Lets work our way up

1 server, no RAC
============
Whilst that server is running, all is fine. If that server crashes, or you need to do maintenance....your application is down.

2 servers, no RAC but has clustering
===========================
Both servers are running, but only 1 server is doing any work. The other sits idle. If the main server crashes, the other one starts a database, transfers over facilities and resumes workload. Typically this could be 5-10mins, or longer if any manual steps are needed.

2 servers, RAC One Node
===================
Both servers are running, but only 1 server is doing any work. The other sits idle however, because it is running RAC, it already "knows" about all of the facilities on the other node. If the main server crashes, the other node can immediately take over facilities and workload. Typically this could be 30 seconds or even less, and applications can be insulated from this, ie, they just see a small delay rather than an error

2 (or more) servers, Full RAC
=====================
*ALL* nodes can be processing queries/DML at the same time. If a node dies, the other node(s) sort out any mess (roll back or reply transactions) so generally users won't even notice that there is/was an service interruption.