A cool application which I can somehow appreciate is Van den Bergh's proof of dimension $3$ case of the Bondal-Orlov conjecture that two birational smooth Calabi-Yau varieties $X,X'$ have equivalent derived category $D(X) \sim D(X')$. Note that since one can [construct the pluricanonical ring from $D(X)$](http://arxiv.org/abs/math/0501094) , this is a generalization of the Batyrev's conjecture that they have the same Hodge numbers. The dimension $3$ case was first proved by [Bridgeland](http://www.springerlink.com/content/d982q37w96qvp7jn/), but Van den Bergh's proof uses non-commutative stuff in a very concrete way. Some references can be found in this [question](http://mathoverflow.net/questions/14350/existence-of-non-commutative-desingularizations) I asked. It goes as follows: by some mimimal model program results, in dimension $3$ any birational $X,X'$ are related by a series of flops $X \to Y \leftarrow X^+$. So we only need to prove $D(X) \sim D(X^+)$. Then one builds a special vector bundle (sum of an exceptional collection of line bundles in this case, some perverse stuff!) on X. Pushforward to $Y$, one gets a coherent sheaf $E$. Let $A= End(E)$. The funny thing is that $D(X) \sim D(A)$. Note that $A$ is non-commutative. Now, do the same thing for $X+$ one gets $A+$, say. But it is fairly easy to prove that $D(A) \sim D(A^+)$ directly on $Y$, so we are done. If we carry out this on an Atiyah's flop or Reid's pagoda, one can see actually that $A$ and $A+$ are the same. This indicates that the non-commutative route can simplify things. There seem to be many experts on this site, so surely you will get a lot of much better answers. But this example seems most down to earth for me, and the conjectures come from complex geometry (motivated by physics?) so I hope it helps you as well.