Below is a simplified picture of the EPS architecture.

The EPS system is made up of the Evolved Packet Core (EPC) and the E-UTRAN. The EPC provides access to external data networks (e.g., Internet, Corporate Networks) and operator services (e.g., MMS,
MBMS).

It also performs functions related to security (authentication, key agreement), subscriber information, charging and inter-access mobility (GERAN/UTRAN/E-UTRAN/ I-WLAN/CDMA2000 etc.). The CN also tracks the mobility of inactive terminals (i.e., terminals in power saving state).

E-UTRAN performs all radio related functions for active terminals (i.e. terminals sending data).

Between the EPC and E-UTRAN there is an interface called S1.

An E-UTRA capable terminal is connected directly to E-UTRAN. However some parts of  the terminal control-plane protocol stack is also terminated in the EPC.

EPS to EPC

A key feature of the EPS is the separation of the network entity that performs control-plane functionality (MME) from the network entity that performs bearer-plane functionality (SGW) with a well defined open interface between them (S11). Since E-UTRAN will provide higher bandwidths to enable new services as well as to improve existing ones, separation of MME from SGW implies that SGW can be based on a platform optimized for high bandwidth packet processing, where as the MME is based on a platform optimized for signaling transactions.

This enables selection of more cost-effective platforms for, as well as independent scaling
of, each of these two elements. Service providers can also choose optimized topological locations of SGWs within the network independent of the locations of MMEs in order to optimize bandwidth reduce latencies and avoid concentrated points of failure.