This isn't correct. What you may be remembering is that some (not all) NP complete problems have limits on how accurately they can be approximated (unless P = NP). But approximation algorithms for NP complete problems form a whole subfield of CS.