Hello - I am wondering what Linux distros, versions, driver version, windows manager, etc your using that works with no issues. Also would like to know what gpu. i was thinking of giving arch Linux a try but curious what other distros also are fine and what settings and such make it better? In terms of disabling dithering, font smoothing and such.

I use latest Lubuntu , 22.04 , on Intel nuc95 as is(even live USB worked well) , chrome with no HW acceleration. The display is: VA249HE . Lubuntu also worked without strain on another desktop, stronger one, Intel integrated graphics, now HW not working, so I can't tell more about it.

A couple of months ago, I was experimenting with the Blackmagic UltraStudio Recorder 3G, recording uncompressed 10-bit video from my MBP M1. At that time, I was already using Stillcolor (the application that disables dithering in macOS). I also had Ubuntu 22.04, which I installed on a Parallels virtual machine in macOS. I decided to try recording and analyzing the desktop in Ubuntu. There was no dithering, neither in macOS nor in Ubuntu 22.04. When I enabled dithering in Stillcolor, dithering also appeared in Ubuntu. I don't remember exactly, but it seems Ubuntu was using Wayland, not X11. My hypothesis is this: dithering is influenced only by a driver and a video adapter. In Parallels, the virtual video adapter driver is used, which does not dither, leading to no dithering in Ubuntu when it is run in the virtual machine. However, there's a question about the presence of hardware acceleration in this setup, which I can't yet comment on. But OpenGL is responsible for hardware acceleration. According to the documentation, OpenGL has the GL_DITHER option. “GL_DITHER - If enabled, dither color components or indexes before they are written to the color buffer.” I think it is possible to recompile the Linux OpenGL library with GL_DITHER forcibly set to false. At first glance, modifications need to be made only in two places in the library code: prohibit enabling dithering from outside: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/mesa/main/enable.c#L519 ; find where ctx->Color.DitherFlag is initialized to set it to GL_FALSE. In the kernel, dithering is enabled during the initialization stage of the video adapter (I'm only talking about UHD Intel here). As I understand it, this happens in this i915 module file: https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/i915/display/intel_display.c . I quickly went through the code, and it seems there is a possibility to force 8-bit and disable dithering. The question of hardware acceleration directly in the browser remains. As far as I know, some browsers use WebGL. The question here is whether this library calls the system's OpenGL library or not. I found information that Intel Arc also uses the i915 kernel module, and I'm interested in whether it's possible to disable dithering not only for UHD but also for ARCs. I recently ordered a mini PC based on the i5-12450H processor. I want to set up an additional workspace using it. Since Windows is bad for my eyes, I want to try Linux and see if I can disable dithering for UHD / ARC.

I installed the latest version of Manjaro XFCE on a mini PC (i5-12450H, UHD Graphics (48EU)). Right out of the box, it was very uncomfortable; after a couple of hours, my eyes started to hurt, and it became difficult to work. After making edits in the i915 module, it got better. Noticeable transitions in gradients appeared: https://ibb.co/q1f6Pw4 As for OpenGL, it's still unclear because Manjaro XFCE has several different OpenGL builds, and some of them are loaded, including in the Firefox process. It's not yet clear where to get the code for them (so far, I've identified and rebuild only one of several libraries). But if I make it so that the OpenGL libraries are not used by the OS (hardware acceleration is switched off), it seems to make it easier on the eyes, but this is still unproven and based on feelings. I am currently on a business trip and have limited access to equipment. In a couple of weeks, I will be home and will try to understand the difference between the adjustments to eliminate subjectivity. I will also continue to explore the possibility of rebuilding OpenGL.

I really don't think you need to do that.. static dithering has never been an issue. It's temporal FRC that people have an issue with. It's drivers that might use FRC or hardware in the panel. This gl_dither thing has primarily been used when working with a colour depth that doesn't match e.g. 4 bit and 8 bit. And really, there's an easy way to test.. just don't use opengl as the compositor and you have your answer. XRender is a software-only path for x.org.

