[protastus]

Yes, on some cards.

I have workstations with dual RTX A6000 that run far too hot (~85 C) and drivers crash with PCIe errors. These are dual slot cards with blower fans, so they are loud, and it seems that NVIDIA chose to let the silicon get extremely hot to keep the noise down. NVIDIA's silicon might be rated to run super hot but the motherboard isn't, and signal integrity apparently suffers.

It's definitely an issue with the cards because the case has excellent airflow and the cards are comfortably spaced.

My solution is to force fan speeds using nvidia-settings. And nvidia-settings requires X11 to run, which is kind of a ridiculous solution for setting fans on headless systems.