madVR - high quality video renderer (GPU assisted)

[madshi]

The nvidia control panel Digital color format only has "RGB" and "YCbCr444" options. Does this suggest that many people with HTPCs are outputting broken color from madvr?
No. Why? RGB is exactly what you need. Just make sure the GPU outputs PC levels.

[rurifan]

No. Why? RGB is exactly what you need. Just make sure the GPU outputs PC levels.

Right, but only when the HDTV displays RGB 4:4:4. It is possible to make my HDTV display RGB 4:4:4 by using the semi-secret PC Mode option. But not all HDTVs have that option, and probably most madvr users don't even know about it. That's why I suggest that many madvr users may be unaware that madvr outputs color incompatible with their display.

Also, PC Mode usually prevents using any kind of fancy processing features such as frame interpolation etc. In my TV's case, it won't even activate PC Mode for 1080p@24hz. So it seems there are legitimate reasons to have the display in the "wrong" color mode.

Am I crazy? :) Maybe madvr should have an option to output compatible color.

I encourage other HTPC/HDTV users to double check their color display using my test pattern (http://i.imgur.com/gTsKnxk.png).

[ThurstonX]

I ran the RGB-via-HDMI test (referenced here (http://forums.guru3d.com/showpost.php?p=4557704&postcount=25)) and sure enough my LG LED LCD was not accepting "Full RGB" from my Radeon HD 6670. Setting the Input Label to PC changed that, and the above mentioned test passed. Now for the noob question:

Does it make a difference if I've got my TV set to TV levels or PC levels? I'm looking at a few files now that I've set it to PC levels. Most of the Advanced Controls are grayed out (Black Level and Gamma are active; set to Low and Medium, respectively). And obviously I changed madVR to "TV expects PC levels".

I've matched my other settings (Backlight, Contrast, et al.) to what I generally use when the TV is running without a label (thus, TV levels), and I'm hard-pressed to tell much difference, but I figured I'd ask the experts.

If it makes a difference, what would that be, and which is preferable?

I'll say this isn't really OT, as my main concern is optimal playback quality with madVR.

Thanks!

[QBhd]

On most TV's a "Low" black level is TV Levels... For LG I am pretty sure PC Levels is called "High" and from what I have read "Normal" is also another name often used by manufacturers. Maybe others can input what "Name" manufacturers use for their black level that is PC Levels (0-255)

QB

[dansrfe]

Is there a look-up table for conversion between prosumer HDTV menu terms to technical terms for black levels, gamma, etc.? I'm tired of having to play the guessing game when I help my friends set up their HDTVs and if I select the wrong option then it stifles the benefits of using madVR in the first place.

[dansrfe]

Ok, so I decided to try putting my second monitor in portrait orientation and madVR just freezes when I drag the MPC-HC window from my primary screen to the portrait one. As an aside, I don't know why madVR only shows one device in the devices list. Both of the monitors are identical but how can I specify different refresh rate settings for the portrait one? I would have to swap the height and width resolutions in the refresh rates list right? Also, how can I apply different 3DLUTs to both monitors? I tried to add the device within madVR but that doesn't work with the second display either.

Here is the log file in which I start the video on the primary landscape monitor and drag it over to the secondary portrait monitor where the video freezes and after a couple of seconds I close MPC-HC.

http://www.sendspace.com/file/0occwa

[nevcairiel]

Ok, so I decided to try putting my second monitor in portrait orientation and madVR just freezes when I drag the MPC-HC window from my primary screen to the portrait one.

This is a known issue with madVR, rotated screens are not supported.

[pie1394]

Im in the market replace my 5770. It needs to be future proof, in that it has to be able to output @ 4K with Jinc 4 AR for chroma and image. Which would you guys advise me? Thanks!

HD7970 (2048 GCN cores) should have been already capable of handling 2K 60fps progressive source with Jin3_AR to 4K@60Hz output... No need to say about the GK110-based card, which contains more than 2300 CUDA cores...

According to my tests, it has been capable of processing the very mem/calculation demanding case like 1440x1080 to 1920x1080 with Jinc3_AR at 120 fps speed + concurrent 1440x1080i at 60 fps vector-adaptive deinterlace.

