madVR - high quality video renderer (GPU assisted)

[maxwell532]

Ran the test with the changes to BIOS you mentioned.

D3D9 Surface speed test:
NV12: upload 110 fps, download 428 fps, trick download failed
YV12: upload 109 fps, download 372 fps, trick download failed
A8R8G8B8: upload 43 fps, download 172 fps, trick download failed

DXVA Surface speed test:
NV12: upload 336 fps, download 425 fps, trick download failed
YV12: upload 109 fps, download 373 fps, trick download failed
A8R8G8B8: upload 47 fps, download 167 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 18 fps, download 179 fps
dynamic: upload 11 fps, download 187 fps, trick download 168 fps

So pretty much the same. I'll try to get that Windows 7 set up by tomorrow.

[nekromantik]

Hi guys
Anyone here using MadVR to auto change refresh rate on their LG LM series TVs to 24hz using MPCHC?
Mine switches fine but TV auto changes to 3D mode and then when I switch back to 2D there is no video output and MPCHC crashes.
Using Windows 7 64bit and got TV connected to PC via HDMI. PC running GTX460 card.

[madshi]

@Guys, in the future if you have a clear madVR bug to report, please use the madVR bug tracker! That really makes things easier for me!

This problem is occurring with 30p content. I don't play any interlaced content on my HTPC (most of it I ripped myself and deinterlaced/IVTCed in AviSynth before encoding with x264/xvid). As such all of my content is 24p, 25p or 30p. The specific content is from an NTSC video DVD that was deinterlaced to 30p then encoded as xvid in avi.

I've since found out that the problem is a little bit more complicated than indicated in my first post. Playing 30p x264/mkv content doesn't cause any problems but 30p xvid/avi does. I added 1920x1200p60 to the display modes and found that the 30p xvid/avi content that was switching to 24Hz now switches to 60Hz. The 30p x264/mkv content still switches to 30Hz. When playing the 30p xvid/avi content at 60Hz switching to exclusive mode triggers a resetting D3D device failed error. After a few seconds the video shows up paused and I can resume playback.

I have found I can fix the problem buy disabling deinterlacing in LAV Video. Then all 30p content plays at 30Hz. But even when deinterlacing is on auto in LAV Filters and the problem occurs, madVR reports deinterlacing off (says upstream). So not sure why changing this setting fixes it.

In any case I think the 60i logic you outlined in your response makes sense, but true 30p content should trigger a switch to 30Hz. Problem may be in LAV though.
That sounds weird. Can you please create a bug entry in the madVR bug tracker:

http://madVR.bugs.madshi.net

And please upload a small sample with which the problem can be reproduced and link to it in the bug tracker entry. Thanks.

The drops usually occur together with frame repeats

Here is a debug log:
http://www.mediafire.com/?gso55yxhjx97vq4
I can see a hint of what is going wrong, but the log is missing some information, unfortunately. I'll add the information in the next madVR build. Can you then please create and upload another log? The next build is not coming in the next couple of days, though.

Madshi,excellent build.The rendering times with Spline 4 taps have gone down dramatically.No frame drops,no glitches anymore.Quite a huge difference compared to 85.7,I must say.
Is 85.8 heavily optimized?
There's no change in v0.85.8 that I would expect to improve performance. So I'm not quite sure where the improvement is coming from for you...

Sorry for the newb question, but I can't get the madVR control panel to launch on my machine in any player. I always get a message along the lines of "Make sure your firewall isn't blocking madVR". I'm on Win 7 64 bit using mostly MPC-HC 32-bit and Zoom Player MAX. I've registered madVR.ax as Admin, but Windows just can't find the file. Is there a hot key combination to open the control panel?
That's weird. Maybe your anti-virus is blocking it from being executed? Or maybe the folder you've stored madVR in doesn't have execute rights for the user MPC-HC runs under? Anyway, the control panel is handled by the "madHcCtrl.exe", which you'll find in the madVR folder. Double clicking it should show the madVR tray icon. After that opening the settings dialog should work. Normally madVR should automatically start "madHcCtrl.exe", but it seems that doesn't work on your PC, for whatever reason...

I have similar problems when using dxva2 decoding. If I open a new file while playing a video in fullscreen exclusive mode on my secondary display, I often get a black screen or a MPC HC (or BE) crash. With software decoding everything works flawlessly.
If that is reproduceable for you, please create a bug entry in the madVR bug tracker and upload a (zipped) debug log. Thanks.

A little use complaint: I'm a bit disappointed by the loss of 1 feature: the ability to rename files for particular screen refresh files (60p, 23p, etc.).
Indeed, sometimes, even if videos are 24fps, I prefer to watch them at 60Hz, because there are short, and I don't need a change of refresh screen.
That features isn't lost! The tag logic just changed. The old logic was too prone to accidental activation. E.g. some files are named 1080p24, but they're actually 23p. So the latest madVR version accepts tags only in a specific format. See changelog for more details.

What exacly are presentation glitches in fullscreen exclusive mode? I dont usually use it but im asking out of curiosity,as i can accumulate really a lot of them during standard playback.
Im using latest drivers for ati hd4870.
Presentation glitches in fullscreen exclusive mode are glitches in Direct3D itself. Basically madVR told D3D when to present which video frame and D3D failed to do so, for unknown reasons. This can happen if you modified your flush settings in a bad way. Try resetting to default settings.

Also,do you think that GeForce gtx260 would be better for video playback overall? Are there any serious differences between ati and nvidia considering video playback?

EDIT:also,what exacly are gpu flushing options used for? and what does presenting few frames in advance do?
These questions have been asked and answered multiple times already. Please try to find it via search. I don't really want to post the same replies every couple of pages...

basically,i see something like current frame+last frame at the same time on screen(not really eye-hurting on movies,but my eyes bleed on anime-its particulary visible when panning over hard edges) -could it be because i forced 72hz refreshrate in powerstrip and gpu video logic still tries to render@60hz?
What dxdiag says doesn't matter. If madVR, PowerStrip and Reclock all agree on the refresh rate then there should be no problem. Are there frame drops noted in the OSD? As a first try I'd suggest to reset madVR to default settings. Then try switching everything to "bilinear", just to make sure your problems are not caused by your GPU being too slow...

