From the FPU to the GPU: Where Are Floating-Point Computing and Modern Processors Heading?
Floating-point computing has evolved from scalar FPU operations to SIMD vectors, massively parallel GPUs, and specialized matrix and tensor engines. Modern processors combine CPUs, GPUs, NPUs, and accelerators, using mixed precision to balance speed, accuracy, and energy. The future is heterogeneous computing focused on both increasing performance and reducing costly data movement.