floating-point unit
A microprocessor that performs floating-point arithmetic. In older models of computer the FPU was not integrated into the central processing unit, but rather as a coprocessor if included at all.
FPU: the chip that handles decimals and tiny numbers fast
A floating-point unit is a specialized processor core, either integrated into the main CPU or operating as a separate coprocessor, that executes arithmetic on numbers represented in floating-point format. Unlike fixed-point arithmetic which handles whole numbers and simple fractions, floating-point notation uses a mantissa and exponent (similar to scientific notation) to represent very large, very small, and fractional values with a fixed number of bits, typically 32 or 64 bits per number.
In industrial control systems, signal processing, and numerical simulation, the FPU performs multiply, divide, square root, and trigonometric operations far faster than the general-purpose CPU could manage through software emulation. A processor executing such instructions without an FPU must break the operation into many integer steps, taking hundreds or thousands of cycles. With a dedicated FPU, the same operation completes in single-digit cycles, cutting execution time from milliseconds to microseconds for compute-heavy algorithms.
Historical separation and modern integration
Through the 1980s and into the early 1990s, floating-point units were physically separate chips, optional coprocessors added to systems that needed them. Intel's 8087 was a standalone FPU for the 8086 processor; the 386 and 486 families came in variants with or without integrated floating-point hardware. By the Pentium generation and across modern ARM and RISC architectures, the FPU became a mandatory part of the main die, eliminating the penalty of inter-chip communication and making mathematical computation routine rather than specialist.
In embedded industrial applications running real-time control loops, feedback calculations, or sensor fusion, the presence or absence of an FPU determines whether a computation finishes within a guaranteed time budget. Systems without an FPU must either use integer approximation (introducing quantization error), run slower, or upgrade to more expensive hardware. Modern microcontrollers commonly include single-precision (32-bit) FPUs; double-precision (64-bit) units appear in higher-performance industrial processors and safety-critical systems where numerical accuracy must exceed typical sensor resolution.
Floating-point arithmetic carries its own gotchas: rounding errors accumulate in long calculations, comparison operations must tolerate small tolerances rather than equality, and some numbers (such as 0.1 in binary) cannot be represented exactly. Industrial software engineers must account for these limitations when writing control algorithms, especially in systems where precision drift could trigger false alarms or instability. The IEEE 754 standard defines how conforming FPUs handle rounding, overflow, and special values like infinity and NaN, ensuring behavior is predictable across different processors.