Output estimation in SPECFEM_GLOBAL

We are trying to run Specfem_Globe using GPU. Our GPU chips only allow max. output of 21Gb/job. But when I use NEX_XI=564, output is about 65Gb, then the job failed eventually. I wonder if there is anyway to estimate size of output?
The other question is we are doing study using PKIKP phase with 2D velocity models in CMB region. Specfem_globe seems quite computational expensive. Do we have other option?