And what into the account of optimization with Crossfire (2xSapphire_HD_7770_Vapor-X_Black_Diamond) ^^?
I'm not sure how Crossfire works exactly, from the view point of a game. There's no special code in madVR for Crossfire setups. I don't have such a setup so I don't know what to do, and even if I knew, I'd have no way to test it.

madshi, thank you for this software. But I have an odd problem. I was seeing presentation glitches only in windowed mode and not in exclusive mode, and also not in EVR renderer. After changing some possibly related settings (GPU flush, queue sizes, separate device..) without success, I found that switching my image and chroma algorithms from Jinc 3 AR (or 8) to Lanczos 3 AR fixed the glitches. Also after further testing, I disabled the anti ringing filter on Jinc, and that caused the playback to be even more, extremely glitchy. I'm using: Windows 8, high end hardware with AMD GPU (latest drivers), 120HZ RR monitor + MPC-HC, ReClock. Rendering time is 0.07ms on both Jinc 3 and 8, with or without AR.
120Hz is not easy to handle, especially in windowed mode. It should work more reliably in exclusive mode, if your GPU is fast enough for that. If you have no glitches in exclusive mode, then just use exclusive mode and be happy.

madshi, this option says (Vista and newer), can it be any use on xp64 ?

I have "stuck window" bug too on xp64 with nvidia and I can say that its usually Chrome or Firefox who gets stuck like that here and I think I fixed Firefox by disabling hardware acceleration in it, but Chrome still gets stuck especially if Google Dictionary (by Google) 3.0.15 extension is used in it...
You can't use it on xp64. The "stuck window" problem on xp64 can't have anything to do with aero/desktop composition because aero doesn't exist on xp64. The problem will have to do with the browsers trying to use Direct3D acceleration, which they can't do when somebody else is in fullscreen exclusive mode. The same problem should occur when using EVR in exclusive mode. There's nothing I can do about it. If disabling hardware acceleration in your browser doesn't help then the browser's hardware acceleration switch is likely broken.

One question @Madshi. Does the OSD add to the rendering time?
Of course it does. Nothing is free. However, the cost should be relatively small.

I have a request for you, if it's not too much trouble:

Would it be possible to have madVR disable yCMS, gamma and colour processing when taking a screenshot?
yCMS should be disabled. Gamma and color processing unfortunately not. There's nothing I can do about that right now (without spending a *lot* of time on it). I plan to improve screenshot functionality in a future version, but not anytime soon.

Let me first say I'm very, very pleased with madVR using default settings. I mostly play commercial movies, both DVD and Blu-ray, using LAV filters.

Being the tweaker I am, however, and being the proud owner of a new GTX 660, I'm wondering if I should fool around with some of the advanced settings. I'm running an i5 3570K with a mild overclock, so I've got decent CPU resources. My display is an LED IPS running at 1920x1080.

Do any of you madVR experts have recommendations for advanced settings tweaks for my (basically vanilla) type of usage, or should I leave well enough alone? ;)
You could try Overlay mode. And you could try different scaling algorithms. Also make sure you setup the display mode changer (if your display supports 24p etc).

I'm using Intel HD 4000 GPU with madVR set to use DXVA2 up-/downscaling, MPC-HC player and LAVFilters.
It works great with majority of files, but I've encountered a small bug when playing a little unusual file - h264, 10bit, 4:4:4 - there's a green line on the left border when using DXVA scaling, sometimes it crashes when starting playback. Doesn't happen with madVR's own scaling algorithms.
After playing with the settings, it seems to only happen when madVR does DXVA scaling, while receiving a 4:4:4 "colour format" (Y410/v410/YV24/Y416). Doesn't happen when LAVFilters output other formats (RGB32, RGB24, etc.).[/CODE]
Thanks, I'll have a look at that. In the future please create a bug entry in the madVR bug tracker for bugs like this. That makes my life a lot easier.

Even though it isn't technically correct information from the file itself but would it be appropriate for madVR to flag the stream as 23.976 fps (when IVTCing) for Reclock? It already does so for soft-flagged 29.970 MPEG-2 streams and it works well.
madVR does not "flag the stream". I don't know where Reclock gets its information from exactly...

I seem to have trouble playing 29.970fps material. it is very choppy and movement and pannings and such look quite terrible. its not specific to madvr though. i get it every player i try. are there any settings i can try? I live in denmark. is it normal to have movies or a series or simlar content in 30fps in stead of around 24fps?
Is it true interlaced material? Or is it movie content? You could try forcing madVR into film mode and then switch your display to 24Hz. Does that make things go smooth? Or do you get combing/interlaced artifacts that way?

Hello guys, I'm new to MPC HC and madVR+LAV filters.

I have one huge issue. If I enable "exclusive mode" when entering full screen I get a red text popping out "Out of memory", although the video in the background is still playing just fine.
Any suggestions?

GPU: Palit Nvidia GeForce GTX460 (1024MB GDDR5)
Weird. Are you using madVR default settings? Make sure your GPU Queue is not set too high. Do you have yCMS enabled? Try disabling it. Also make sure you disable stuff like Anti-Aliasing, Anisotropic filtering etc in your GPU control panel (or set to "application preferred" or however that option is named). You could also try to use some GPU tweaker tool to check GPU RAM consumption. Maybe it's already very high before you even start video playback?

Hi guys,
i recently configured MPC-HC, with the LAV filters & madVR.
for some reason though, when i press "control-J", it shows the composition rate to be 30Hz, when it SHOULD be 60Hz. playback is obviously not correct with this rate.
what can i do to correct this?
thanks videophiles!
Do you use the MPC-HC refresh rate switcher? Or any other refresh rate switcher other than madVR's? If so, try using madVR's display mode switcher instead. If you don't use a display/refresh mode switcher at all, then I don't know where the problem comes from. Might be a bug in your GPU drivers. Or maybe your PC is in energy-power-save mode, trying to save power by reducing the composition rate? In any case, madVR has zero control over the composition rate. I don't even know exactly who decides it. Might be D3D (OS) or might be the GPU drivers, I don't know...

