If you run an LLM with optimization turned on on a NVIDIA GPU then you can get non-deterministic results.
But, this is a choice.