/brightness in pstrip, it also modifies that graphic card CLUT.

Isn't that a "problem" of cr3dlut ? Are those 3x 1D LUTs not a subset of one 3D LUT ? I thought you just need 1 3D LUT to cover it all in madVR, let cr3dlut intigrate not only 3 primaries into the YCbCr to RGB convertion, but instead 3x 1D LUTs, or another 3D LUT (17x17x17 like professionals do, or smaller ones) for calibration reasons. Am i wrong that 1 3D LUT can't handle it all ?

EDIT: Just for the Microsoft ICM thing - there is a reason why everyone is using "Adobe(ACE)" in Photoshop and not the "Microsoft ICM"...

[lunkens]

follow up:

i tested it on my computer screen and i can with out a problem play the movie at 1600x1200@60Hz in fullscreen and there is no problem at all....

VGA (computer screen)
DVI (projector)
HDMI (nothing)

what could this be?


:goodpost:

i solved it... i first tryed with a strait hdmi-hdmi cable instead of the dvi-hdmi i used before... no change.. then i tryed to update the nvidia driver from 180.xx to 185.85 and it is now FIXED!!

[cyberbeing]

all the PC video renderers let it affect the picture(except Overlay), this would be most welcome that mVR does it too(if that's not already the case)

This is already the case. madVR has always applied the graphic card luts before displaying the picture. The only renderer that doesn't is overlay like you mentioned. Whatever Ikarad is talking about, it's not the graphics card lut.

If he is talking about graphics card lut, I'm curious what graphics card he is using, since it sounds like he is the only person who has experienced this problem.

[leeperry]

oh ok, well AFAIK PowerDVD is not color managed...so I don't see how its VMR9 Renderless presenter could take simple .ICM files in account :confused:

anyway, case closed...if mVR does take the graphic card's CLUT in account, there's nothing to see here :)

@ikarad: I also have the Lacie BlueEyePro(rebadged i1d2), their software is awful...try ARGYLLCMS, I've discussed its usage thoroughly on HCFR w/ MLill ;)

something like "dispcal -v -yc -t6500 -g2.3 -qh test"

@Grmpf: well ideally, if we could input the ARGYLLCMS LUT data in cr3dlut, indeed we'd get an "all in one" 3D LUT(RGB conversion/gamut conversion/display color calibration), but again the graphic card's LUT works in 10bit(you can measure its accuracy w/ ARGYLLCMS...on a CRT it gave 9bit on a HD3850 and 10bit on a GF9600) so better let the graphic card do that part of the work ;)

[yesgrey]

Isn't that a "problem" of cr3dlut ?
Yes.
Am i wrong that 1 3D LUT can't handle it all ?
Yes you are.

Stay tuned...;)

[cyberbeing]

The more I use madVR 0.10, the more I'm starting to realize that it's still not as smooth as Haali Renderer. Haali seems silky smooth, but madVR seems to have slight micro stuttering which makes it seem not as stable on panning scenes. This is particularly noticeable on animated content with strong hard lines. With madVR the lines seem to vibrate during panning scenes more so then Haali.

In any case, madshi, I hope you still have something in mind to improve the smoothness even more to at least match Haali with exclusive mode disabled.

[tetsuo55]

We've talked about interpolation upsampling refresh rate before.

I just saw another thread about a panasonic TV that takes any signal and interpolates that to 600hz.

Now 600 is the lowest value by which all current interlaced and progressive refresh rates are devisable.

I wonder how taxing it would be to interpolate ANY > 600.
then based on the static output setting (24p/50p/60p) the best image would be selected and displayed.

So for a 60P display 1 out of 10 interpolated frames are displayed, the result should be smooth judderfree(even on diffucult pans) playback. However film effect would ofcourse be lost.

I bet a i7 with its 8 threads and a 4770 or faster GPU would be able to do this