madVR - high quality video renderer (GPU assisted)

[ikakun]

Originally Posted by baii

0.88.2 seem to broke next/skip file for me, tried default setting, both mpc-hc and pot player does it. revert back to 0.88.1 and it is fine.

Same here.

[SecurityBunny]

Originally Posted by ryrynz

Exactly. Set to 10 bit, avoid dithering to gain some performance if u want as chances are u won't see any benefit enabling it (tests required) and call it a day.
I think dithering looks a bit messy on a plasma less noise = cleaner/clearer picture.

Are you saying I could set the dithering algorithm to none if madVR display bitdepth is set to 10 bit? It won't cause any graphical downgrade in video playback and I can save the performance cost?

If dithering is recommended to always have, which setting produces the highest quality without grain?

[James Freeman]

Originally Posted by ryrynz

Overkill that costs you nothing. Disable dithering and gain performance.

Nonsense.
By overkill I meant to visual gradient but NOT for mathematics correctness; Meaning it should never be turned off.

Dithering is important, nothing falls exactly in place mathematically even in 10bit which is only 1024 places.
I use Ordered Dithering which takes nothing from the performance nor from the quality of dithering (imo).

[ryrynz]

Originally Posted by James Freeman

Nonsense.
By overkill I meant to visual gradient but NOT for mathematics correctness; Meaning it should never be turned off.

Sure, for correctness.. but that would be only able to be seen with ramps right? In real world content some displays may look "better" with it disabled, correctness be damned.

So if 10 bit "looks" like 8 bit + dithering then the point I'm trying to make here is that you can stand to attain the same PQ (perhaps higher) and also gain performance if you decide to disable dithering.
Alternatively of course one could keep dithering enabled, I was simply pointing out there was an upside to something you claimed was overkill (pointless?) not that enabling 10 bit was sufficient enough to warrant disabling dithering for every situation, I'm quite well aware of it's "importance"

BTW not everyone has negligible impact from dithering, If I disable it I have 8% reduction in GPU load on the HD4000 and it then clocks down to 350Mhz half the time thus saving power.

Originally Posted by SecurityBunny

Are you saying I could set the dithering algorithm to none if madVR display bitdepth is set to 10 bit? It won't cause any graphical downgrade in video playback and I can save the performance cost?

If dithering is recommended to always have, which setting produces the highest quality without grain?

You could, enabling 10 bit certainly makes dithering less important, but how much so would have to be tested.

[James Freeman]

Originally Posted by ryrynz

enabling 10 bit certainly makes dithering less important, but how much so would have to be tested.

You may be correct about 10bit being enough to not show gradients (mathematical errors) but only for 1000:1 (low) contrast ratio LCD displays of current generation and a super smooth gradient test pattern, anime/cartoons or super compressed video where the studio dithering is completely gone.
With HDR/Plasma (high contrast ratio display) gradients may be visible in 10bit also, so dithering is important.
Ordered Dithering is as good as Direct Compute (IMO) and has no performance hit for me.

I will test my plasma (Panasonic ST60) if it can accept 10bit signal, and whether the picture is "calmer" without dithering if it can.
Plasma display technology by itself is based on dithering, this is how the image is generated, which can be very distracting.
Maybe it can be even calmer without madVR sending another dithering...?

[ryrynz]

I'm quite interested in seeing actual results of 8 bit + dithering vs 10 bit, is madTestPatternSource working for you?

Originally Posted by James Freeman

I will test my plasma (Panasonic ST60) if it can accept 10bit signal, and whether the picture is "calmer" without dithering if it can.
Plasma display technology by itself is based on dithering, this is how the image is generated, which can be very distracting.
Maybe it can be even calmer without madVR sending another dithering...?

I tested my VT50 with 10 bit ouput using MPDN at the time and it looked better (certainly a lot calmer) without dithering, plasma's have enough noise IMO.
Certainly with 8 bit output based on madshi's posts I wouldn't disable it but once I enabled 10 bit that dithering went straight off. Stable pixels makes for a better picture.

[tyto]

Guys. I cant hold all this placebo.
How the heck do I take a proper screenshot? Using mpc builtin takes it before madVR. Or should I just take my time and use hotkeys?

