(most) Chess engines without end-game database will play horrible endgame to the extend that a lowly IM can beat the strongest chess engine on the best consumer hardware. At least this was still true last time I was keeping tabs on Computer Chess scene. But with 4-5-6 pieces endgame its not only perfect but it also effects game quality in the middle-games, where most of the games are decided between human-comp games.

Nowadays, Computer Chess scene is not as exciting as it was 10 years ago (not because of Deep-blue). Now it has come to a stage that it is less about knowledge algorithm tweaking and more about, fine-tuning and narrowing down opening books to few sets where a particular engine is strong at and use extensive positional and book learning from previous games on those or similar lines of moves and win games.

Now when you (or GMs) play against a computer engine, you are not as much playing against an engine, but playing against a huge database that has been fine-tuned over 100,000s of games.

This is why no chess engine author will agree to an official tournament games against a top GM, without opening-book, EGTB, pos-learning, book-learning.