I disabled dithering on Apple silicon + Introducing Stillcolor macOS M1/M2/M3

markdotpeters5

markdotpeters5 OK so this is totally weird.

I now seem to be able to use MacOS Taho 26.4 on the same M4 Mac Mini with the same Omen 27Q monitors without any real problems. Previously this setup would start giving me dizziness (dithering symptoms) in a few minutes.

Using Stillcolor and betterdisplay at 1080p with HiDPI… SDR sRGB FULL (if it goes to limited it starts to hurt again).

So what's changed? Has anyone seen 26.4 bring any benefits, or is it me that's changed?

Donux

markdotpeters5 Can you please provide a bit more context to it:

1) Have you switched just recently and discovered you feel ok, but been using for relatively short time to give reasonable feedback? (It is common to feel good after being away from machine longer time, yet to find later, this was placebo)
2) Have you just recently updated and the update is clearly a primary cause of change?
3) What is your mac setup - icc profile, bit output, dither flags, any other workarounds
4) What is your setup on the monitor? - Color mode, any special settings you have applied. What cable are you using?

Thanks

AshX

markdotpeters5

Major firmware update with Tahoe 26.4. If the Neo I had didn’t cause a seizure in 26.3.2 I was going to try to update to 26.4.

Apple is changing the firmware almost every point release. The display engine, according to non-Apple professionals in the display space, is being altered every point release and has been going back to Big Sur. It was apparently especially problematic on Ventura, for a lot of video/photo oriented professionals’ workflows (breaking hardware compatibility, for example).

Someone posted on Reddit iOS 26.4 made their iPhone 17e unusable and they had to DFU restore to iOS 26.3.1.

I’ve been checking periodically since the 17 series was released and initially found 26.0 bad, then 26.1 seemed usable, 26.2 seemed the same, 26.3 has been bad, and I’ve not yet tested 26.4. So it’s likely they’re tweaking the display engine and possibly the dithering algorithm and for some of us it is influencing things in a positive or negative way.

Firmware updates for your reference:

And FYI, if you perform a security update for say an older OS like Sequoia 15.7.2, you will receive the firmware update that is currently shipped with the latest version of MacOS at the time of that security or point versions release. I have confirmed this with several users who are on Tahoe firmware despite never installing Tahoe. The only way to avoid this is a DFU Restore with a second Mac using an IPSW file from a point release from the era in which that OS was the main release, and therefore you will receive the proper, older firmware for your Mac.

markdotpeters5

Donux

1 - I have a Mac in the house but don't often use/try it since it's very unconfortable. I picked it up the other day to try on my '8-bit only' monitor and it was useable
2 - I'd upgraded to Tahoe since last using it in anger so quite possibly
3 - icc profile doesn't seem to matter but I'm using sRGB IEC61966-2.1
4 - Nothing special, just reduce the contrast and brightness because that bothers me on any machine. It's OK on HDMI and displayport cables.

markdotpeters5

AshX on the one hand, it's good that things are improved, but it just means you have to be VERY careful going forward if you have managed to get a useable system!

Also that absolutely explains what happened with my M1 Air was became totally unusable when I installed Sequoia even after I reverted back to Sonoma

Donux

markdotpeters5 Allright, so nothing special there. I think on mac keeping latest OS is a good approach. Downgrading gives short pracebo feel, but it all comes back the same just like any other OS. In fact I have been looking at this trend of "going back", it has been there for 10, 20 and more years. I would say 80 percent of cases, people want to go back out of psychological reasons, not real somatic symthoms. At least this is how I feel about mac os.

AshX

markdotpeters5

Yeah. At least you can easily block updates unlike with Windows. Just be sure to kill the new automatic background security updates they started doing again.

I’m hopeful for you! I’m taking a break for a few weeks after that last seizure using the Neo. Been awhile since that’s happened with a MacBook. Please keep us updated on how it goes.

markdotpeters5

AshX I will do, and so sorry you had to go through that. These people should be ashamed of themselves!

