Running Gnomix for 600K individuals

[vicbp1]

Dear all
I face memory issues when running gnomix for 600K individuals (with no rephasing).
We were thinking of two strategies to deal with this, and I would like to know if they make sense.

Splitting chromosomes: We considered splitting the chromosomes into two overlapping segments.
For example:
start end
region 1: 1 - 20,000,000
region 2: 15,000,000 - 40,000,000

Subsetting the dataset: Subsetting the dataset in groups of 100K individuals

Which of these two strategies will you recommend?
Would subsetting the data but running the same model for all the batches provide the best results?
Will the first strategy have memory requirements similar to those of running the entire dataset?

Thank you for your time!

Vic