I’m quite convinced - wether this is a hardware or a software issue - this must be an issue that’s rooted in the mode the GPU is in. Once this issue appears, it will continue appearing within minutes even after a reboot if I don’t turn off the device for at least like half an hour, reproducibly. Though there’s exactly one scenario that seems to be reliably preventing it: playing a video. I can’t tell if it just needs to be in the foreground or if it needs to be full screen (I only have it playing full screen), but no matter how I play it, be it locally through mpv or like a YouTube video through Firefox, it will always prevent it from happening. Though I can’t tell if watching the video or if the hardware accelerated playback is doing the fix. Also, in tty mode it barely ever happens.
EDIT: just having the video play in the foreground is enough, doesn’t need to be fullscreen. Though I haven’t yet tested software vs hardware decoding.
And I can confirm this for Debian Testing, Gnome (up to 49.2 currently) and BIOS 4.02, on basically any kernel of the past year or so. Though I haven’t tried suspending for fixing it yet, all I tried was rmmod -f amdgpu && modprobe amdgpu (which either fails or takes a very long time to complete, no idea) and just using the SysRQ key combination for rebooting (which fails sometimes, while the combination for shutdown basically never works).