An application container cannot access GPUs although the node has them. What should be investigated?

1. Only the class distribution of the application input
2. Driver, runtime integration and device allocation
3. Only the inference batch size
4. Only the Service load-balancing algorithm