I once worked on a supercomputer in the olden times - this was before Linux. You basically wrote your calculation application on a front-end system with a cross-compiler. It was then transferred to the target machines’ RAM and ran there. Your application was the only thing running on that machine. No OS, no drivers, no interrupts (at least not that I knew of). Just your application directly on the hardware. Once your program was finished, the RAM was read back, and you could analyze the dump to extract your results.