Yeah, my NEC 24WMGX3 supports 1080p/50 over HDMI (not DVI for some reason) but it just introduces tearing. Clearly no native support for anything other than 60 Hz.
Well, you should *love* madVR v0.86.0, then.

CRU seems to work much better than Nvidia's custom resolution utility for me. It adds another refresh rate and resolution option as if it was supported by the monitor.

You can also edit the "normal" 60Hz to 59 or anything you want.
One big problem with the EDID is that the pixel clock can only be defined very roughly. And unfortunately the EDID doesn't support a 1.001 refresh rate divisor, either. Which means EDID tweaks will probably work perfectly for 24.000, 25.000, 50.000 and 60.000, but I think results for 23.976, 59.940 etc will be sub-optimal...

My question is: If I uncheck the box (no auto-calibrate) in the GUI window shown below when generating the .3dLUT, will the .3dLUT be made to work properly along with the ICC profile loaded by Windows 7 (both the ICC profile and the .3dLUT in madVR will be active):confused:
I don't really know, unfortunately.

I recently read a discussion with one of the big names in LUT creation tools, and they said that using 5-point cube was enough for proper gamut correction (with data was sourced from 17-point measurements) and a 21-point greyscale correction was sufficient. I would argue that with consumer displays and meters, you are probably better off with 9-point greyscale correction.
Do you mean 17 measurements or do you mean 17*17*17 = 4913 measurements (ouch!!)?

@maxwell532

Assuming that same thing occurs in Windowed Mode with all madVR scalers set to Bilinear, yet all 8-bit videos are fine, it probably means there is something wrong with either your PCI-E interface or GPU memory bandwidth.
Good thinking!

Hi guys
Anyone here using MadVR to auto change refresh rate on their LG LM series TVs to 24hz using MPCHC?
Mine switches fine but TV auto changes to 3D mode and then when I switch back to 2D there is no video output and MPCHC crashes.
Using Windows 7 64bit and got TV connected to PC via HDMI. PC running GTX460 card.
That sounds weird. Are you using the MPC-HC refresh rate changer or the madVR display mode changer? Maybe creating a 1080p24 custom resolution could help?

[maxwell532]

Okay, got Windows 7 installed. Here are the results:
D3D9 Surface speed test:
NV12: upload 1051 fps, download 487 fps, trick download failed
YV12: upload 872 fps, download 415 fps, trick download failed
A8R8G8B8: upload 323 fps, download 180 fps, trick download failed

DXVA Surface speed test:
NV12: upload 363 fps, download 480 fps, trick download failed
YV12: upload 869 fps, download 416 fps, trick download failed
A8R8G8B8: upload 334 fps, download 174 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 136 fps, download 260 fps
dynamic: upload 265 fps, download 9 fps, trick download 186 fps
Also tried playing 8-bit and 10-bit 1080p video and got 0 dropped frames after playing them for a minute or two.

Looks like I'm going back to Windows 7.

Thanks for your help.

[Jong]

@madshi, of course "relatively small" is a relative term! Do you mean <0.2ms, around 1ms or somewhere in between (with current typical HTPC GPUs)? It is just that using jinc-3 for both c&l is still marginal. So tight it would certainly be useful to know if we had an extra 1ms to work with or not. If it is "relatively" significant it would be useful to be able to log the OSD stats without displaying them to get a true picture of rendering time. Of course, these days it may not be significant at all, in which case it would just be nice to know.

[nekromantik]

[b]
That sounds weird. Are you using the MPC-HC refresh rate changer or the madVR display mode changer? Maybe creating a 1080p24 custom resolution could help?

Nope I am using MadVR changer only.
I shall try that and get back to you. Thanks
This is a 2012 TV model so should not cause issues for MadVR, I will also try updating MPCHC.

[cyberbeing]

Okay, got Windows 7 installed. Here are the results:
D3D9 Surface speed test:
NV12: upload 1051 fps, download 487 fps, trick download failed
YV12: upload 872 fps, download 415 fps, trick download failed
A8R8G8B8: upload 323 fps, download 180 fps, trick download failed

DXVA Surface speed test:
NV12: upload 363 fps, download 480 fps, trick download failed
YV12: upload 869 fps, download 416 fps, trick download failed
A8R8G8B8: upload 334 fps, download 174 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 136 fps, download 260 fps
dynamic: upload 265 fps, download 9 fps, trick download 186 fps
Also tried playing 8-bit and 10-bit 1080p video and got 0 dropped frames after playing them for a minute or two.

Looks like I'm going back to Windows 7.

Thanks for your help.

Those Win7 results look normal enough, which is good news!

Hmm... Really makes me wonder what went wrong in Win8. Did you have the latest Windows 8 AMD chipset drivers (http://www2.ati.com/drivers/13-1_vista_win7_win8_32-64_sb.exe) and DirectX Runtimes (http://www.microsoft.com/en-us/download/details.aspx?id=35) installed?

I can only suspect that your PCI-E link was getting stuck in a lower power state for some reason and was misreporting its link-speed.
Could be a BIOS compatibility issue of some sort, so it probably wouldn't hurt to contact ASUS Technical Support if only to inform them of the potential issue.

[iSunrise]

Originally Posted by DragonQ
Yeah, my NEC 24WMGX3 supports 1080p/50 over HDMI (not DVI for some reason) but it just introduces tearing. Clearly no native support for anything other than 60 Hz.
Well, you should *love* madVR v0.86.0, then.
Wow, that sounds very promising. Cannot wait to test 0.86.0!

[maxwell532]

Did you have the latest Windows 8 AMD chipset drivers (http://www2.ati.com/drivers/13-1_vista_win7_win8_32-64_sb.exe) and DirectX Runtimes (http://www.microsoft.com/en-us/download/details.aspx?id=35) installed?
Yep.