[James Freeman]

Originally Posted by ryrynz

I'm quite interested in seeing actual results of 8 bit + dithering vs 10 bit, is madTestPatternSource working for you?

I tested my VT50 with 10 bit ouput using MPDN at the time and it looked better (certainly a lot calmer) without dithering,

I have tested with a grey gradient from AVS709HD and Debanding on High which creates a super smooth gradient in 16bit floating in madVR, which is similar to madTestPatternSource.
Tested with my Dell U2410 which supports 10bit (8bit+FRC).
I saw no difference between:
* 8bit+dithering(madVR)->U2410 8bit.
* 10bit+no dithering (madVR)->U2410 10bit (8bit+FRC).

Originally Posted by tyto

Guys. I cant hold all this placebo.
How the heck do I take a proper screenshot? Using mpc builtin takes it before madVR. Or should I just take my time and use hotkeys?

Printscreen.

[nevcairiel]

Originally Posted by tyto

Guys. I cant hold all this placebo.
How the heck do I take a proper screenshot? Using mpc builtin takes it before madVR. Or should I just take my time and use hotkeys?

There is a function in windows, called "print screen", there is even a key like that on your keyboard....

[Qaq]

Is DX11 (current implementation?) compatible with DXVA2 native?

First time I tried it with 10 bits (0.88.0) and got fullscreen static pattern. Seems like a video driver crash. Had to reboot the whole system. 10 bits worked fine with software decoding before that.

Next time I tried it with 8 bits (0.88.2) and got high CPU load on MPC-HC process (very old build though). Both LAV and madVR showed DXVA. Reverted madVR back to some old build and everything went normal.

Win7 x86, 7750, 13.12.

[ryrynz]

Originally Posted by tyto

How the heck do I take a proper screenshot?

Just make sure you aren't in FSE mode (when using printscreen) also 10 bit output can't be captured accurately so don't bother trying to make screenshot comparisons with it enabled.

Alt+I is another alternative in MPC that'll capture FSE mode just fine.

[iSunrise]

First things first, thanks a lot for keeping madVR up-to-date madshi! Will provide useful feedback when I have more time, currently an awful lot of work...

Originally Posted by ryrynz

I tested my VT50 with 10 bit ouput using MPDN at the time and it looked better (certainly a lot calmer) without dithering, plasma's have enough noise IMO.
Certainly with 8 bit output based on madshi's posts I wouldn't disable it but once I enabled 10 bit that dithering went straight off. Stable pixels makes for a better picture.

You really have to be careful about that.

Just because something is "calmer" and more stable doesn't automatically mean it's more accurate. That's why we (foremost madshi, but also everyone that helped tweaking the dithering algorithms in this very thread) wanted an extremely accurate but as noise-free as possible solution and I think we succeeded in that. We tested with up to 16bit inputs and only with madVRs final dithering we could get away with extremely high accurateness. madVR's dithering is so accurate that you can lower bitdepth to sub 6bit and barely notice anything if you don't watch very closely. This of course also depends on the dithering type you select, that's why madVR offers some very high-quality choices tweaked to preferences of CRTs, plasmas, LCDs and last but not least projectors.

Secondly, and rather more common, there are still a lot of sub 8bit displays out there and lots of displays that claim to support >8bit, but don't, let alone people that don't even understand "what to look for".

Sadly, my Eizo CG243W only supports 10bit input via DisplayPort and I am not even sure if DVI in general even supports it. From the top of my head I only recall Displayport and HDMI. So there's that.

Also, as long as we cannot output in 16bit (which is the final image madVR has available before it gets sent to the display), theoretically dithering will always be required.

Also, in cases when the display does something it shouldn't do (e.x. disabling the dithering when there's higher precision inputs than a certain threshold, e.x. due to a firmware bug or just because no one expected it), dithering is a good way of being safe in that regard.

[ryrynz]

Originally Posted by iSunrise

You really have to be careful about that.

Granted I was only using my eyes for testing but I was very close to the TV and honestly whatever mathematical accurateness near invisible to the eye was gained by having dithering
enabled did not amount to enough of a improvement vs the drawbacks of additional noise in the picture. I'll give it another look though with madVR when I can.

