> RISC vs CISC. Why you think a mainframe is so fast?

This hasn't been true for decades. Mainframes are fast because they have proprietary architectures that are purpose-built for high throughput and redundancy, not because they're RISC. The pre-eminent mainframe architecture these days (z/Architecture) is categorized as CISC.

Processors are insanely complicated these days. Branch prediction, instruction decoding, micro-ops, reordering, speculative execution, cache tiering strategies... I could go on and on but you get the idea. It's no longer as obvious as "RISC -> orthogonal addressing and short instructions -> speed".

> The pre-eminent mainframe architecture these days (z/Architecture) is categorized as CISC.

Very much so. It's largely a register-memory (and indeed memory-memory) rather than load-store architecture, and a direct descendant of the System/360 from 1964.