In short, in my discipline, if you don’t free your memory you can eat a TB in 10 minutes. Moreover, if you fragment memory too much, then you’ll start to get se

In short, in my discipline, if you don’t free your memory you can eat a TB in 10 minutes. Moreover, if you fragment memory too much, then you’ll start to get segfaults when you want large segments of contiguous memory. Lastly, the same fragmentation will cause advanced libraries like Eigen to use many small memory areas, and jumping between them kills locality, hence causing you performance losses. I’m running tasks on (and administering) clusters with similar resources to yours, yet I always treat them like 486DXs with 4MB RAM, because I can’t restart them every week.