DMVPN Phase 1 uses HUB-and-spoke tunnel deployment. The tunnels through which inter-branch connections are made are only built through the central DMVPN hub and the individual spokes, working much like a traditional VPN system.
DMPVN Phase 2 uses spoke-to-spoke tunnel deployment, meaning that data doesn’t have to travel to a central hub first, so long as there are specific routes in place for the spoke subnets.
DMPVN Phase 3 allows for spoke-to-spoke tunnel deployment, but without the specific pre-made routes in place, but rather uses NHRP traffic indication messages from the hub to secure those routes on the fly.
Hub and Spoke Network Architecture
As mentioned, the Hub and Spoke Network Architecture is a way to efficiently manage the endpoints that are being secured. There are three different phases, each of them suitable for different configurations. However, in general, the hub is used to configure the protocols by which the connections are secured, and these protocols are then applied to the spokes (or endpoints) of the network. This system can be centralized to ensure that allow spoke-to-spoke tunnels first go through the hub, or can be decentralized to cut the hub out entirely.

DMVPN Components
Multiple GRE tunnel interfaces: a single GRE interface that can secure several IPsec tunnels, reducing the overall scope of the DMVPN configuration
IPsec tunnel endpoint discovery: meaning that static crypto maps between individual IPsec tunnel endpoints do not have to be configured
Routing Protocols: which can allow the DMVPN to find routes between different endpoints much more effectively
NHRP: which can deploy spokes with assigned IP addresses that can then be connected to from the central DMVPN hub.
multipoint