SAS glimmix overdispersion

#1
Let's say I have an experiment with a total of 10 petridishes each with 20 seeds. 5 of these petri dishes are variety A and 5 are variety B. I count the number of germinated seeds in each dish. There seems to be to be two main ways to model potential over dispersion:

proc glimmix data=a;
class rep variety;
model germinated/n=variety / d=binomial link=logit ;
random _residual_ ;
run;

proc glimmix data=a;
class rep variety;
model germinated/n=variety / d=binomial link=logit ;
random intercept /subject=rep(variety);
run;

I understand that the first method affects the R-side matrix and the second affects the G-side matrix but what does this mean in practice and how do I decide which is most appropriate?