05-21-2020, 02:59 AM
The case at hand doesn't have much to do with I/O but hashcat's scheduler/parallelisation. Without having any idea about how hashcat's implementation actually works I think you can imagine it to schedule one word (the famous "left side", important for hashcat's speed) per core, meaning hashcat will only work with twelve cores of your GPU, leaving most of the cores unused (a typical GPU has several thousand cores). Such kind of unwanted "optimisation" cannot happen when using straight wordlist attacks. You can observe similar slowdowns when using very small wordlists with lots of rules.