SIMD (Single Instruction, Multiple Data) is a class of parallel computing hardware and instruction sets in which one instruction operates simultaneously on multiple data elements packed into a wide register, rather than processing them one at a time. It is implemented in CPU vector extensions and underlies the lock-step execution model of GPU cores. Numerical libraries such as NumPy and GPU programming frameworks rely on SIMD execution to accelerate array and tensor operations.