madVR - high quality video renderer (GPU assisted)

[huhn]

Y to RGB limited is lossless but this image still produces a ton of banding when doing that without dithering (using passthrough obviously).

something is not working correctly with it.

it's a good idea sadly it's HDR and the bitrate is very questionable.

[el Filou]

Originally Posted by Alexkral

The upscaling settings are not used in your config, try Bicubic60 to downscale, I doubt that you can tell the difference.

In that situation I would first try to lower the chroma upscaling quality to see if I can keep SSIM.

Originally Posted by Asmodian

Try enabling the trade quality for performance option "scale chroma separately, if it saves performance". In the right situation it can save a lot of performance, arguably without a quality hit (the output is slightly different but not necessarily lower quality).

This works only when chroma res = output def so for 4K->1080 OK, but not 1440 unfortunately.

[Alexkral]

Originally Posted by el Filou

In that situation I would first try to lower the chroma upscaling quality to see if I can keep SSIM

I'm a big fan of catrom (actually 60 is closer to the sweet spot to maximize sharpness without introducing ringing), in the past I did some comparisons with SSIM and the results were very image dependent. That's why I think that in this situation the most important thing is the chroma upscaling quality. Anyway, it's a matter of taste.

[Asmodian]

Originally Posted by el Filou

This works only when chroma res = output def so for 4K->1080 OK, but not 1440 unfortunately.

There are situations where it can scale chroma to 1440 while scaling 4K luma to 1440. Probably doesn't work with SSIM though.

[huhn]

works for 2160 to 1440p too. you just have to make sure the chroma scaler isn't a x2 algorithm even with SSIM. it will use the x2 algorithm if the target resolution is above ~1420p this is odd could be a bug but without performance measuring hard to tell.
if there is a luma < in the OSD it is working.

[tp4tissue]

Originally Posted by LordX2

This would seem to indicate to me that 10bit is being taken advantage of on my setup, and that dithering is working with 10bit video.

Is this the right conclusion to draw with the above scenario?

Congratulations, your projector does not suck.

as for the shadow bloat, does the base black become elevated when set to full rgb, or does the low-mid black elevate only.

So for example, if you just had a 0,0,0 background, is THAT also elevated on RGB full on the proj. If base black isn't elevated, it might just be the case that rgb full is causing the projector to emulate SRGB gamma, which has a brighter rampup near black.

[tp4tissue]

Originally Posted by Asmodian

For BT.2020 -> DCI-P3 or BT.2020/DCI-P3 -> BT.709 it seems to work perfectly. Using pure math to map BT.709 to a almost DCI-P3 display usually comes out a bit desaturated, which isn't great since BT.709 is already a lot smaller than the native gamut. The issue is that most displays don't quite hit the DCI-P3 primaries, not that the math is wrong.

Thx for the deets Asmodian.

Do you know if Madvr turns off its math-bending when using a rec2020 lut ,

For example, on the osd, if it says rec2020->dcip3 does that mean it's mathing ?

Because it says that regardless if i use a rec2020 lut or a p3 lut.

Is there another indicator to check what it's doing gamut wise.

[LordX2]

Originally Posted by Asmodian

If 10 bit looks the same or better use 10 bit, it is only when 10 bit is worse that you do not want to use it. The only reason I suggested using 8 bit was because you had worse black clipping with 10 bit but if it makes banding worse keep 10 bit.

1. Ok, so I tested 8bit vs 10bit with dithering enabled - they were both smooth, but the 10 bit was definitely smoother... went back and forth muuuuultiple times to make sure. So that is good! Obviously my PJ is processing 10bit. (also, the black levels are working fine now with my current settings over 10bit... don't know what I changed, but not complaining)

So I wanted to know WHY I had all the problems I did, and I had some posts answered on the PJ page on a different forum.

2. My PJ is limited to 13.5Gbps over its HDMI ports (vs 18Gbps). Stupid design, and one that has been bashed repeatedly (and of course fixed on the next release!).

