Use all samples or group samples for GenotypeGVCFs?

I have created g.vcfs for all my samples (48). Half of my samples (24) are control individuals and the other half (24) is treated individuals. Should I run GenotypeGVCFs on all my samples once to generate one raw vcf? Or twice, once for treated samples and once for control samples producing two raw vcfs? Ultimately I am looking for differences between my groups.