Could be a BIOS compatibility issue of some sort, so it probably wouldn't hurt to contact ASUS Technical Support if only to inform them of the potential issue.
I'll do that.

[madshi]

@madshi, of course "relatively small" is a relative term! Do you mean <0.2ms, around 1ms or somewhere in between (with current typical HTPC GPUs)?
What is a current typical HTPC GPU? I can't really say an exact number, but with e.g. a 660 I would expect it to be less than 0.2ms.

[DragonQ]

Okay, got Windows 7 installed. Here are the results:
D3D9 Surface speed test:
NV12: upload 1051 fps, download 487 fps, trick download failed
YV12: upload 872 fps, download 415 fps, trick download failed
A8R8G8B8: upload 323 fps, download 180 fps, trick download failed

DXVA Surface speed test:
NV12: upload 363 fps, download 480 fps, trick download failed
YV12: upload 869 fps, download 416 fps, trick download failed
A8R8G8B8: upload 334 fps, download 174 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 136 fps, download 260 fps
dynamic: upload 265 fps, download 9 fps, trick download 186 fps
Also tried playing 8-bit and 10-bit 1080p video and got 0 dropped frames after playing them for a minute or two.

Looks like I'm going back to Windows 7.

Thanks for your help.
That is strange. I wonder if it's a GPU driver issue? I haven't had any problems with MadVR in Windows 8. Remember that Windows 8's "desktop mode" is very similar to Windows 7 (just as Windows 7 was very similar to Windows Vista) in terms of program compatibility.

[6233638]

Do you mean 17 measurements or do you mean 17*17*17 = 4913 measurements (ouch!!)17³ which actually doesn't take that long if you are using a pro-grade meter like the Klein K10 (http://www.youtube.com/watch?v=MbrVVDHvnls). Totally unfeasible with consumer-grade hardware though, as it will drift too much over the length of time it takes to make those measurements. You could maybe do it with the i1Display Pro/SpectraCal C6 colorimeter, but it would still take hours.

The reason they do this, is that LightSpace profiles the display once, and then uses that data to create any future LUTs you require, rather than profiling the display each time, as most packages do.

As mentioned, they then "average" that to lower precision LUTs for use, because that's actually too much data to work well.

To be honest, I think I prefer profiling the display each time, especially on a consumer-grade display.

[leeperry]

you should *love* madVR v0.86.0, then.
Ouh, some sort of motionflow would seriously kill :devil:

Barely any of the displays I recently tried would support 48Hz, but this sort of feature always does wonders on 29.97fps@59.94Hz when set to its least agressive settings IME.....24fps@60Hz might be pushing it, but you always find new ways to surprise us so thanks for the teaser ^^

[yahyoh]

ok i did the test and i got very low download :/ :(

IDK why

Windows NT New x64
ATI Mobility Radeon HD 5400 Series

D3D9 Surface StretchRect:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-255)
D3DFMT_X8R8G8B8: lossy (0-255)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: lossy (0-255)
D3DFMT_A2R10G10B10: lossless (0-255)
D3DFMT_A2B10G10R10: lossless (0-255)
D3DFMT_A16B16G16R16: lossless (0-255)
D3DFMT_A16B16G16R16F: lossless (0-255)
D3DFMT_A32B32G32R32F: lossless (0-255)

D3D9 Surface VideoProcessor:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (16-242)
D3DFMT_X8R8G8B8: lossy (16-242)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: VideoProcessBlt failed
D3DFMT_A2R10G10B10: lossy (16-243)
D3DFMT_A2B10G10R10: lossy (16-243)
D3DFMT_A16B16G16R16: VideoProcessBlt failed
D3DFMT_A16B16G16R16F: VideoProcessBlt failed
D3DFMT_A32B32G32R32F: VideoProcessBlt failed

DXVA Surface StretchRect:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-255)
D3DFMT_X8R8G8B8: lossy (0-255)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: lossy (0-255)
D3DFMT_A2R10G10B10: lossless (0-255)
D3DFMT_A2B10G10R10: lossless (0-255)
D3DFMT_A16B16G16R16: lossless (0-255)
D3DFMT_A16B16G16R16F: lossless (0-255)
D3DFMT_A32B32G32R32F: lossless (0-255)

DXVA Surface VideoProcessor:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (16-242)
D3DFMT_X8R8G8B8: lossy (16-242)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: VideoProcessBlt failed
D3DFMT_A2R10G10B10: lossy (16-243)
D3DFMT_A2B10G10R10: lossy (16-243)
D3DFMT_A16B16G16R16: VideoProcessBlt failed
D3DFMT_A16B16G16R16F: VideoProcessBlt failed
D3DFMT_A32B32G32R32F: VideoProcessBlt failed

D3D9 Surface speed test:
NV12: upload 384 fps, download 4 fps, trick download failed
YV12: upload 391 fps, download 4 fps, trick download failed
A8R8G8B8: upload 181 fps, download 2 fps, trick download failed

DXVA Surface speed test:
NV12: upload 380 fps, download 4 fps, trick download failed
YV12: upload 359 fps, download 4 fps, trick download failed
A8R8G8B8: upload 183 fps, download 2 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 151 fps, download 214 fps
dynamic: upload 326 fps, download 6 fps, trick download 159 fps



BTW i'm using WIN 8

[DragonQ]

I get a lot of failures on Windows 8 too. Probably not a coincidence.