Sunspark You are right about OpenGL, I can confirm that rebuilding OpenGL libs with dithering disabled did not yield results. Visible changes only occurred after modifying the i915 code. Currently, MBP M1 + Stillcolor seems slightly more comfortable for my eyes than Manjaro with i915 module tweaks (although I'm not 100% sure yet). And really, there's an easy way to test.. just don't use opengl Without using OpenGL, the text looks different and is easier for me to read. However, I suspect this is not because of specific OpenGL libraries, but rather due to the system rendering fonts differently without any hardware acceleration. I think I'll dig into font rendering and see what can be done here.

who can use Linux but not windows? Share your Linux setup

DisplaysShouldNotBeTVs

Sunspark static dithering has never been an issue. It's temporal FRC that people have an issue with

this is mostly true with older devices — but some more modern versions of "spatial dithering" are "semi" dynamic where even though they're still if nothing is happening, the dithering will slightly move around if you're moving your mouse cursor across things (or any similar kind of change to the screen)

for example, the dithering in e-ink devices is not really temporal as it's "usually" still, but areas of the screen will "regenerate" an entirely different dithering pattern when any motion happens, which can still cause issues

WhisperingWind

I took a closer look at the i915 module code. It appears that dithering is disabled by default and only enabled for 6-bit displays:

https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/i915/display/intel_display.c#L4777C2-L4782C37

In other words, my edits theoretically should not have had any effect.

Hmm, that's strange. Why do my eyes seem to strain a bit less after the edits? It might be related to my hardware setup: Tecno MEGA MINI M1 PC -> Thunderbolt 4 -> Belkin USB-C Video Adapter -> DVI cable -> BenQ GL2450 monitor (6-bit+FRC). I might have accidentally changed some settings elsewhere, which had such an effect, or it could just be the new kernel version. There are many questions.

I think I need to start by getting a monitor without dithering and see how Linux looks by default on it (maybe I'm not fighting Linux, but my current monitor). And I need to understand how all this works. Otherwise, there will be a lot of magic and unicorns here), which is not good.

I think I should first learn how to detect dithering in the video card's output signal (I think I can create some kind of detector in Python) and then build further work based on that.

I will hold off on fully presenting the results until I am completely sure that I have achieved some outcome and it is 100% not a placebo.

DisplaysShouldNotBeTVs

Can you please tell me if you have any developments in the field of image processing for detecting dithering? As I understand it, temporal dithering can be detected using the method proposed by aiaf. But what about other types of dithering? Do you possibly have any links to some materials?

DisplaysShouldNotBeTVs

WhisperingWind experimenting with the Blackmagic UltraStudio Recorder 3G, recording uncompressed 10-bit video from my MBP M1

can you send me some still PNG frames of some macOS apps taken directly from the lossless capture? i want to analyze the colors themselves to see if there are any other (non-temporal) anomalies like color fringing or shadows/halos where there shouldn't be

i want to see if there's anything different about the colors that actually get sent to the video output, compared to what e.g. a normal screenshot of the same app looks like

WhisperingWind

DisplaysShouldNotBeTVs

The HDMI recorder is unfortunately at my home. But I am currently on a business trip. At the end of September, I will return home and will be able to record and send the images.

twomee

WhisperingWind what did you changed in the OpenGL and in i915 code? How did you deployed it? I want to do it too, don’t have a guide?

WhisperingWind

twomee

I think I can put together a guide and upload the code to GitHub next week.

jordan

WhisperingWind that would be useful. I have a Intel arc a770 LE and can experiment for you if you can help me turn the dithering off. I think it's spatial but I still would like it off so there's no chance of it bothering by scrolling or cursor movement. I did order a epiphan dvi2usb 3.0 framegrabber to test output of the cards I have. I do have a dvi2pcie as well but haven't set it up in a PC yet so gonna try the USB one since it was cheap

WhisperingWind

jordan I have a Intel arc a770 LE and can experiment for you if you can help me turn the dithering off.

That would be great!

jordan

WhisperingWind Did the edit possibly turn off the +2frc on your monitor maybe? Only running in 6bit?

autobot

WhisperingWind if i understand you correctly, you want to use a cature card, cature some video frames straight from the display connection and than analyze them in python?

JTL

WhisperingWind I think I should first learn how to detect dithering in the video card's output signal (I think I can create some kind of detector in Python) and then build further work based on that.

Already exists as an (in progress) project.

https://github.com/ledstrain/videodiff

WhisperingWind

jordan

Yes, I did.

The Thunderbolt 4/USB-C 4 output is recognized as a DisplayPort. Therefore, it is possible to set it to 6-bit through xrandr. In this case, the image on the monitor is less straining on the eyes (in the edits I disabled dithering for 6 bits, at least, I hope it works as I expected).

I removed this lines https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/i915/display/intel_display.c#L4777C2-L4782C37 from the code. I want to return it later and see if there are any changes.

jordan

WhisperingWind interesting. I own a sun vision 32" rlcd that's 6+2frc that I never tried after learning its not true 8 bit. I wonder if it's truly safe as 6bit only in Linux with the arc after driver edit

WhisperingWind

jordan

Does the monitor have a DisplayPort? Otherwise, it might not work, as HDMI (at least in Linux) does not support 6-bit.

