A cluster is a collection of, usually, homogeneous compute nodes. They are usually split into MPI and SSI, Message Passing Interface of Single Server Image. The latter is a bunch of machines trying to emulate a single system and is not commonly found in the HPC world. You are more likely to find MPI setups where each bit of processing can be broken into smaller pieces and distributed to each node.
For a render farm you can have machines with no knowledge of each other as they can each work on a separate set of frames. If the rendered is MPI aware it can ask neighboring nodes for data.
ROCKS Clusters are MPI based but can be used as individual machines. ROCKS strength is its simplicity in managing the OS image on each of the nodes. You plug a machine into the network, flip it on, wait for ROCKS to find the MAC broadcast during PXE boot and then assign it to a predefined image group. Minutes later you have a machine installed exactly as all other nodes on the cluster are. This is sometimes vitally important as certain software may not work with a different revision of the OS.
ROCKS allows for easy management and easy expandability. If you need more compute power, plug in new nodes, collect the MACs and image.