3. All RGB content (or 444 full) is processed at 8bit on the PJ - which explains why having FULL settings through the chain (video card, madvr, and PJ) never worked right. The PJ was downgrading the 10bit signal to 8bit, and gave it that washed out look. So at least I know WHY that was happening now.

4. However, the PJ CAN process 10bit limited (4:2:0) signals up to 30Hz.

5. Even though my Nvidia is set to RGB FULL, Having MadVR set to Limited must be putting out the 10bit 4:2:0 video stream directly to the PJ in Fullscreen windowed mode - which would explain why the 10bit is being processed correctly by the PJ.

So I think I finally have a handle on why and what is actually happening. I think...

6. Final question though - from the conversation previously about color spaces and gamut etc - since my PJ is getting the full 10bit signal (madvr is doing the tone mapping), regardless of whether it is reporting bt.709 color space, and that my windows system is in SDR - am I theoretically getting all the colors that I possibly can to my PJ?

The picture looks astounding, so I don't care either way - just curious.

[tp4tissue]

what's the projector model again?

[Asmodian]

Originally Posted by tp4tissue

For example, on the osd, if it says rec2020->dcip3 does that mean it's mathing ?

Isn't that info about the source? I believe that is telling you that the metadata indicates it is DCI-P3 range video encoded as BT.2020. In that case it simply clips everything outside DCI-P3 when using DCI-P3 output, or does nothing to it for BT.2020 output.

Originally Posted by LordX2

4. However, the PJ CAN process 10bit limited (4:2:0) signals up to 30Hz.

10 bit is usually not worth using 4:2:0, at least on displays I have tested (not projectors). Converting to 4:2:0 after madVR is bad. Do you know what 4:2:0 means? Limited range and 4:2:0 are completely independent things.

4:2:0 loses a lot of spatial resolution for the color detail. Dithering to 8 bit preserves a lot more precision than simply resizing all the color data to 1/4 the number of pixels.

Can you set the GPU to output 8 bit YCbCr 444?

Originally Posted by LordX2

5. Even though my Nvidia is set to RGB FULL, Having MadVR set to Limited must be putting out the 10bit 4:2:0 video stream directly to the PJ in Fullscreen windowed mode - which would explain why the 10bit is being processed correctly by the PJ.

No, madVR cannot do that.. When madVR outputs limited range it is outputting limited range RGB. The GPU must not be set to output full range 10 bit RGB, since the projector cannot accept that. Limited and full range video takes the same amount of HDMI bandwidth. What is the GPU outputting?

Originally Posted by LordX2

So I think I finally have a handle on why and what is actually happening. I think...

Your described config doesn't seem possible given the bandwidth limitations of your projector.

Originally Posted by LordX2

6. Final question though - from the conversation previously about color spaces and gamut etc - since my PJ is getting the full 10bit signal (madvr is doing the tone mapping), regardless of whether it is reporting bt.709 color space, and that my windows system is in SDR - am I theoretically getting all the colors that I possibly can to my PJ?

Let's see if I can help make sense of this for you.

The bit depth does not control what colors are available. It simply doesn't so don't think of it that way. You get the same color with 8 or 10 bit. Higher bit depth makes getting smooth gradients easier but it does not change the colors displayed.

The gamut is what controls the colors able to be displayed. With the display set to BT.709 it cannot display any colors outside BT.709. DCI-P3 is larger than BT.709 so there might be colors in DCI-P3 content your projector cannot display when set to BT.709. Most content doesn't use many colors near 100% saturation so with a good conversion almost all the colors will be the same as they would be if the display was set to DCI-P3 with video tone mapped to DCI-P3. It is only colors outside the gamut of the projector that will change.

Originally Posted by LordX2

The picture looks astounding, so I don't care either way - just curious.

I have thought this about a completely wrong image before. Don't stop yet.

I suspect you can figure out something even better with your current system.

[LordX2]

I also FORCED the PJ into Limited mode. Forgot to mention that.

I wonder if that is what is helping me here - since I noticed a gigantic improvement when I first made that switch on the projector.

Perhaps the PJ can process RGB in limited mode at 30hz or lower with 10bit.