In addition to computation power / price factors (when I bought mine), the reason I changed from nVidia to AMD is the support of 10-bit color depth, 4K@30Hz with HDMI, 4K@60Hz with DP 1.2. Yet the non-dithered 10-bit D3D D3DFMT_A2R10G10B10 output mode only works with FSE. It is also only supported with newer top-end AMD HDx700~x900 and nVidia Quadro product line. So it does not seem a very high-priority feature in madshi's plan. Yet I still wish madshi could have some extra free time to do it. :D

The HDMI output on nVidia GTX6xx does not allow 4K resolution -- It supports up to 2560x1600. This restriction has been eliminated on GTX760/770/780/Titan.

[Asmodian]

When playing 4096x2160p@59.94fps (Reclock to 60.000Hz) resized to 2560x1440 for display with Jinc3AR+Jinc3AR, Smooth Motion Off, FSE, and Windows 8 my Titan is at ~95% usage @836MHz, ~86% @1006Mhz. If I use an external 3DLUT instead of selecting "this display is already calibrated" I get dropped frames when at 836Mhz and ~90% usage if @1006Mhz.

Source and destination resolution both seem to slow down Jinc a lot but a native 2160p display would require much less GPU power.

Edit: Sorry, of course I am not using Jinc3 for luma, Jinc is disabled when downscaling... Catmull-Rom+AR+LL was used for image. :o

You need something really high end for Jinc3 chroma 2k@60p. I wonder how much faster 1080 to 2160 is vs 1080 to 1440 for the chroma resize? I remember 2x resizes have an optimization and are much faster than other ratios?

[rurifan]

No. Why? RGB is exactly what you need. Just make sure the GPU outputs PC levels.

My previous response seems to have gotten lost in moderator approval land.

The gist was that not everybody with an HDTV can run in a "PC Mode" that displays RGB 4:4:4. At least it's worth being aware of in our quality-conscious crowd. :)

It's not clear to me if this is something madvr could account for. Are HDTVs converting the input RGB 4:4:4 to YCbCr 4:2:2, or actually misinterpreting the input? I suppose in the former case the end result should be close to correct, just converted through some addtional lossy steps.

[Asmodian]

Sometimes it is the video card that does the conversion, somehow the TV convinces the video card it wants YUV 4:2:2. Maybe the EDID?

There are also TVs that accept 4:4:4 but use internal processing in 4:2:2 so they fail test patterns.

Edit: I think MadVR shouldn't try to do anything about it. I believe the drivers want to start with RGB and even convert to it first if given something else?

[pie1394]

When playing 4096x2160p@59.94fps (Reclock to 60.000Hz) resized to 2560x1440 for display with Jinc3AR+Jinc3AR.

Is your source video content 4:4:4 or 4:2:2 or regular 4:2:0?

For your case:

4:4:4 -- U/V = 4096x2160 (H/V-dir will be both down-scaling)

4:2:2 -- U/V = 2048x2160 (H-dir is up-scaling, V-dir is down-scaling... I am curious which scaling algorithm will be selected by madVR -- chroma upscaling or image downscaling setting? :p)

4:2:0 -- U/V = 2048x1080 (H/V-dir will be both up-scaling)

If I don't remember it wrong, 4:4:4 contents are often used on "original source". 4:2:2 is used on high-end digi-cam. 4:2:0 video is everywhere -- including DTV / BD / DVD. But a different sub-sampling mode 4:1:0 is used on DV and WMV3 complex profile.

[nevcairiel]

4:2:2 -- U/V = 2048x2160 (H-dir is up-scaling, V-dir is down-scaling... I am curious which scaling algorithm will be selected by madVR -- chroma upscaling or image downscaling setting? :p)


madVR always first upscales chroma to 4:4:4 and then applys image scaling, AFAIK.

[Asmodian]

It is a 4:2:0 source.

[madshi]

It's not clear to me if this is something madvr could account for. Are HDTVs converting the input RGB 4:4:4 to YCbCr 4:2:2, or actually misinterpreting the input? I suppose in the former case the end result should be close to correct, just converted through some addtional lossy steps.
If they misinterpreted the input the image would be all screwed up. Usually they internally convert the incoming RGB 4:4:4 to YCbCr 4:2:2 because some internal chip is limited to 4:2:2. This is a limitation which has nothing to do with HTPCs or madVR. Whichever source device you use, these displays are always converting anything higher than 4:2:2 down to 4:2:2. Not a big problem for 4:2:0 sources (only a bit quality goes lost), but quite annoying for true 4:4:4 sources, like Xbox/PS3/PC games/desktop. PC mode sometimes helps.