Windows NT New x64
NVIDIA GeForce GTS 250

D3D9 Surface StretchRect:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-255)
D3DFMT_X8R8G8B8: lossy (0-255)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: creating GPU texture failed
D3DFMT_A2R10G10B10: lossy (0-255)
D3DFMT_A2B10G10R10: lossy (0-255)
D3DFMT_A16B16G16R16: StretchRect failed
D3DFMT_A16B16G16R16F: luma lossless (0-255), chroma lossy
D3DFMT_A32B32G32R32F: StretchRect failed

D3D9 Surface VideoProcessor:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-255)
D3DFMT_X8R8G8B8: lossy (0-255)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: creating GPU texture failed
D3DFMT_A2R10G10B10: lossy (0-255)
D3DFMT_A2B10G10R10: lossy (0-255)
D3DFMT_A16B16G16R16: lossy (0-255)
D3DFMT_A16B16G16R16F: luma lossless (0-255), chroma lossy
D3DFMT_A32B32G32R32F: luma lossless (0-255), chroma lossy

DXVA Surface StretchRect:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-204)
D3DFMT_X8R8G8B8: lossy (0-204)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: creating GPU texture failed
D3DFMT_A2R10G10B10: lossy (0-204)
D3DFMT_A2B10G10R10: lossy (0-204)
D3DFMT_A16B16G16R16: StretchRect failed
D3DFMT_A16B16G16R16F: lossy (1-190)
D3DFMT_A32B32G32R32F: StretchRect failed

DXVA Surface VideoProcessor:
D3DFMT_R8G8B8: creating GPU texture failed
D3DFMT_A8R8G8B8: lossy (0-204)
D3DFMT_X8R8G8B8: lossy (0-204)
D3DFMT_A8B8G8R8: creating GPU texture failed
D3DFMT_X8B8G8R8: creating GPU texture failed
D3DFMT_A2R10G10B10: lossy (0-204)
D3DFMT_A2B10G10R10: lossy (0-204)
D3DFMT_A16B16G16R16: lossy (1-204)
D3DFMT_A16B16G16R16F: lossy (1-190)
D3DFMT_A32B32G32R32F: lossy (1-190)

D3D9 Surface speed test:
NV12: upload 503 fps, download 523 fps, trick download failed
YV12: upload 394 fps, download 361 fps, trick download failed
A8R8G8B8: upload 269 fps, download 221 fps, trick download failed

DXVA Surface speed test:
NV12: upload 473 fps, download 571 fps, trick download failed
YV12: upload 366 fps, download 360 fps, trick download failed
A8R8G8B8: upload 311 fps, download 221 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 118 fps, download 307 fps
dynamic: upload 315 fps, download 32 fps, trick download 213 fps

[6233638]

I wish I still had a record of my tests on Windows 7, but things seem fine here on Windows 8? At least, I haven't had any performance complaints.
Windows NT New x64
NVIDIA GeForce GTX 570

D3D9 Surface speed test:
NV12: upload 1048 fps, download 1003 fps, trick download failed
YV12: upload 1056 fps, download 832 fps, trick download failed
A8R8G8B8: upload 343 fps, download 402 fps, trick download failed

DXVA Surface speed test:
NV12: upload 727 fps, download 929 fps, trick download failed
YV12: upload 976 fps, download 832 fps, trick download failed
A8R8G8B8: upload 350 fps, download 353 fps, trick download failed

A8R8G8B8 Texture speed test:
default: upload 275 fps, download 300 fps
dynamic: upload 427 fps, download 30 fps, trick download 216 fps

I'm not sure how this corresponds to madVR's current state though. This test is from a long time back.


Ouh, some sort of motionflow would seriously kill :devil:

Barely any of the displays I recently tried would support 48Hz, but this sort of feature always does wonders on 29.97fps@59.94Hz when set to its least agressive settings IME.....24fps@60Hz might be pushing it, but you always find new ways to surprise us so thanks for the teaser ^^I would expect frame blending for 24p @ 60Hz to try and reduce judder.

[Libeluratio]

Hello, I did'nt see this before but I have a strange problem with madvr !

Two photos taken from mpc-hc that speaks from themselves:

Video with Madvr (http://img850.imageshack.us/img850/8234/ergoproxymadvr.jpg)

Video With System Default Renderer (http://img705.imageshack.us/img705/3073/ergoproxysystemdefault.jpg)

I triple-checked with other video players: the eyes should be blue and not green as they are with madvr.
Furthermore, when I encode the video with handbrake, then madvr displays the eyes blue as it should. The problem is, I do not want to recode all of my videos.


Somebody knows about this weird problem ???


Thank you so much !!!!!

I run windows 7 64bits, and the madvr 0.85.8

[dansrfe]

Hello, I did'nt see this before but I have a strange problem with madvr !

Two photos taken from mpc-hc that speaks from themselves:

Video with Madvr (http://img850.imageshack.us/img850/8234/ergoproxymadvr.jpg)

Video With System Default Renderer (http://img705.imageshack.us/img705/3073/ergoproxysystemdefault.jpg)

I triple-checked with other video players: the eyes should be blue and not green as they are with madvr.
Furthermore, when I encode the video with handbrake, then madvr displays the eyes blue as it should. The problem is, I do not want to recode all of my videos.


Somebody knows about this weird problem ???


Thank you so much !!!!!

I run windows 7 64bits, and the madvr 0.85.8

Do you have a small sample to test with?

[Libeluratio]

Yeah,

Here (https://mega.co.nz/#!WUVTgbyT!Sx6-lBBr9IGT0IP8HdibEBwcsm9U5WDXjYMAGQPhFqk) you can download a quick sample I just made (3.6 MB)

(look at the girl's makeup, it should be blue, for me, it displays green with madvr, and correct blue with "system default" renderer of mpc-hc)


Thank you !

[leeperry]

I would expect frame blending for 24p @ 60Hz to try and reduce judder.
yep, and hopefully for 25fps@50Hz and 29.97fps@59.94Hz as well =)

I vaguely remember madshi stating that FI wasn't part of his masterplan but also mentioning that it might not be too wise to buy a "barely" fast enough GPU for Jinc3AR because he could add more demanding algorithms later on. I can only imagine the kind of crunching power Jinc3AR FI would require....can you say GTX670 :devil:

[Nekomata]

It can't seem to show the "Opaque box" in subtitles option on Potplayer x64, with a previous version of madvr the opaque box showed fine in the subtitles, not now.....

[QBhd]

Yeah,

Here (https://mega.co.nz/#!WUVTgbyT!Sx6-lBBr9IGT0IP8HdibEBwcsm9U5WDXjYMAGQPhFqk) you can download a quick sample I just made (3.6 MB)

(look at the girl's makeup, it should be blue, for me, it displays green with madvr, and correct blue with "system default" renderer of mpc-hc)


Thank you !

