I have tried both using ACS/census and they seem to underestimate minorities as I have filtered with Hispanic/subtracted to get white/black only, hispanic and other.
In general if you are performing your analysis at the person level you should use the person weight, and if you are performing your analysis at the household level you should use the household weight. I am not certain if your research question is at the person our household level, given the information you’ve shared so far. I can imagine an analysis that either counts households or counts individuals.
Regarding the “underestimate”, do note that we do not generally expect to exactly replicate official statistics with public use data. This is primarily due to the fact that official statistics use restricted use data and the data available via IPUMS is public use data. Public use data is limited by confidentiality restrictions. That being said, we do expect to get relatively close and within some margin of error of officially published population counts. For more information on this see this blog post.