You can try using a video adapter with a DP input, like the one I have, but I'm not sure which one will work in this case. Mine has a USB-C connector, and it won't be suitable for Intel Arc.

WhisperingWind

jordan

Today, my ASRock Arc A770 arrived. It seems that HDMI on it is routed through DP. Check your A770 LE in Ubuntu Desktop (X11); if it's the same for you, you won't have any issues setting 6 bits with the xrandr utility when connected via HDMI.

They mention here that LE also doesn't have a 'true' HDMI (https://www.vegascreativesoftware.info/us/forum/intel-arc-general-discussion-of-newest-gpu-competitor--140323/), and most likely, inside the graphics card, it is connected via a converter to DP.

You can check with the command

xrandr --verbose

You should see something like "DP-X connected primary…". If you see something like that, you can set it to 6 bits with the command

xrandr --output DP-X --set "max bpc" 6

where X is the number of the DP output.

My HDMI connection is detected as DP-1, which indicates that the HDMI port in it is not genuine:

DP-1 connected primary 1920x1080+0+0 (0x4b) normal (normal left inverted right x axis y axis) 531mm x 298mm

Clones: HDMI-1

max bpc: 6

range: (6, 12)

subconnector: HDMI

P.S. Regarding dithering, I can't say for certain yet. It seems that after switching to 6 bits, everything is okay, but more time is needed for testing.

WhisperingWind

I came across an interesting thing: https://github.com/skawamoto0/ditherig/blob/master/ditherig/database.csv. As I understand it, this is for integrated graphics cards from Intel and AMD. We can play with addresses to disable dithering. This can be done in Linux using the devmem2 program, which can be downloaded from https://github.com/VCTLabs/devmem2. It can be built by running make in the directory with the source code.

The app can be run like this (you need to substitute the addresses of your own chip, presumably where it says "Disabled"):

./devmem2 0x70030 w 0x00000000

As I understand, the address can be taken from the 8th column, the size is 4 bytes. For simplicity, you can try filling it with zeros (although, of course, one should properly apply a mask from the 10th column).

I wonder if this will work.

UPD. In reality, as usual, it turned out to be more complicated: in the 8th column, it's not the actual address but an offset relative to it. The actual address is calculated based on the values of Bar1/Bar2.

ryans

WhisperingWind

You might also find intel_reg useful, there's a thread about Intel graphics dithering registers + Linux here: https://ledstrain.org/d/1017-registers-to-control-intel-graphics-dithering-on-linux-located/17.

WhisperingWind

autobot

Yes, I want to try doing this. Because a couple of times during experiments, I mistakenly started thinking that my eyes were feeling better, but that wasn't the case.

Donux

jordan I use the most mainstream linux distro - Ubuntu with old Lenovo X1 laptop, Intel UHD 620. Also very important, running X11, NOT WAYLAND. Wayland for me introduces some tension. Not as bad as Windows 11, but kind of feels over time. Some suggested to use Y, X, Z in terms of distros. Well I have tried some of them, and I can clearly suggest, if you really about getting work done, do not branch out. It is as messy as it can be, and overall best experience is on mainstream Ubuntu with most features within UI working quite well. Of course it is still buggy, you can not compare it to windows or especially Mac Os, but it is as good as you can get right now. My suggestion for best machine would be lenovo X1, but definitely not brand new. Also, do not expect this to be a permanent solution if you are a macbook user. Macbooks really make things simple, nice to use and just rock solid like calculators.

autobot

Donux Thank you, i had problems with ubuntu in the past(in a good hardware), but i didn't try running X11. I'll try that.

jordan

Donux did you have a negative experience with a newer igpu with same software setup? Also so x11 is default on latest Ubuntu? I haven't used Linux in soooo long but it seems very promising. I am previously a windows user. I just want something that works I don't care for any bells and whistles anymore tbh haha.

WhisperingWind that happens to me too sometimes and catches up to me later

zlhr is the font hinting a usual setting in settings on Linux distros ? Or is it something that has to be added?

eDenon-2 did you alter any settings? DP or HDMI?

zlhr

Stock Fedora (40, Wayland, Gnome) improved a previously unusable device for me (running HD 3000, now with i915 driver).

The only tweaks of note is that I always set font antialiasing to grayscale and turn off the hinting, as it helps with my astigmatism.

Suprisingly on distros that use X that many people have success on here, I fare much worse.

eDenon-2

I'm using Debian 12 Gnome, I previously use Archlinux but installation is a hassle.

Display: ASUS VG258QM
Graphics Card: AMD RX6400