Issue with CUDA on Drive PX2

Autonomous Vehicles DRIVE PX 2 DRIVE Hardware
,
1

Hi!
We have an issue with initializing CUDA (9.2) on Tegra A (and B). When we are trying to run a sample from /usr/local/cuda-9.2/samples/3_Imaging/EGLStreams_CUDA_Interop/ the system goes into a reboot on cuInit invocation.
This issue happens on one of our boards.

dmesg last words are:

[ 306.438810] nvgpu: 0000:04:00.0 __nvgpu_check_gpu_state:60 [ERR] GPU has disappeared from bus!!
[ 306.440282] nvgpu: 0000:04:00.0 __nvgpu_check_gpu_state:61 [ERR] Rebooting system!!

2

Dear @ProgDevel ,
Is the GPU is working after reboot. Please check running CUDA deviceQuery, matrixMul samples to confirm. Could you check flashing the target again to see if the issue persists?

4

Thank you for answering.

  1. Yes, GPU is working after reboot
  2. matrixMul causes reboot too. deviceQuery works good. Please, see log: dq (4.5 KB)
  3. Issue is flickering but happens almost always
5

Dear @ProgDevel ,
looks like issue is occurs when we use GPU. Can you flash the target again to confirm if HW issue?

6

OK, I’ll try flashing later when I have physical access to the board. And let you know

8

Sorry for the late reply. We’ve checked cuInit after flashing the TegraA and faced random rebooting again.