How important is a Read Group for GATK?

[Niranjanks]

Hello I am doing Variant Calling on a sample -

I received a total of 3 files to analyse.
They were divided into 2 runs
So Run1 > 1 paired end file
and Run2 > 2 paired end files run on different lanes

GATK Best Practices Variant Calling pipeline required that I add Read Groups to the files which I did using Picard AddOrReplaceReadGroups.

I tried merging the 3 files and then performed Marking Duplicates and this next step needs Read Groups. I can give it a single read group right now

OR

I can go back and assign 3 different read groups to the 3 set of files and then merge them

What is preferable?