We apply the GARCH model when there is volatility clustering in our data set, other wise if our aim is to measure variability we can simply exert standard deviation technique

Depends entirely on what you want to do with it. If you are looking accurate estimation of tail quantile of financial portfolio losses for example (VaR or similar), you need to consider volatility being return dependent and look at clustering etc - hence use a garch model or one of its modifications. GARCH model has more parameters to calibrate, and I find that its likelihood has lots of local maxima (i.e., estimation of parameters is quite sensitive to initial values), for stock market data from developed world. This may or may not be the case in all applications.

It depends on your Area and factors which you are using to estimate or forecast. If your model is symmetric the ARCH, GARCH parameter may be significant. But in asymmetric model not all type GARCH model like PGARCH, GJR-GARCH, EGARCH will give you same results. Its entirely depends on both empirical and theoretical background of your Model, in which you want to measure volatility.