Originally Posted by James Freeman

I have tested with a grey gradient from AVS709HD and Debanding on High which creates a super smooth gradient in 16bit floating in madVR, which is similar to madTestPatternSource.
Tested with my Dell U2410 which supports 10bit (8bit+FRC).
I saw no difference between:
* 8bit+dithering(madVR)->U2410 8bit.
* 10bit+no dithering (madVR)->U2410 10bit (8bit+FRC).

So if 8 bit + dithering was super accurate enough.. and nobody (so far) can see a significant difference between 8 bit + dithering and 10 bit with no dithering why big deal about enabling dithering on 10 bit?
Are we at a point where we don't really need it enabled?

[toniash]

@madshi
any requirements for image and upscaling enhancements?
I'm in XP

[James Freeman]

Originally Posted by ryrynz

Are we at a point where we don't really need dithering enabled?

Again, at low contrast ratio display.
In this case yes, 8bit+dithering which visually can be equivalent to 10bit+ where our human eye is at its limit discerning such small light difference.

* Taken that the visual system is linear (which is most definitely not), just for practical example:
A monitor with peak white at 120[cd/m2] divided by 256 (8bit) will have 0.47[cd/m2] light difference between steps.
A monitor with peak white at 1000[cd/m2] (HDR) divided by 256 (8bit) will have 3.9[cd/m2] light difference between steps.
A monitor with peak white at 120[cd/m2] divided by 1024 (10bit) will have 0.11[cd/m2] light difference between steps.
A monitor with peak white at 1000[cd/m2] divided by 1024 (10bit) will have 0.98[cd/m2] light difference between steps.

[ryrynz]

madTestPatternSource is only working on x86 ATM, but I ran it on the plasma and the U2412. Basically 10 bit without dithering looks near identical to 8 bit without dithering.. so there's a big difference in the graduation of the bars
when switching from 10 bit to 10 bit + dithering.. so I guess that ends that.. I was expecting 10 bit to show some sort of difference.

Madshi, sometimes when switching dithering options or even the bit depth output on the HD4000 sometimes resulted in a black or partial black screen which when playback continued disappeared.

[James Freeman]

ryrynz I can't understand your post.
Can you please read it and fix it?

Originally Posted by toniash

any requirements for image and upscaling enhancements?

A video card.

[ryrynz]

Originally Posted by James Freeman

ryrynz I can't understand your post.
Can you please read it and fix it?

Reads fine, using smallramp I believe I can see a small improvement when switching between 8 and 10 bit but it's not hardly worth mentioning, I still see each individual bar in 10 bit until dithering is enabled.
I don't know how anyone can say 8 bit + dithering is anywhere near 10 bit.. as I'm not seeing that at all.

7 bit has a huge effect on the ramps, 8, 9 & 10 bit look almost exactly the same (I certainly couldn't tell the difference without looking up close)

Can someone else jump into an x86 version of MPC / BE and look at the smallramp & tell me if they see any difference between 8 and 10 bit output?

madshi, I'm getting strange results switching bit depths.. I switched to 8 bit from 7 bit (which didn't change bitdepth btw) and when I clicked okay on 8 bit it switched to 7 bit.. I had banding on smallramp
so it definitely wasn't 8 bit.

I duplicated it, switched to 7 bit no change in output, went and changed to 8 bit and boom.. banding all over the place. I also changed to 6 bit (updated as expected) went to 8 bit (did nothing) then went back to 6 bit and it changed to 8 bit. Testing on the HD 4000, 10.18.10.4176, Windows 7.

Now it's behaving properly.. (restarted the player) no idea what caused it, can't duplicate it again.

[James Freeman]

First sentence: looks near identical -> big difference in the graduation ???
Or: there's a big difference in the graduation -> 10 bit to 10 bit + dithering ???
???

My experience is 180 deg different.

I can clearly see the difference between UNDITHERED 8bit vs 10bit.
But Dithered 8bit vs 10bit has no difference.

It has to be compared on a ditherless display technology, aka don't compare on a plasma.
Use an LCD preferable with native 10bit panel or 8+FRC which is next best thing.

[toniash]

Originally Posted by James Freeman

A video card.

It doesn't work for me No effect