When I was in academia I was stuck using the HPC facilities at my institution. I was lucky to get about 30 cores and my jobs took many months to run to get out

When I was in academia I was stuck using the HPC facilities at my institution. I was lucky to get about 30 cores and my jobs took many months to run to get out of the queue (overall I used about a decade of CPU time). I'm quite sure that 2 months of my time waiting for results (and the impact on colleagues occupying their HPC cluster) was worth the $9k or so it would have cost to rent enough VMs to get my results in about 4 hours or so instead of 2 months. That's not counting the hours of machine and my time wasted building enough support for the awful queueing and execution environment it used which was almost impossible to mock up locally.