When to run unit-tests?

We are currently setting up our build-process within an automated continious integration environment and facing the fundamental question, when to run unit-tests?

One way would be to run the unit test with every build task. So as soon as one unit-test fails, the whole build fails. This has the advantage, that the developer is always forced to keep the unit-tests green, as s/he is otherwise not able to run the application. On the other hand, you are always distracted by fixing the tests during a development process - which might force you to work in very small iterations. Besides that the time to run your application always increases, as you have to wait for the tests every time.

The other way would be, to let the CI-Server run the tests after each new commit and let the developer simply know, that something went wrong. In this way the developer is pretty free, at what time to care for the unit-tests, but also other developers on the same branch might suffer, because they cannot be sure, that all parts of the software work as expected and have to check theirselves, if the failing tests might also influence their work.

So do you have any best-practices or recommendations, which would be a good time to run the tests?

BTW: of course we also run bigger integration-tests, which are handled in a seperate CI-process.