What is an ensemble?
An ensemble is a set of several model simulations. One simulation of an ensemble is denoted as ensemble member.
A multi-model ensemble is an ensemble created by simulations with several models (e.g. each model is run twice).
Alternatively, one might create an ensemble by performing several simulations performed with one model. Each simulation is disturbed by some noise so that the model produces slightly different result for each model run. We denote each run as a realization. The generation of the appropriate noise is model dependent (the results should cover the natural variability) and a scientific topic itself. Additionally, we might use other initial conditions or different physical parameterizations.
The individual simulations of a CMIP5 model are identified according to the rip nomenclature: rXiYpZ (r: realization; i: initialisation; p: physics; X, Y, and Z are integers), e.g. r12i1p1 or r1i1p1.
It is also explained here: https://portal.enes.org/data/enes-model-data/cmip5/datastructure
Summarizing ensemble simulations
If you calculate the mean of your models runs, you call it the "ensemble mean". If you calculate the standard deviations, you call it "ensemble standard deviation". The "ensemble" prefix just means that you calculated the mean/std.dev./etc. from an ensemble of simulations.
Example: This plot of the 5th Assessment Report of the IPCC (synthesis report of Topic 2 - Future Climate Changes, Risks and Impacts, online version) shows the ensemble mean (solid thick colored lines) and the space covered by the 5% to 95% percentiles (shaded area).

Please have a look here for the caption and more details: http://ar5-syr.ipcc.ch/topic_futurechanges.php#figure_2_1