LLM are memory-bandwidth bound so higher core frequency would not help much.