Numba cannot see Blackwell GPU and index out of range

Hello,

We found some rather weird issue while trying to run the sample DFP Azure jupyter notebook in Morpheus 25.06. The host shows GPU is present but on Numba.

nvidia-smi shows my RTX PRO 4000 GPU is present and also in “docker exec morpheus bash” the GPU is present with the same command.

Driver: 580.95.05

CTK: 1.18.1

However in my simple Numba test or launching the example jupyter notebooks there is an error: index out of range. Here is the Numba test result after isolation:

CUDA_VISIBLE_DEVICES: Not set

Numba sees devices: <Managed Device 0>

Numba device count: None

CuPy device count: 1

Further output:

---------------------------------------------------------------------------IndexError Traceback (most recent call last) Cell In[1], line 4 2 numba.cuda.current_context() # Force init + context attach 3 print(“Numba devices after force-init:”, list(numba.cuda.gpus)) # Should list <Managed Device 0> ----> 4 print(“Current device:”, numba.cuda.gpus.current)File /opt/conda/envs/morpheus-dfp/lib/python3.12/site-packages/numba_cuda/numba/cuda/cudadrv/devices.py:58, in _DeviceList.current(self) 56 devnum = ac.devnum 57 if devnum is not None: —> 58 return self[devnum]File /opt/conda/envs/morpheus-dfp/lib/python3.12/site-packages/numba_cuda/numba/cuda/cudadrv/devices.py:40, in _DeviceList._getitem_(self, devnum) 36 def _getitem_(self, devnum): 37 ‘’’ 38 Returns the context manager for device *devnum*. 39 ‘’’ —> 40 return self.lst[devnum]

May I know is there any hit before or advice?

Thank you!

The OS is Ubuntu 22.04.