How to assign memory and CPU cores for genotypeGVCF?

We followed the workflow in http://gatkforums.broadinstitute.org/gatk/discussion/3893/calling-variants-on-cohorts-of-samples-using-the-haplotypecaller-in-gvcf-mode to run GATK hyplotypecaller for a large whole exome sequencing project (~6200 samples). However, in step 3, using genotypeGVCF to merge 31 gvcf became very slow. So, we decided to run each chromosome in parallel. But even for an individual chromosome, our cluster always run out of memory.
To avoid the memory issue, I want to know how much memory and CPU cores we should assign for an individual genotypeGVCF job? How to set java -Xmx and -nt in a command line?
Thanks,