How does it compare to simple binary I/O? Many studies have shown that the C function atof(), which your read_csv() function is likely calling at low level in the C based runtime, is where time is being lost. Avoiding that with binary I/O has been going on since the 1960s ;-) plus you gain the benefit of precise storage. Of course, there are machine architecture portability issues, which incentivized non-binary exchange formats, but if you’re talking about speeding up your own work that’s not an issue.