Sometimes it is the video card that does the conversion, somehow the TV convinces the video card it wants YUV 4:2:2.
This is usually not something that happens secretly. If the display reports that it doesn't support RGB *at all*, the GPU control panel should not offer that option. So it's very visible in the GPU control panel that RGB won't work. However, I've yet to see a display which doesn't accept RGB.

madVR always first upscales chroma to 4:4:4 and then applys image scaling, AFAIK.
Correct.

[leeperry]

Are HDTVs converting the input RGB 4:4:4 to YCbCr 4:2:2, or actually misinterpreting the input? I suppose in the former case the end result should be close to correct, just converted through some addtional lossy steps.
Sammy TV's run in low-latency "game mode" when you rename the active HDMI input to "PC", which pretty much disables every post-processing option.....it's a possibility that some of those TV's would do their processing tricks in 4:2:2.

[chano22000]

Hello,

Sorry for this very noob questions but I did not find any answer neither on doom9 nor the web.

Q1 : Queue utilizations are reported by the OSD as a-b/s. I suppose s is the queue size but what is the precise meaning of a and b ?

Q2 : Which settings (CPU/GPU queue sizes, frames presented in advance ...?) impacts which queue sizes and how (in exclusive mode)?

Q3 : If "delay playback start until render queue is full" is disabled, does it mean that the render queue might never be (nearly) full even if there is sufficient CPU / GPU power to catch up ?

Thanks you for kindly sharing your knowledge !

[Mangix]

The nvidia control panel Digital color format only has "RGB" and "YCbCr444" options. Does this suggest that many people with HTPCs are outputting broken color from madvr?

I made a very simple test pattern (http://i.imgur.com/gTsKnxk.png).

If you are getting crushed colors on nvidia hardware, madshi has an nvidia levels tweaker that he includes with madVR.

The better option for me though is to use CRU and uncheck "Include extension block". See: http://www.monitortests.com/forum/Thread-Custom-Resolution-Utility-CRU

[zero9999]

which project is the right place to report madvr crashes with xySubFilter being used?
http://i.imagebanana.com/img/trdf7zq9/Untitled1.png

[6233638]

I wonder how much faster 1080 to 2160 is vs 1080 to 1440 for the chroma resize? I remember 2x resizes have an optimization and are much faster than other ratios?Chroma is always scaled to the luma resolution, and then the image is scaled using your luma settings, so chroma performance shouldn't change no matter what resolution you are scaling the video to.

Downsampling 4K using Catmull-Rom, Linear Light scaling, and Anti-ringing may give the best quality, but it's very demanding.

What is GPU load like on your system if you disable linear light scaling and anti-ringing?
Q1 : Queue utilizations are reported by the OSD as a-b/s. I suppose s is the queue size but what is the precise meaning of a and b ?The queue size should be mostly unchanging. If there's a large range being displayed, then it's likely that your GPU is struggling and you might be best to reduce the queue sizes.

Unless you have a very slow CPU/GPU, where the numbers fluctuate a lot and it can be helpful to have a large number of frames queued up, I generally find it best to adjust the queue sizes until they are full at all times.

If they are larger than this and your CPU/GPU can't fill them, they will be stuck at 100% usage throughout video playback.

Q2 : Which settings (CPU/GPU queue sizes, frames presented in advance ...?) impacts which queue sizes and how (in exclusive mode)?CPU queue size = Decoder Queue (perhaps this should be renamed now that we have CUVID/DXVA/QuickSync for decoding)
GPU queue size = Upload/Render Queue size
Number of backbuffers (windowed) or number of frames presented in advance (exclusive mode) should be self explanatory.

Q3 : If "delay playback start until render queue is full" is disabled, does it mean that the render queue might never be (nearly) full even if there is sufficient CPU / GPU power to catch up ?It should fill up within a second or so if your GPU has enough power. This option should just prevent possible dropped frames at the start of playback or after seeking, but it causes the video to pause until the buffers are full.
With high buffer/queue sizes, you probably want to disable this, unless you don't mind waiting.


Right, but only when the HDTV displays RGB 4:4:4. It is possible to make my HDTV display RGB 4:4:4 by using the semi-secret PC Mode option. But not all HDTVs have that option, and probably most madvr users don't even know about it. That's why I suggest that many madvr users may be unaware that madvr outputs color incompatible with their display.

