A Comprehensive Evaluation of GPT-4V on Knowledge-Intensive Visual Question Answering