Once the mars globe in the main menu is at full size, on my Laptop (with Ryzen 7640U w/ RDNA3) I get these blue flickering specks; I wonder if everyone gets them.
@Doomed_Daniel I see this problem with discrete RX 7900 GRE, but it is fine on renoir APU. Both on X11. Haven't checked on more modern APU on 7950X - ping me if you're interested, I'm not sure how to test it properly without replugging monitor to motherboard display port, so haven't done it.
Steam Deck OLED would be very interesting, as it has a RDNA2 GPU, which is only one generation behind my RDNA3 GPU (and at least two generations newer than the Vega GPU in the 4800H)
On my laptop this happens with Ubuntu 24.04 and also a very recent Arch Linux live image.
It does not happen on my old laptop with Intel GPU (with same Mesa and Kernel version), and not on my Desktop with nvidia GPU.
A friend couldn't reproduce it with his Radeon 5700 either.
But so far no one has tested it with RDNA3 (or even RDNA2), or AMD iGPUs in general (except for me on my Framework 13 AMD Laptop).
I wonder if this is a driver bug or maybe defective hardware or something.
Doesn't happen on my secondary desktop with Radeon RX 580 (with the same Arch live system) either.
On the laptop the issue happens with X11, Wayland and XWayland.
Anyway, I'd be thankful about any reports, including "works on my system", ideally with information on your Linux kernel and Mesa version ($ glxinfo | grep "OpenGL renderer string") and which GPU/CPU your device has (for CPU: $ grep "model name" /proc/cpuinfo | head -n 1).
So far the issue could only be reproduced on Ryzen 7x40 CPUs - it doesn't seem to happen with GCN5/Vega iGPUs (in Ryzen 4000 and 5000 APUs), nor with dedicated Navi/RDNA1 or RDNA2 GPUs (like RX 5700 or RX6700M).
I'm still especially interested in:
RDNA2 iGPUs (Ryzen 6000 mobile or 7000 desktop)
RDNA3 iGPUs (mobile Ryzen 7x40 and 8x40 and 8x00G for desktop), esp. in other devices than Framework laptops
@Doomed_Daniel I may be not correct - in Vega iGPU NAN and INF from shaders were "red" on screen when screen connected to GPU, but latter they removed this feature from drivers, now it black.
Maybe it also something to do with nan or inf.
And if it Vulkan - there many "hidden" coflicts with memory types for AMD drivers, and validation layers dont catch that, AMD very strict with it - maybe somewhere in code used VK_ATTACHMENT_LOAD_OP_DONT_CARE instead of VK_ATTACHMENT_LOAD_OP_CLEAR.
@danil Hmm, it's neither red nor black (but it's not Vega but RDNA3, maybe there it's blue), but maybe it could be that NAN or INF are involved - though OTOH, shouldn't that cause glitches on other GPUs/drivers as well?
This isn't Vulkan, but OpenGL 1.4 with ARB assembly shaders
it does not have inf and nan support.
You need to "catch them" https://www.shadertoy.com/view/lsjBDy - look there define isNaN/inf - this you can use to catch them in old graphic api.
But - in general shaders translated/compiled to same "binary" - and modern GPU do not have support for 8-bit float and 16-bit float, only 32-bit, and maybe your shaders logic include some 8 or16 bit hardcoded logic that break on 32-bit shaders.
RDNA3 - is architecture with only 32-bit floats.
@Doomed_Daniel But more realistically - I think it just "broken OpenGL drivers in AMD"... if you try to fix this bug - you will run in million other bugs.
As option that may work - try zink, that is OpenGL to Vulkan translation - zink work on Windows also it just opengl dll that you put at exe. But zink not always work, so try. It junk-solution ofc.
Ofc ultimate solution should be - just port/move to Vulkan.
@Doomed_Daniel I read your comment about pow and clamp
what if you replace clamp(0.,1.) with clamp(0.001, 1.) - still bug?
clamp should always return 0. positive that equal to lower clamp region, but maybe there some weird stuff that made clamp return negative 0 somehow, I can be wrong ofc just guess.
@danil
I hope that some AMD driver developer figures this out =)
I guess if POW was generally broken someone would have noticed by now, but if the issue is OpenGL ARB shader specific, it's more likely that no one noticed because those are pretty rare (and it's not like POW always turns the pixels blue, only some of them, and they change all the time, giving a flickering effect)
@slembcke thanks, that's good to know - at least my hardware is (likely) not broken =)
Of course it's still unclear if it's framework-specific, Ryzen 7040 specific, RDNA3-specific (or just RDNA3 iGPU)... but I'm glad that someone can reproduce it at all
Add comment