Floating-point unit (FPU)
The part of the processor that performs calculations on floating-point numbers (decimal numbers with a movable decimal point) according to the IEEE 754 standard. Separate from the integer unit (ALU) because floating-point math is more complex. Crucial for science, graphics, AI, and anything that computes with "real" numbers.
Background: integer operations are simple, but floating-point (representing 3.14159, very large/small numbers) requires special hardware to handle the mantissa, exponent, and rounding per IEEE 754. In the past the FPU was a separate chip (the Intel 8087 "math coprocessor") you bought separately; then with the 486DX it was integrated into the CPU. Today each core has one or more FPUs, often interwoven with the SIMD units (AVX, NEON) that do floating-point operations on several numbers in parallel — the basis for vectorized numerical computation. Performance is measured in FLOPS (floating-point operations per second). The FPU/SIMD capacity is central to HPC, simulation, rendering, and AI inference on the CPU. The complement to the ALU (integers). Related to ALU and SIMD.