LLM Inference Speed for Prompt Processing