Also, PC Mode usually prevents using any kind of fancy processing features such as frame interpolation etc. In my TV's case, it won't even activate PC Mode for 1080p@24hz. So it seems there are legitimate reasons to have the display in the "wrong" color mode.

Am I crazy? :) Maybe madvr should have an option to output compatible color.RGB is always full resolution. YCC can be 4:2:2 or 4:4:4.
As you say, many HDTVs process the image in 4:2:2 though - but outputting a 4:4:4 signal is not likely to make a difference to that.

It's rare that a set will display full chroma resolution with a 4:4:4 signal and not display full chroma resolution with an RGB input.
If anything, it's typical that you will only get full chroma resolution with an RGB input and the display set to PC or Game mode. (which often disables many interpolation or color management options, as they operate in 4:2:2)

As always, there are exceptions. I think there are a small number of Panasonic plasmas which display full chroma resolution with a 4:4:4 input, but not with an RGB input.
In that case, it may be worthwhile to set your graphics card to output a 4:4:4 signal rather than RGB.


However, computers are rendering everything in RGB internally. To output YCC, the RGB signal is converted to YCC by the graphics card, which is a potentially lossy process that could introduce banding into the image. I would only use YCC if RGB is not an option.


And if your display is not rendering full chroma resolution, there's no point in using anything above Lanczos 3 - likely no point in going above Bicubic 75 even, so you may as well reduce those settings to increase performance.

[cyberbeing]

which project is the right place to report madvr crashes with xySubFilter being used?

Probably neither, unless you are testing your own VS2010 build from latest git, had XySubFilter PDB debug symbols, and saved the entire madVR crash report text file. Quite a number of commits where reverted a couple days ago and re-pushed with fixes, so a crash from a older build isn't particularly useful. None the less, what were you doing when the crash occurred? If it is easily reproducible, you may as well go ahead and report it on madshi's tracker.

[nevcairiel]

Whenever you start "officially" testing, you should really consider a thread for that somewhere, or it'll keep popping up all over the place. :)

[QBhd]

And if your display is not rendering full chroma resolution, there's no point in using anything above Lanczos 3 - likely no point in going above Bicubic 75 even, so you may as well reduce those settings to increase performance.

And what about downscaling almost everything to 1024x768? I use this resolution on my 720p LG Plasma (circa 2007) since it's the physical resolution of the pixels... and I can also run it at 60Hz AND 72Hz. Should I be wasting resources on Jinc3 AR? My 5770 handles everything I watch and the most demanding has been 1080i sources that are "mostly film" hybrids, but still play back silky smooth. Is usuing Jinc3 AR for chroma going to hurt quality or just be a waste? If just a waste, I don't mind since I rent and don't pay electricity :P

BTW, I am fairly certain that the plasma is only outputting 4:2:2, but it's so damned hard to tell.

QB

[zero9999]

None the less, what were you doing when the crash occurred? If it is easily reproducible, you may as well go ahead and report it on madshi's tracker.

easily reproducible (r470; fine in xy-vsfilter r214), happened on a ASS line with asterisks near an override tag (put there by a gradient script as a hack to undo its effects later). so it's probably not related to madvr at all, is it?

{\c&HEAE7E8&\fs35\blur0.6\frz44.04\pos(882,440)} {*\frz42.392}R{*\frz40.743}e{*\frz39.095}s{*\frz37.447}c{*\frz35.798}u{*\frz34.15}e {*\frz30.853}S{*\frz29.205}q{*\frz27.557}u{*\frz25.908}a{\frz24.26}d

EDIT: nevermind, seems fixed in r486

[6233638]

And what about downscaling almost everything to 1024x768?It probably doesn't make much difference when downscaling. It certainly won't hurt image quality to use Jinc 3 AR when it's not necessary - only performance.

BTW, I am fairly certain that the plasma is only outputting 4:2:2, but it's so damned hard to tell.http://www.abload.de/img/chroma-smallp1ihz.png (http://www.abload.de/img/chromabucxa.png)

On a display which supports full chroma resolution, both halves of this image should be roughly the same magenta color.

On a display which only supports 4:2:2, the right half of the image will be a much darker color. The difference should be obvious, not subtle.

This pattern will only work when it is shown at 100% size on a 1:1 mapped display. Click the image for a full 1920x1080 pattern.