And what makes you think the system default render is doing the correct job in color reproduction and levels? I for one actually think the makeup is supposed to be green... Your madVR image looks far superior.

QB

[ryrynz]

I for one actually think the makeup is supposed to be green... Your madVR image looks far superior.


The makeup is indeed supposed to be blue (http://1-media-cdn.foolz.us/ffuuka/board/a/image/1337/82/1337828539925.jpg). Could be related to the graphics card settings.

[zamadatix]

And what makes you think the system default render is doing the correct job in color reproduction and levels? I for one actually think the makeup is supposed to be green... Your madVR image looks far superior.

QB

How precise an image is calculated has no causal effect to how accurate it is. I for one think the series cover (http://www.imdb.com/title/tt0791205/) shows the rendered color is incorrect, more proof can be seen on Google.

[Libeluratio]

Indeed, the makeup is supposed to be blue, and in every other player that is not using madvr, it is, and when i encode the video, madvr accepts to display it blue as well, but not with the original (or sample) file, where other renderers do.

I guess it has something to do with the video's colors profile or something (I don't really know about that at all in fact), that's very strange...

As if, with this file, madvr cannot do the right color conversion, right ?

[QBhd]

Fair enough... It does seem to be an interesting situation.

QB

[cyberbeing]

Here (https://mega.co.nz/#!WUVTgbyT!Sx6-lBBr9IGT0IP8HdibEBwcsm9U5WDXjYMAGQPhFqk) you can download a quick sample I just made (3.6 MB)

That sample has the YCgCo matrix specified.
Is that sample really supposed to be YCgCo and not standard BT.709 YCbCr as it appears to be?
If so, whatever you used to convert to YCgCo didn't do so correctly.


madVR is the only renderer which supports the YCgCo matrix. Using an Avisynth filter to convert the video from YCgCo to RGB shows the same colors as madVR, so it's your video which is incorrect.

[cyberbeing]

Is the file really an incorrect encode?
...
CTRL+SHIFT+ALT+M to change the matrix fixes the colors.
...
Remuxing with eac3to fixes the colors

The H.264 bitstream of his file specifies YCgCo, which means whoever encoded this file in x264 specified that matrix explicitly.

eac3to uses Haali to mux MKV, which appears to strip all the color matrix, primary, and transfer information from the bitstream. That could be considered a bug in Haali's muxer.

[Libeluratio]

Ok, thank you for your explanation, but why remuxing with mkvmerge won't solve the problem if the color matrix is specified in the bitstream of the sample and stripped just by eac3to ?

And why recoding with handbrake fix this color matrix problem ? isn't the color matrix a "hard" specification ?

(I do not understand how, when recoding whit handbrake, it says "ooooh that should'nt be green, that should be blue with madvr", and how other renderers can display the right colors of the original file ?)

Thank you, I'm learning interesting things !

[ryrynz]

(I do not understand how, when recoding whit handbrake, it says "ooooh that should'nt be green, that should be blue with madvr",


Maybe Handbrake defaults to BT.709 YCbCr?


and how other renderers can display the right colors of the original file ?)


Cyberbeing answered that one.


madVR is the only renderer which supports the YCgCo matrix.

[cyberbeing]

How did you encode the sample file which you linked above? (Note: Doom9 is quite strict about Rule #6 (http://forum.doom9.org/forum-rules.htm) violations, if this is something you downloaded from the internet)

By default, no matrix is specified when encoding with x264. It's something which is done explicitly at encode time, so you should know if you've done this.

If you incorrectly specified YCgCo as the matrix when the source video you were encoding was BT.709, that is why this problem has occurred.

If you have a 1920x1080 video flagged with the YCgCo matrix, only madVR will use the YCgCo matrix. All other renders will usually use BT.709, which normally would be incorrect. In your case the video was either encoded incorrectly, or flagged at encode time with the incorrect matrix.

[robpdotcom]

Originally Posted by robpdotcom (paraphrase of deleted post)
Is the file really an incorrect encode?
...
CTRL+SHIFT+ALT+M to change the matrix fixes the colors.
...
Remuxing with eac3to fixes the colorsThe H.264 bitstream of his file specifies YCgCo, which means whoever encoded this file in x264 specified that matrix explicitly.

eac3to uses Haali to mux MKV, which appears to strip all the color matrix, primary, and transfer information from the bitstream. That could be considered a bug in Haali's muxer.

Sorry, looks like I was deleting as you were posting. :sly:

[Libeluratio]

I could'nt really say since I'm living in a university residence and that's a file on it's own server. So, indeed, I didn't encode it myself.

What I find interesting is, as you said, muxing with eac3to produces a correct mkv, whereas muxing with mkvmerge gives me the same type of file with wrong matrix. Do you know why ?

I usually use Tsmuxer/Mkvmerge when I dump my blueray and I begin to ask If I should use eac3to instead of mkvmerge ?? But, hey, I know, that's no more a madvr issue there

[leeperry]

PotP [..] its pan&scan feature is great, but not accurate at all as most of the time it will shrink or stretch the picture by as much as 12 pixels(just like MPC).......a far cry from ffdshow with its 0.01 AR accuracy :(
Just for the record, PotP's coder told me to forget about the whole thing but the newest versions allow extremely accurate pan & scan with mVR, which works really great to fix ITU-compliant material: http://thumbnails102.imagebam.com/23674/4035db236734493.jpg (http://www.imagebam.com/image/4035db236734493)

[robpdotcom]

muxing with eac3to produces a correct mkv, whereas muxing with mkvmerge gives me the same type of file with wrong matrix.

Actually mkvmerge has it right. It's whoever encoded the video that got it wrong.

Muxing with eac3to (which uses Haali to mux) doesn't produce a correct mkv, it just gets it wrong in a way that hides the real problem. See cyberbeing's post above for why: http://forum.doom9.org/showthread.php?p=1615242#post1615242

[QQ]

is it possible to make madvr work over remote desktop connection?

