nvidia driver: 460.32
cuda toolkit: 11.2
platform arch: arm64
Nvidia Card: NVIDIA Corporation TU104GL [Tesla T4]
GPU card can be detected and GPU driver can be loaded normally. but the kernel panic at the same virtual address several times after executing nvidia-smi command, the physical address corresponding to the virtual address is reserved on my arm64 server.
Call trace in attachment.
I have a question:
The nvidia driver uses the fixed physical address ？