https://bugzilla.kernel.org/show_bug.cgi?id=206475
--- Comment #17 from Andrew Ammerlaan (andrewammerlaan@riseup.net) --- (In reply to Alex Deucher from comment #16)
When the GPU is in reset all reads to the MMIO BAR return 1s so you are just getting all ones until the reset succeeds. 511 is just all ones. This patch will fix that issue: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/ ?id=9271dfd9e0f79e2969dcbe28568bce0fdc4f8f73
Well there goes my hypotheses of the broken thermal sensor xD.
I did discover yesterday that the fan of my GPU spins relatively slow under high load. When the GPU reached ~80 degrees Celsius, the fan didn't even spin at half the maximum RPM! I used the pwmconfig script and the fancontrol service from lm_sensors to force the fan to go to the maximum RPM just before reaching 80 degrees Celsius. It's very noisy, *but* the GPU stays well below 70 degrees Celsius now, even under heavy load. As this issue seems to occur only when the GPU is hotter then ~75 degrees Celsius, I'm hoping that this will help in preventing the problem.
I'm still confused as to why this is at all necessary, the critical temperature is 91, so why do I encounter these issues at ~80?