[nevcairiel]

yep, and hopefully for 25fps@50Hz and 29.97fps@59.94Hz as well =)

I vaguely remember madshi stating that FI wasn't part of his masterplan but also mentioning that it might not be too wise to buy a "barely" fast enough GPU for Jinc3AR because he could add more demanding algorithms later on. I can only imagine the kind of crunching power Jinc3AR FI would require....can you say GTX670 :devil:

If i remember madshis comments correctly, all he wanted to do is replace the judder by blending, so for 24p@60Hz, it would be "1, 1, 1+2, 2, 2" so that instead of a 3:2 judder pattern, you get every frame shown for "2.5" periods, but not with interpolation (unless you consider blending two frames interpolation already). Doubtful this would be any good for clear integer multiples.

[madshi]

17³ which actually doesn't take that long if you are using a pro-grade meter like the Klein K10 (http://www.youtube.com/watch?v=MbrVVDHvnls). Totally unfeasible with consumer-grade hardware though, as it will drift too much over the length of time it takes to make those measurements. You could maybe do it with the i1Display Pro/SpectraCal C6 colorimeter, but it would still take hours.

The reason they do this, is that LightSpace profiles the display once, and then uses that data to create any future LUTs you require, rather than profiling the display each time, as most packages do.

As mentioned, they then "average" that to lower precision LUTs for use, because that's actually too much data to work well.
Well, if they do 17³ measurements then I don't see the point in averaging them down. madVR can handle 3dluts up to 256³. I would rather interpolate the measurements *up* instead of rounding them down. But that's probably a limitation of the hardware they're using to apply the 3dluts. That hardware is probably not able to handle really bug 3dluts.

ok i did the test and i got very low download :/ :(

IDK why
That's totally normal for older generation AMD GPUs. And it doesn't matter at all because madVR never "downloads" with AMD GPUs.

I get a lot of failures on Windows 8 too. Probably not a coincidence.
Those "trick download" failures are not really important.

The only thing that is important is that the *up*load framerates are fast enough. The *down*load is only used for Intel and NVidia GPUs, and madVR uses a much faster algorithm for that than the madNv12Test test tool. So really the main thing you should be looking for in madNv12Test results is the upload speed.

Two photos taken from mpc-hc that speaks from themselves:

Video with Madvr (http://img850.imageshack.us/img850/8234/ergoproxymadvr.jpg)

Video With System Default Renderer (http://img705.imageshack.us/img705/3073/ergoproxysystemdefault.jpg)

I triple-checked with other video players: the eyes should be blue and not green as they are with madvr.
Here's the detailed answer:

http://bugs.madshi.net/view.php?id=4

That sample has the YCgCo matrix specified.
Exactly. The video bitstream is broken/incorrect. The only reason why other renderers show this video correctly is because they don't even know what YCgCo is.

This problem can be worked around by adding a "[matrix=bt709]" tag to the file name.

I vaguely remember madshi stating that FI wasn't part of his masterplan
Exactly.

If i remember madshis comments correctly, all he wanted to do is replace the judder by blending, so for 24p@60Hz, it would be "1, 1, 1+2, 2, 2" so that instead of a 3:2 judder pattern, you get every frame shown for "2.5" periods, but not with interpolation (unless you consider blending two frames interpolation already).
Correct. However, the final solution will blend 2 out of 5 frames for 24@60, for reasons to be explained later.

It can't seem to show the "Opaque box" in subtitles option on Potplayer x64, with a previous version of madvr the opaque box showed fine in the subtitles, not now.....
You're using the PotPlayer internal subtitle renderer, correct? Does it work with MPC-HC? If so, it is very probably a bug in PotPlayer because both MPC-HC and PotPlayer use the same madVR interface to draw subtitles.

is it possible to make madvr work over remote desktop connection?
I'm not sure. Maybe. But it's not a priority for me at the moment.

[Niyawa]

Hey madshi, got 2 questions for ya.

1. We at the KCP (kawaii codec pack) (http://haruhichan.com/forum/showthread.php?7545-KCP-Kawaii-Codec-Pack) team want to make a filter in the pack that recognizes the GPU of the user and says if it's compatible or not with madVR itself. If possible, could you give some more details in what we should consider when filtering GPUs aside from what you already mentioned in the madVR requirements? I can't really trust those either, no offense. I'm using madVR with a onboard Intel that has less than 128 MB of dedicated memory yet it works perfectly so...

2. I'm not sure if this is a bug of MPC-HC, or madVR so I'll just throw this and see what you can say about it.

"My scenario: two monitors that are 1900x1200. One is landscape 1900x1200 the other is portrait 1200x1900. Moving from the primary monitor(landscape) to the secondary(portrait) will crash mpc or opening a file where the mpc instance starts on the secondary monitor will crash mpc. It happens when using madVR (evrcp is fine). Switching the secondary monitor to landscape and then moving between screens does not crash mpc when using madVR. I know in the past LAV + madVR has issues like this when used together. But yea I am not gonna point blame at anything and wanna know if anyone else can reproduce this problem? Using latest catalyst drivers, might be an ati issue too."

"I have them @1920x1080 and just test this with Intel iGPU: mpc-hc is not crash on both scenario but video is just stop rendering. The playback is still going, audio is playing and every time you minimize and maximize the mpc window, video will play briefly (my guess is due to left over in render queue) and get updated, then hang again. The only case it will crash is when you hit stop playback, hit pause or close mpc will not crash it. EVR-CP is fine, put secondary monitor on landscape mode is fine. My guess is madVR doesn't support this, do you want to ask madshi about this?"

Any thoughts? I should probably ask them to make a madVR bug report in the bug tracker now that I think about it.

[madshi]

When using all bilinear scaling algorithms, madVR can work with almost any DX9 GPU. The only requirement then is to have DX9 PS3.0 *hardware* shader support (software shader support, as some old Intel GPUs have, is not enough). The amount of GPU RAM needed depends on how high the GPU queue is set, whether you're using higher algorithms than bilinear, whether you've activated the "use 10bit textures" option in madVR etc. You can probably get away with even 64MB RAM GPUs if you set all the madVR options accordingly. It's hard for me to give you a list of all GPUs madVR works with or doesn't work with because I simply don't have so many different GPUs here to test with. But pretty much the only hard limit is PS3.0 hardware shader support, I think.

