Pick sync or async…

In LLM serving, what is the primary driver of end to end latency for a generation request?