amdgpu is shit for me, my friend. Funny story: my headless server with a small Navi 1 workstation card (repurposed) couldn't be SSH'd into last week. Went and plugged in a monitor, rebooted, and the framebuffer got stuck during stage 1 while fsck'ing my disk. OK, I think to myself, it's probably taking a while to fsck after N boots, happens once every few months. Doesn't change for 48 hours. Turns out my machine had just run out its DHCP lease, so it had a new IP and I didn't realize it, which is why I couldn't log in. So I log in, and..
What was wrong? What actually happened was that on bootup, the amdgpu driver would panic and fault during boot exactly when fsck was happening, and the framebuffer would be stuck forever. So it just looked like a filesystem issue but in reality my graphics output was merely fubar; the system itself was otherwise fine tnough.
This is reliable and reproducible for me; it always faults at almost the exact same spot at boot every time, for this kernel version at least. In reality amdgpu has been unreliable for me for 5+ releases at this point on a card less than like 7 years old.
Really considering moving over to a small cheap nvidia card and just running Nouveau instead. At least then I might have a reliable framebuffer.
What was wrong? What actually happened was that on bootup, the amdgpu driver would panic and fault during boot exactly when fsck was happening, and the framebuffer would be stuck forever. So it just looked like a filesystem issue but in reality my graphics output was merely fubar; the system itself was otherwise fine tnough.
This is reliable and reproducible for me; it always faults at almost the exact same spot at boot every time, for this kernel version at least. In reality amdgpu has been unreliable for me for 5+ releases at this point on a card less than like 7 years old.
Really considering moving over to a small cheap nvidia card and just running Nouveau instead. At least then I might have a reliable framebuffer.