I'm not moving to the Mac as my main machine, that'll still be Windows 10 for now, but i'll keep playing with it to see if I can get it to 100% comfort (mainly the scaling I hate in MacOS!)

Newscreensarehorrible

TCON dithering and FRC need to have accessibility toggles with many types of dithering modes, these internal displays are extremely uncomfortable. Apple can do this easily, but they don't care. Mac OS UI elements are extremely uncomfortable same with Windows 11, they need to be as comfortable as Windows 10.

https://forums.macrumors.com/threads/recording-artist-cant-use-new-apple-devices.2476464/

Donux

Newscreensarehorrible Yes of course, they should do it. Apple designs its own TCON, its not like thay are buying panels from suppliers like we can do, they give full instructions what and how to produce IMO.

Newscreensarehorrible

There seems to be 3 forms of dithering happening at the same time on MacBook displays all the way back to the first Retina MacBook in 2012. Mac OS temporal dithering for UI elements and OS to the P3 color gamut, TCON and FRC dithering in the display and GPU dithering, this is an observation of mine. We simply need Apple to address this and provide a solution fast, MacBook's are ruined by severe eyestrain..

Donux

Newscreensarehorrible Yes please. And if they are struggling, all they need to do is to reverse engineer UHD620, UHD630 models. Thats it. No need to reinvent the wheel. Just take blueprint from these models/chips, jack them up with extra cores, and that is it. Bells and whistles could be sold separately in the store for brave spirits with big eyes made from glass.

xenu

Newscreensarehorrible

Thinking of external monitors, I wonder if someone could build a circuit that averages adjacent frames so that all this dithering could be eliminated e.g. a 120Hz HDMI signal goes into the circuit, it processes it, and what comes out is HDMI again but is
1) effectively 30Hz because every 4 adjacent frames are averaged and smoothed along the time axis.
2) reduced to 6 bit so that a monitor's FRC has nothing to do.
Or perhaps what is needed is an HDMI-to-VGA converter followed by a VGA-to-HDMI converter.

ryans

Has anyone tried this on an M5 (external display)?

photon78s

ryans

~~Any particular M5 model?~~

Maybe at some point the 13 inch silver m5 air base model as suggested here:

https://ledstrain.org/d/4006-new-macbook-neo-a-new-hope-for-eye-strain-community/187

simplex

xenu

FGPA plate

Cuttin off low bytes is enough to disable GPU's pixel walk in -2…+2 values range

xenu

simplex
Here's a company that sells both HDMI to VGA and VGA to HDMI. With both in series, no need for an FPGA.
https://www.amazon.com/Benfei-Input-Adapter-Support-Resolution/dp/B075GZ8DX7?
According to ChatGPT: "VGA supports 18 bits of color depth (6 + 6 + 6), which allows for 262,144 possible colors (2⁶ × 2⁶ × 2⁶).".

WhisperingWind

I have found many methods related to dithering in the scaler driver: AppleM2ScalerCSCDriver.

AppleM2ScalerCSCDriver is the driver for the M2 Scaler (scaler + Color Space Converter). This is a separate block in the Apple Silicon SoC that sits between the GPU and the TCON.

GPU -> DCP -> M2 Scaler (CSC) -> TCON -> panel

In AppleM2ScalerCSCDriver, I discovered several classes:

DitherControl
DitherControlMSR
DitherControlMSR4, MSR7, MSR8.

And a couple of interesting methods:

DitherControlMSR8::setDither_gatedContext(M2ScalerCSCRequest*)
DitherControlMSR8::isDitherUseful(M2ScalerCSCRequest*)

What MSRX means is also unclear for now - versions of registers?