EDIT: If you have a Pioneer Kuro, they appear to use 4:2:0 subsampling, rather than 4:2:2, in which case both sides of the pattern will also appear to be the same. I don't know of any other displays which use 4:2:0

[TheElix]

On a display which supports full chroma resolution, both halves of this image should be roughly the same magenta color.

On a display which only supports 4:2:2, the right half of the image will be a much darker color. The difference should be obvious, not subtle.I am watching this on a CRT monitor (which is RGB) and the right side is much darker. Is it because of aperture grille?

[6233638]

I am watching this on a CRT monitor (which is RGB) and the right side is much darker. Is it because of aperture grille?A CRT will not be converting to 4:2:2 (unless it's a television, where I suppose it might?)
I don't know that you will get anything meaningful out of using that pattern on a CRT. I'd only use it on a display with fixed pixels.

It probably looks darker due to the softer image of a CRT blurring things together a bit.

[fairchild]

On a display which only supports 4:2:2, the right half of the image will be a much darker color. The difference should be obvious, not subtle.

This pattern will only work when it is shown at 100% size on a 1:1 mapped display. Click the image for a full 1920x1080 pattern.

On my 2013 Panasonic S60 plasma, I can get that pattern to display properly and have full chroma resolution by enabling 1080p pixel direct mode and tell the display to expect nonstandard (0-255 range) and then set the video card to output full RGB 0-255. There is a downside though...

If I do all the above to get full chroma resolution I end up losing the standard color gamut (which is what I use to calibrate display to Rec.709 D65) and it defaults to the native color gamut which is an expanded gamut. As a consequence the color accuracy of the calibration suffers. (I could calibrate the colors to get as close as possible but then my other sources will suffer...)

Which should provide a better overall PQ when calibrated, the full chroma resolution or the more accurate colors/grayscale? (strictly speaking video quality and blu-ray quality)

[truexfan81]

It probably doesn't make much difference when downscaling. It certainly won't hurt image quality to use Jinc 3 AR when it's not necessary - only performance.

http://www.abload.de/img/chroma-smallp1ihz.png (http://www.abload.de/img/chromabucxa.png)

On a display which supports full chroma resolution, both halves of this image should be roughly the same magenta color.

On a display which only supports 4:2:2, the right half of the image will be a much darker color. The difference should be obvious, not subtle.

This pattern will only work when it is shown at 100% size on a 1:1 mapped display. Click the image for a full 1920x1080 pattern.

yay, they are the same color on my video monitor :D

[6233638]

Which should provide a better overall PQ when calibrated, the full chroma resolution or the more accurate colors/grayscale? (strictly speaking video quality and blu-ray quality)For desktop computer use or playing video games, I would not be happy with anything less than RGB/4:4:4 chroma. Those are RGB native sources, so the loss of chroma resolution is very obvious to me.

If it means better color accuracy, motion handling, or anything else on your set though, I would absolutely sacrifice full chroma resolution with video playback.

Video is natively 4:2:0, and while madVR's upscaling is arguably better than 4:2:2 resolution permits, the difference is not nearly as dramatic.

[nevcairiel]

My TV also loses some of its features when i run it in game mode, so i don't bother and let it run in Cinema mode instead, which means only 4:2:2 but instead i get its clear motion features, which do help a lot on 24p content. :)

[Andy o]

yay, they are the same color on my video monitor :D
They are the same color on my Pioneer Kuro KRP-500 as well, but it's displaying 4:2:0 (http://www.avsforum.com/t/1162100/the-official-pioneer-9g-north-american-krp-500m-krp-600m-owners-discussion-pt-ii/6270#post_20822997).

[6233638]

They are the same color on my Pioneer Kuro KRP-500 as well, but it's displaying 4:2:0 (http://www.avsforum.com/t/1162100/the-official-pioneer-9g-north-american-krp-500m-krp-600m-owners-discussion-pt-ii/6270#post_20822997).Wow, I remembered problems resolving detail on the Kuros, but I didn't remember them downsampling chroma to 4:2:0. If it's really doing that, then it will be blurring both halves of the image rather than just the right hand side. (both are probably a dark purple rather than magenta-ish?)

For anyone else wanting to check that their display is not downsampling to 4:2:0, get up close and you should be able to see that the left half is actually made up of alternating red and blue lines.
4:2:2 blurs horizontally, so they should be unaffected. The checkerboard pattern on the right is affected with 4:2:2, but not 4:4:4, which is why it should appear darker if your set is processing chroma in 4:2:2.

