CUDA Init fails on ARM platform

Hi everyone.
When trying to init CUDA I get

INIT CUDA
NVRM: failed to register with the ACPI subsystem!
NVRM: failed to unregister from the ACPI subsystem!
NVRM: failed to register with the ACPI subsystem!
nvidia-uvm: Loaded the UVM driver, major device number 236.
NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x23:0x40:624)
NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 0
NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x23:0x40:624)
NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 0
NVRM: failed to unregister from the ACPI subsystem!
info: no CUDA-capable device is detected - 100

I’ve tried T1000 and RTX3000 GPUs, double-checked at CUDA GPUs - Compute Capability | NVIDIA Developer that they are indeed supported.
I’ve tried several different driver versions from 460 to to 470 of Linux aarch64 driver package and SDK.
I’ve tried several linux distributions including building a very current kernel from scratch.
I’ve tried swapping UEFI for U-Boot.
I’ve tried different hardware platforms.

It’s always the same messages.
Ironically normal graphics operation works ok. Everything else seems to be fine. Only when trying CUDA I get the failure.

Has anyone got an idea what causes the failure?
Is ACPI relevant in this case and how can it be fixed?

Has anyone got any ideas or suggestions what’s wrong?
Thanks!