I just used the OS. Not the same sort of problem you are doing but up to 4,000,000 nodes. For Pastix I had to get a version compiled in I8 for much beyond 500,000 nodes and for lots of nodes, beyond 1,000,000 or so it is slower. Apparently the index numbering on an array goes beyond 4i at about that point. I also have run 1 problem “Out of Core” which requires some set up. I routinely run 2,000,000 nodes, but I have 64GM of ram. Thinking about what to do for my next PC, it needs more memory, but also more memory channels as for some of these big problems it is just too slow, and each channel can’t effectively feed more than 3 cores. (I run an AMD 3700x with 64 GB).