A display downsampling chroma to 4:2:0 is a very rare case though - typically it will be 4:2:2 or 4:4:4.

It would be interesting to go back and have a look over one of those sets again today. I had one when they came out, but ended up returning it for a number of different reasons.

[Asmodian]

Chroma is always scaled to the luma resolution, and then the image is scaled using your luma settings, so chroma performance shouldn't change no matter what resolution you are scaling the video to.

Downsampling 4K using Catmull-Rom, Linear Light scaling, and Anti-ringing may give the best quality, but it's very demanding.

What is GPU load like on your system if you disable linear light scaling and anti-ringing?

Good point.

Using Jinc3+AR chroma and only Catmull-Rom (no AR, LL, or 3DLUT) down-scaling I get a steady 75% GPU usage @1006MHz, 82% @836MHz.

Using Nearest Neighbor on GPU texture units for down-scaling I get 71% @1006MHz, and 77% @836MHz which should be a good approximation for native 2160p out. It is still not easy but there is more headroom.

[njfoses]

What are the pros/cons of disabling "use inverse telecine" in the nvidia control panel? With madvr now having its own version of ivtc should i just leave the option unchecked?

[Moragg]

Hi madshi,

Just posting to say thanks, and a quick question:

GPU usage isn't going above 57% yet I still get dropped frames. This is Jinc3+AR+SmoothMotion

Upping it to Jinc8 AR uses the whole 100%, so why can't it use 100% at Jinc3 to not drop frames?

Just SM uses 35%, Just Jinc3+AR uses 41%, so surely my card should be able to do both together?

I have a 7870 Tahiti LE, stock clocks. CPU is a 1055T (hex core 2.8GHz) if that is causing a bottleneck.

[ryrynz]

Moragg, is that with Fullscreen exclusive mode?
Also try Bicubic 75 for Chroma, Jinc 3 3taps for Chroma can be a bit overkill.

[Moragg]

Moragg, is that with Fullscreen exclusive mode?
Also try Bicubic 75 for Chroma, Jinc 3 3taps for Chroma can be a bit overkill.

I turned fullscreen exclusive off since I didn't like it.

It may be overkill, but the vast majority of what I watch is anime and that works perfectly with my settings.

But with this video usage should go up to ~80% - it staying at 57% is the problem that is causing dropped frames. I just wondered if this was a madvr problem or physical limitation of my GPU (though madvr used 100% with Jinc8).

[huhn]

It may be overkill, but the vast majority of what I watch is anime and that works perfectly with my settings.

it's not an over kill i use an 6770 with jinc 3 ar /jinc 3 ar.

it doesn't make sense that animes work and "not anime" not madvr doesn't care about the content. maybe 60i content or 480p?

make a screen from the osd control + j and post it here.

your cpu is not the problem!

[Moragg]

it doesn't make sense that animes work and "not anime" not madvr doesn't care about the content. maybe 60i content or 480p?

your cpu is not the problem!

It's more that anime is "easier" to scale.

OSD is attached
13566
48fps 1080p, I know that should be hard to upscale to 1440p and smooth motion.

I wouldn't mind if my card was running 100%, but it doesn't even reach 60% and madvr's dropping frames instead of using the other 40%.

[Soukyuu]

I had the same issue (dropped frames even though resources were free) and it turned out to be related to the CPU (decoding) queue. Try increasing it in madvr settings. Can't see your attachment since it's pending approval, but try keeping an eye on the queues and see if any drops to 0

[QBhd]

On a display which supports full chroma resolution, both halves of this image should be roughly the same magenta color.

On a display which only supports 4:2:2, the right half of the image will be a much darker color. The difference should be obvious, not subtle.

This pattern will only work when it is shown at 100% size on a 1:1 mapped display. Click the image for a full 1920x1080 pattern.

Well that did the trick! Thanks. I have done other tests that suggested it was not 4:4:4, but this test is very conclusive! The right side of the image is clearly darker, no if's and's or but's! Now I know 100% for sure, even though I was 99% sure before :)

The only other question I have about the test is... does the rectangular pixels affect the result?

QB

[mindbomb]

What are the pros/cons of disabling "use inverse telecine" in the nvidia control panel? With madvr now having its own version of ivtc should i just leave the option unchecked?

Leave it checked. The pro is that the nvidia one works automatically. The con is that if you are using madvr to change display modes (1080p23 for example), madvr won't know that ivtc is being done and will choose the wrong refresh rate.

