1559465 : GPU Python environment crashes

Created: 2025-12-09T09:35:42Z - current status: new

Anonymized Summary: A user reports that their Python script with Numba on GPUs, which previously worked, now crashes with a segmentation fault when running CUDA code. They have tried recreating their Mamba environment and confirmed that GPUs are detected correctly. They are seeking guidance on the recommended setup for Python + Numba + GPUs on the current system.

Solution: The user should ensure they are using the correct modules and environment setup for GPU computing on Maxwell. Based on the context, they can try the following steps:

  1. Load the necessary modules for GPU support: bash module load maxwell mamba/3.9 cuda/11.2
  2. Initialize Mamba and activate an appropriate environment: bash . mamba-init mamba activate /software/jupyter/.conda/envs/tensorflow-2.11/ # or another relevant environment
  3. Verify GPU detection and CUDA compatibility: bash python3 -c 'import tensorflow as tf; print(tf.test.is_built_with_cuda()); print(tf.config.list_physical_devices("GPU"))'
  4. If the issue persists, consider creating a custom environment with the required packages: bash mamba create -n numba_env numba cudatoolkit=11.4 mamba activate numba_env

References: - Using pre-installed tensorflow - Using pytorch - Installation