GPT-4 API result not stable with temp 0

I am using gpt-4-0613 model with temp set to 0 but it seems like given the same prompt it still sometimes genereate different results? Is it normal with GPT-4? Been using 3.5 for a while. The result from 3.5 seems to be much more stable.

The restriction on unexpected output can further be enhanced by also using the top-p parameter. Setting it to 0.0 out of 0.0-1.0, where 1 is the default, vastly curtails the alternates that can be chosen.

GPT-4 is a particular mystery to outsiders, in that internal parts of the model could have different interpretations or quality from temperature.