Now, the madvr one works well with the display mode switching, but you have to manually enable it. Also, if you are using an intel atom or similiar, you will have issues with 1080i since madvr's ivtc runs on the cpu.

[Moragg]

I had the same issue (dropped frames even though resources were free) and it turned out to be related to the CPU (decoding) queue. Try increasing it in madvr settings. Can't see your attachment since it's pending approval, but try keeping an eye on the queues and see if any drops to 0

Okay, setting CPU queue to 128 helped - at one point the queue fell to 40/128 :scared:
All the points were playback would "hang" had this problem (though none that badly).

However, smooth motion doesn't make any difference to the CPU queue - I can scale up slightly from the original without issue but after a point I start dropping frames - this time with GPU running at 45%.

Running a low bitrate anime through with smooth motion Jinc3+AR:

1x speed: stable 23% GPU usage
2x speed: stable 47% GPU usage
4x speed: no dropped/delayed frames, but extremely unstable, peaking at 95%.

The problem video Jinc3+AR, 1x speed:
Original Size: extremely unstable between 2% and 37%, doesn't go anywhere in between.
Slightly larger: stable 39%
Larger: stable up to 44%
Even larger: frame drops, but still stable up to 56% (increases with size).

I'm very :confused: as to the problem.

[6233638]

Well that did the trick! Thanks. I have done other tests that suggested it was not 4:4:4, but this test is very conclusive! The right side of the image is clearly darker, no if's and's or but's! Now I know 100% for sure, even though I was 99% sure before :)I'm not sure if that was a typo, but if the right side is darker, then it's 4:2:2

Here's a couple of quick photos from my Sony TV for reference. (note: the camera colors are wrong, the pattern appears darker on my screen than it looks here)

PC/Game mode. (4:4:4)
http://www.abload.de/img/_dsc30695esoj.jpg

And Cinema mode. (4:2:2)
http://www.abload.de/img/_dsc30682ms0w.jpg

I threw the lens out of focus so you're only looking at the color and not the pixels.
When you are looking at the pattern normally, there is a slight difference between both halves in PC/Game mode (4:4:4) but that is to be expected.
The difference with 4:2:2 chroma should be an obvious one, as the photos illustrate.

The only other question I have about the test is... does the rectangular pixels affect the result?It shouldn't make a difference. Chroma resolution is related to image processing, not the pixel structure.


Okay, setting CPU queue to 128 helped - at one point the queue fell to 40/128 :scared:
All the points were playback would "hang" had this problem (though none that badly).Are you using software decoding, or hardware decoding? If you're using hardware decoding, what are you using?

Is deinterlacing active?
Are these 10-bit files? (requires software decoding, and some systems have difficulty with them apparently)

[njfoses]

Leave it checked. The pro is that the nvidia one works automatically. The con is that if you are using madvr to change display modes (1080p23 for example), madvr won't know that ivtc is being done and will choose the wrong refresh rate.

Now, the madvr one works well with the display mode switching, but you have to manually enable it. Also, if you are using an intel atom or similiar, you will have issues with 1080i since madvr's ivtc runs on the cpu.

I do use madvr to auto switch display modes. Should i still leave ivtc checked in the nvidia control panel then? I have an i7 cpu so cpu power is not an issue.

[dansrfe]

Is it normal for madVR to only detect one of my two connected and identical make/model monitors? I only see one device in madVR's device list...

[corporalgator]

Link on first page downloads .86.4. I was able to get .86.6 from the old versions link.

[pie1394]

It's more that anime is "easier" to scale.

OSD is attached
13566
48fps 1080p, I know that should be hard to upscale to 1440p and smooth motion.

I wouldn't mind if my card was running 100%, but it doesn't even reach 60% and madvr's dropping frames instead of using the other 40%.

What is your flush setting? If you turn off all of them, what rendering time does it show? With 120Hz output mode, it is obviously that the rendering time needs be less than 8.333 ms -- usually much safe with the value near 5 ~ 6ms.

How about the case with 60Hz output mode?

[mindbomb]

I do use madvr to auto switch display modes. Should i still leave ivtc checked in the nvidia control panel then? I have an i7 cpu so cpu power is not an issue.

if you are gonna be using madVR's ivtc, i think it is irrelevant whether you check it or not. Remember, the hotkey to enable it is ctrl+alt+shift+T, and it doesn't work with dxva native afaik.