Is there any performance (or other) benefit to loading less columns in pyspark dataframe?

Is there any performance (or other) benefit to loading less columns in pyspark dataframe?

Basically my use case is that I have a large table (many rows, many columns) that I am loading in as dataframe to filter down another table based on matching keys in both so something like...

filter_table = sparksession.read.load("/some/path/to/files").select("PK").dropDuplicates()
table_to_filter = table_to_filter.join(filter_table.select("PK"), "PK", "leftsemi")

My question is: Is there any benefit to loading the table like this

filter_table = sparksession.read.load("/some/path/to/files").select("PK")

vs

filter_table = sparksession.read.load("/some/path/to/files")

I suspect I am getting confused on how spark's lazy evaluation works (very new to using spark), but I would think that since I only ever use the table with .select("PK") there would be no difference (unless the entire dataframe is stored in memory once loaded (and not only on evaluation))?