A probe placed on isDitherUseful does not trigger:

sudo dtrace -n 'fbt::*isDitherUseful*:entry { trace("isDitherUseful called"); }'

So for some reason, the method is not being called by anyone. Perhaps it is called once during OS startup? But I cannot place a probe at such an early boot stage.

On my M1, this driver is loaded:

kmutil showloaded | grep com.apple.driver.AppleM2ScalerCSCDriver
94 4 0xfffffe0007430280 0x372ad 0x372ad com.apple.driver.AppleM2ScalerCSCDriver (265.0.0) 08FBC36A-1F8D-3B82-9210-05A7910BAACA <41 35 27 7 6 5 4 3 1>

Also, there is an interesting line in the driver:

"Quadra Bayer: Setting Dither Clamp registers to max \n"

This tells us that there are hardware registers for dithering. "Clamp" might control the dithering amplitude, and this log line could mean that the maximum possible amplitude is set. But I believe it's not FRC, it's spatial dithering.

So we are seeing hardware-level dithering control.

In AppleM2ScalerCSCDriver, there is a IOKit registry property: EnableFiltersNoRewriteMode. By default, it is False. I don't know what it means. But its value can be changed.

Update

As soon as I started watching a video on YouTube, calls to isDitherUseful and setDither_gatedContext started appearing: isDitherUseful returned zero, meaning dithering is not being applied in this case. But when I played the video in QuickTime, isDitherUseful started returning one.

The client of AppleM2ScalerCSCDriver is not only WebKit/QuickTime, but also WindowServer. WindowServer is the window compositor - it handles the composition of the GUI interface. But probes showed that there wasn't a single call from WindowServer the entire time - only from video players.

I need to take a closer look at the code of the MSR classes.

xenu

WhisperingWind "Quadra Bayer: Setting Dither Clamp registers to max \n"

This appears to have to do with the ISP not the DCP. It seems Quadra Bayer is synonymous with quad Bayer i.e. the 2x2 subpixels that make up the Bayer image sensor mosaic pattern. Setting the dither clamp to high allegedly ensures the raw input data is passed in without modification, which means temporal dithering is used to smooth individual pixels coming from an image sensor.

WhisperingWind

The scaler uses spatial dithering. It dithers a specific surface, not the entire screen. For instance, when video is playing on screen, it exists as a separate IOSurface instance, and dithering is applied only to that surface.

However, the scaler doesn't apply dithering for every video playback case. In many instances, it reports that dithering is unnecessary. For example, when playing YouTube on my M1, the scaler actively reported that dithering wasn't needed (it only appears in Picture-in-Picture mode, but there's heavy downscaling there and that's normal).

The numbers in the DitherControlMSRX class names indicate dithering algorithm versions - from the simplest in MSR4 to the more complex in MSR23. On M1, only DitherControlMSR, DitherControlMSR7, and DitherControlMSR8 are available.

The scaler driver receives calls from video players, and although WindowServer is also registered as a client, throughout all my testing I never saw a single call from it.

Nothing interesting for us here.

WhisperingWind

xenu

Interesting point! I think we might be talking about different components - you're talking about the ISP and camera, I'm talking about the display pipeline. "Quadra Bayer" might be the same name used for different purposes in different hardware blocks. According to the Asahi Linux firmware overview, the ISP is a separate coprocessor, while the scaler appears to be part of the DCP.

On a related note, looking at AppleM2ScalerCSCDriver::start(), the log strings like "[IOSA][Boot ] MSR Driver Waiting for IOSurfaceRoot" suggest that MSR stands for hardware blocks, each implementing a different dithering algorithm. The driver waits for IOSurfaceRoot at boot, which further ties it to the display pipeline.

I think the Quadra Bayer dithering algorithm, in the case of the scaler, could be used for handling 2x2 blocks from the ISP and repacking them for display (whatever that means) - the IOSurfaceAcceleratorCapabilitiesQuadraBayerRepacking property in the code hints at this - but I haven't dug that deep.

« Previous Page Next Page »