I feel like GPUs don't thermal throttle as gracefully as CPUs do. I have a desktop GPU and when it's at its max temperature for too long, eventually the graphics driver crashes. Meanwhile, CPUs are perfectly happy to run with inadequate cooling; they just thermal throttle.
I worry that it might actually be a software issue. GPU clock starts running slowly, software assumes that it will be running faster, two commands are issued but only one is processed, driver gets confused, driver crashes.
Seems like that wasn't working fully, since they report having GPU issues after stress testing the computer (external GPU worked fine). And that is worrying. Is the thermal stress shutdown from OS, or onboard the GPU/CPU?