DescriptionA new data type called a posit is designed as a replacement for IEEE Standard 754 floating-point numbers (floats). Unlike earlier forms of universal number (unum) arithmetic, posits do not require interval arithmetic or variable size operands; like floats, they round if an answer is inexact. However, they provide compelling advantages over floats - larger dynamic range, higher accuracy, identical results across systems, simpler hardware, simpler exception handling. Posits never overflow to infinity or underflow to zero, and “Not-a-Number” (NaN) indicates an action instead of a bit pattern.
Posit math unit (PMU) takes less circuitry than IEEE float FPU. Using lower power and smaller silicon, posit operations per second (POPS) supported by a chip can be significantly higher than FLOPS using similar hardware resources. GPU accelerators and Deep Learning processors, can do more per watt and per dollar with posits, yet deliver superior answer quality.
Series of benchmarks compares floats and posits for accuracy produced for a set precision. Low precision posits provide a better solution than “approximate computing” that try to tolerate decreased answer quality. High precision posits provide more correct decimals than floats of the same size; in some cases, 32-bit posit may safely replace 64-bit float.