Separate names with a comma.
Do you need full CUDA support for TensorFlow, or would CPU/alternative acceleration be enough for your use case?