I mean something is happening with 10bit since I am noticing a difference on the test video. If it were all 8bit, it would (should) look the same either way I tested it.

Since I love the picture now - I may just leave it alone.

But would it be worth it to try and set the PJ from BT.709 to DCI-P3 if it will let me? And see if that makes any difference?

Projector model is: VPL-VW285es - Sony

[tp4tissue]

it makes no sense that you'd Not buy a colorimeter on a $5000 projector.

Buy the colorimeter, then we'll know exactly what's going on. Get the i1studio it's only $150. Don't buy a datacolor probe, they're busted.

I also find it difficult to grasp that such a projector can't do RGB Full. It's probably doing it correctly. it's just the gamma tracking is different in that mode.


The P3 gamut is that projector's native gamut, if any setting puts it in rec709 mode, the projector is using its internal CMS to clamp the colors. This operates independently of 8bit/10bit.

8bit / 10bit is a mapping system, it is not the color space, you can have 2bit rec2020 colors, or 100bit rec709 colors.

[Asmodian]

Originally Posted by LordX2

Perhaps the PJ can process RGB in limited mode at 30hz or lower with 10bit.

You should not need to use 4:2:0 unless you want to send 4K60. Even 4K30 12 bit RGB/444 is within the bandwidth limits. Oddly that projector seems to work in 8 bit internally with 4K60 anyway.

ProjectorCentral - Sony VPL-VW285ES 4K SXRD Projector review:

Unfortunately, this projector will down-res all 4K HDR 60 fps material to 8-bit regardless of chroma subsampling.

What is your GPU set to output?

Edit: For limited range video I prefer to use YCbCr 444 over RGB, limited is a normal concept for YCbCr while it is weird for RGB. Limited RGB uses 16-235 for all three planes while YCbCr uses 16-235 for Y but 16-240 for Cb and Cr. You would need to set madVR to full range when setting the GPU to YCbCr 444. 10 bit YCbCr 444 limited at 30 Hz or less should be a great option for that projector.

Originally Posted by LordX2

I mean something is happening with 10bit since I am noticing a difference on the test video. If it were all 8bit, it would (should) look the same either way I tested it.

Things can change between bit depths even if something isn't right.

Originally Posted by LordX2

Since I love the picture now - I may just leave it alone.

Hard to argue with that but something seems off to me.

Originally Posted by LordX2

But would it be worth it to try and set the PJ from BT.709 to DCI-P3 if it will let me? And see if that makes any difference?

Projector model is: VPL-VW285es - Sony

I wouldn't bother to be honest, not without a colorimeter to create a 3DLUT. Your projector has very good coverage of BT.709 but it is less good for wide gamut, most colors will be more accurate using the BT.709 mode on the projector.
I do not necessarily suggest copying their settings but they do show the gamut coverage pretty well.
Sony VPL-VW285ES Calibration Results

[LordX2]

Originally Posted by Asmodian

You should not need to use 4:2:0 unless you want to send 4K60. Even 4K30 12 bit RGB/444 is within the bandwidth limits. Oddly that projector seems to work in 8 bit internally with 4K60 anyway.

Yeah, I am aware of the 'above 30hz' cut off for my PJ - My system always boots to 23.976Hz, and I have MadVR set to change the Hz for whatever the video content is.. 99% of everything I watch is 23.976 or 25fps (planet earth etc). So I am always in the 'sweet spot' in that regard.

MadVR shows that the 4k HDR Remux videos I play are "P010, 10bit, 4:2:0 (DXVA11)" in the statistics. That is where I am getting the 4:2:0 from.

My actual settings are: Nvidia is RGB Full 12bit @ 23.976, MadVR output set to Limited, PJ set to Limited.

I think it is just ME saying things wrong that is leading to the confusion!

"Your projector has very good coverage of BT.709 but it is less good for wide gamut, most colors will be more accurate using the BT.709 mode on the projector."

I noticed this as well. Both on reviews and in person. When switching to HDR content, colors are crap. But in SDR they are vivid and rich.

So with madvr tone mapping to SDR, and outputting at 10bit, I think I am getting the best I can with the setup I have.

