SIMD

SIMD (Single Instruction, Multiple Data) lets a CPU do the same math on multiple numbers at once. If you have a list of 1,000 numbers and need to double each one, a normal CPU does it one at a time. SIMD does 4 or 8 at a time in a single step.

Where it matters: video codecs, game engines, ML inference, image processing—bulk numeric work on uniform data. Libraries like numpy and ffmpeg use SIMD under the hood.

Why not use it everywhere?

• Only works for the same operation on independent data—most logic is sequential or branching
• Memory is usually the bottleneck, not compute
• SIMD hates if statements
• Data must be packed contiguously in memory

You benefit from SIMD without touching it. It's plumbing—good to know it exists, rarely something you write yourself.