Since Window Server 2008, the role of what we used to call the “Quorum Disk” in a Failover Cluster has changed significantly and it’s now referred to as a “witness” because its only job is to help the nodes in the cluster decide who should be running the cluster resources in the event that they can’t talk to each other i.e. which nodes plus the witness can form a quorum.

Best practice before Windows Server 2012 R2 was that if you had an even number of cluster nodes then you needed to add a witness resource to enable an odd number of votes.

An often overlooked best practice since Windows Server 2012 R2 is to always configure a witness resource, regardless of how many nodes you have in your cluster. This is because failover clustering uses a dynamic quorum and will decide for itself whether or not it needs to use the witness at any point in time depending on how many nodes are up.

This post on TechNet shows how dynamic quorum works: https://blogs.technet.microsoft.com/askcore/2016/03/21/behavior-of-dynamic-witness-on-windows-server-2012-r2-failover-clustering-3/