Counting the intersection of two large sets meant building the intersection first, which allocates a set you immediately discard.
# before
SINTERSTORE tmp set:a set:b
SCARD tmp
DEL tmp
# 6.2
SINTERCARD 2 set:a set:b
SINTERCARD 2 set:a set:b LIMIT 100
# LIMIT stops counting early, which turns "do these
# overlap at all" into an O(1)-ish question.The LIMIT is the part that changes what is affordable: asking whether two hundred-thousand-member sets share any member no longer costs a full intersection, because the count stops at the limit. Without it this is still a full intersection internally and saves only the allocation and the round trips. Both sets being large and mostly disjoint is the case where the difference is dramatic.