If you look at the first page of this thread, you will find one limitation mentioned there: Which is rotated (landscape vs portrait) displays. madVR currently doesn't support rotated displays. I've had a quick look at how to make it work - and failed to find the cause of the problem. It's on my list of things to look at (again), but it's not high priority at the moment. Currently if you play a video on a rotated display, usually you'll just get a black screen, I think. Or maybe a crash if you let it run very long, not sure...

[Niyawa]

When using all bilinear scaling algorithms, madVR can work with almost any DX9 GPU. The only requirement then is to have DX9 PS3.0 *hardware* shader support (software shader support, as some old Intel GPUs have, is not enough). The amount of GPU RAM needed depends on how high the GPU queue is set, whether you're using higher algorithms than bilinear, whether you've activated the "use 10bit textures" option in madVR etc. You can probably get away with even 64MB RAM GPUs if you set all the madVR options accordingly. It's hard for me to give you a list of all GPUs madVR works with or doesn't work with because I simply don't have so many different GPUs here to test with. But pretty much the only hard limit is PS3.0 hardware shader support, I think.
Okay. Some days ago someone passed by in the KCP forums with a problem, we figured out it was because madVR couldn't be used since his GPU didn't have PS3.0 hardware shader support. You don't really need to make a list, I'm already working on a GPU database so I can give to the KCP developer and make him implement it.

I'll make sure to give you the list later, it might be useful in the future.

If you look at the first page of this thread, you will find one limitation mentioned there: Which is rotated (landscape vs portrait) displays. madVR currently doesn't support rotated displays. I've had a quick look at how to make it work - and failed to find the cause of the problem. It's on my list of things to look at (again), but it's not high priority at the moment. Currently if you play a video on a rotated display, usually you'll just get a black screen, I think. Or maybe a crash if you let it run very long, not sure...
Oh, I saw that. I'll pass that to them.

As always, thanks for your help and the wonderful software.

[maxwell532]

I decided to do a quick and dirty Windows 8 installation without getting Windows updates etc. just to be sure. Only installed GPU driver, DirectX redistributables and MPCHC+MadVR+LAV.

Got similar results to my Windows 7 tests on page 871 with madNV12Test and 10-bit 1080p played without framedrops.

So dunno what was the cause to the performance decrease with my first Windows 8 install.

[Gyzome]

First of all I'd like to thank madshi for his great work. I've been using madVR for ages now and it's one of the few pieces of software I consider to be flawless. However, there is one feature that I'd like to request.

I have multiple computers with diverse specifications which I all use to watch media. They all have madVR on board. In order to get the most out of each, I've tuned their settings (mainly the upscaling algorithms) to appropriate levels for each computer, which guarantees a comfortable viewing experience for every kind of content—except it doesn't. Most video files have the same processing footprint on the renderer, regardless of their resolution, bitrate, or other properties. There is, however, one property that guarantees to substantially increase the load and that is frame rate. I can throw any 0-30fps file at any of my computers and it'll play without stuttering, but even the most gently encoded 60fps video is bound to have stuttering every few seconds. The obvious choice is to ease up the settings for high fps files but doing this every time is an incredible drag and using lower settings permanently is frankly a compromise of quality for 95% of the content I play, which is something that doesn't seem to stroke with madVR's aims.

I, and possibly/probably others, would like a solution that isn't necessarily simple to set up, but is effortless in use. Personally I would recommend dividing each scaling algorithm page (chroma up, image up, image down) into two or more tabs indicating the frame rate range at which they're to be used. This way I could for example use Jinc-3 at 0-30fps and SoftCubic-50 at 31-60fps, which would ameliorate this minor inconvenience with sacrificing a minimum of quality.


Thanks again for the time and effort you put into this fantastic piece of software. I hope my remark helps. It's not a bug so I'm not going to file it on the bug tracker (which doesn't seem to be used much anyway?).

[huhn]

is already on the "to do" list.

u can use the hotkeys for the time being.

madvr only cares about color-space resolution and fps. bitrate doesn't matter at all.

the bug tracker is new.
about 4 days old?

[cyberbeing]

So really the main thing you should be looking for in madNv12Test results is the upload speed.

@madshi
Which methods (if any) from this legacy test does madVR use to upload video textures nowadays?

Could you clarify this? Which upload method(s) from madNV12Test are still in use, and/or most relevant to current versions of madVR?

[QQ]

they are standard ffmpeg libraries,just like if you would use ffdshow to decode i guess

yes, but is there any reason to use those instead of, say, mpc-hc ones?

or let me ask another way:

what was the motivation for including decoding into already complex project?

[Dodgexander]

I'm a bit confused by this talk of converting input colourspaces to RGB. Supposing you want your computer to do all the work and feed the display with its native format?

I view on a computer monitor connected via DVI. So from this, I can assume the native display and input of the display will be RGB 0-255.

When I leave LAV video at is default settings, viewing 4:2:0 my OSD always shows:
http://aluminas.net/images/d24cbe.png.

Lav Video Settings:
http://aluminas.net/images/384bfe.png

Now I had a brief fiddle about, but presuming I want all the conversion done before it goes to my display. How should i configure lav?

[nevcairiel]

madVR does all the conversion for your display, you should not change LAVs config at all.

[konakona]

Now I had a brief fiddle about, but presuming I want all the conversion done before it goes to my display. How should i configure lav?

your graphic card feeds the screen with the full 4:4:4 rgb signal after madvr chroma upscaling,pc screens cant usually handle YCbCr input.

edit:i got ninja'd by nev

[Dodgexander]

madVR does all the conversion for your display, you should not change LAVs config at all.

your graphic card feeds the screen with the full 4:4:4 rgb signal after madvr chroma upscaling,pc screens cant usually handle YCbCr input.

edit:i got ninja'd by nev

Thanks guys! :D