PS - Asmodian, wanna come over and give an in person review? It's only a 3000 mile drive and a 2 week quarantine! hahaha

[Asmodian]

Originally Posted by LordX2

MadVR shows that the 4k HDR Remux videos I play are "P010, 10bit, 4:2:0 (DXVA11)" in the statistics. That is where I am getting the 4:2:0 from.

My actual settings are: Nvidia is RGB Full 12bit @ 23.976, MadVR output set to Limited, PJ set to Limited.

I think it is just ME saying things wrong that is leading to the confusion!

Haha, that makes sense! You are correct, I think you found your ideal settings.

That line is telling you about the source, madVR received 10 bit 4:2:0 video in the P010 format from DXVA11 hardware decoding.

[tp4tissue]

My guess @ what's actually happening to lordX2 is that the projector goes into a pc mode of sorts when fed 444 rgb full, on many systems this disables the dynamic brightness/contrast image manipulation of the projector itself. and it may well clamp to rec709 instead of native p3 gamut. This could account for the loss in perceived saturation both from the contrast and gamut difference. Potentially also following a different gamma curve

For an image to be rendered too brightly due to levels mismatch, The PC has to send the wrong 16-235 levels while the Projector is expecting 0-255. If that is NOT the case, then the projector is doing full rgb correctly.

This is why you need to check the 0,0,0 black, if that black is about the same, while comparing both modes with all dynamic contrast/iris settings disabled, THAT MEANS the projector is doing the RGB Full correctly.

[tp4tissue]

@ Asmodian.

If Madvr is doing math to go from rec2020 to P3, then isn't it always better to feed it a rec2020 lut ?

Because that way, Madvr can do 1 less mathing of the gamut, we rely on argyll cms to math the gamut ONCE, And this side is more reliable because they are compressing using actual measurement data, WHEREAS, if Madvr does their math even once, it's Blind math, made with assumptions that don't always hold up on real color systems.

[Asmodian]

Well, not in practice, but theoretically you do want a custom 3DLUT for every source color space.

The difference between BT.2020 and DCI-P3 is a purely theoretical and an absolutely precise mathematical relationship, both sides are their theoretical color spaces. Assuming you are working with DCI-P3 encoded in BT.2020, the normal case. I haven't seen anything mastered in BT.2020 since there are no displays capable of displaying it. When mapping to a real display we need to make tradeoffs, and making tradeoffs twice is bad, but when mapping DCI-P3 range in BT.2020 to DCI-P3 you do not need to make any tradeoffs.

When I used software 3DLUTs I did make all the SDR LUTs, since that is so quick, but the difference was not measurable with an i1d3. I use passthrough for HDR so I haven't messed with HDR 3DLUTs. That said I would want a BT.709 and a DCI-P3 3DLUT, assuming my display's native gamut was very close to DCI-P3, and not worry about the pure math BT.2020 -> DCI-P3 conversion.

[huhn]

not sure if madVR does that or if it is worth any processing power but if you are tone mapping you can can do that to the gamut conversation too. so the tone mapping could do a better job if it knows that the gamut needs to be compressed too.

[tp4tissue]

Originally Posted by Asmodian

Well, not in practice, but theoretically you do want a custom 3DLUT for every source color space.

The difference between BT.2020 and DCI-P3 is a purely theoretical and an absolutely precise mathematical relationship, both sides are their theoretical color spaces. Assuming you are working with DCI-P3 encoded in BT.2020, the normal case. I haven't seen anything mastered in BT.2020 since there are no displays capable of displaying it. When mapping to a real display we need to make tradeoffs, and making tradeoffs twice is bad, but when mapping DCI-P3 range in BT.2020 to DCI-P3 you do not need to make any tradeoffs.

The definitions may be precise, but there are always idiosyncrasies in real systems that the math can't account for. My question was whether feeding madvr the rec2020 lut would save 1 transform.

Absent real offsets it's still very possible to swing wide even if by definition it should hit.

The data may be perfect but when you cast it, /problem.

I believe that movie Shadow is rec2020, but the bulk of the movie is greyscale, so idk what they were going for.