AVX-512
A set of 512-bit SIMD instructions (Intel, later AMD) that process large data vectors in a single sweep. Powerful for numerically heavy workloads, but notorious for early Intel CPUs dropping the clock frequency sharply when the instructions ran.
Idea: SIMD = "Single Instruction, Multiple Data"; AVX-512 works on 512 bits at a time (e.g. 16 floats) → enormous throughput for matrices, encryption, AI, simulation. Problem: on Skylake-X and others the instructions drew so much power/heat that the whole core clocked down ("AVX offset"), so mixing a little AVX-512 into regular code could make everything slower — Linus Torvalds famously raged against it. Fragmented support: Intel removed/complicated it on consumer CPUs (E-cores lack it), while AMD Zen 4/5 implemented it cleanly and efficiently. AMD's comeback made AVX-512 relevant again. Related to SIMD and superscalar.