Programming Massively Parallel Processors, 4th Edition

Appendix A

Numerical considerations

In the early days of computing, floating-point arithmetic capability was found only in mainframes and supercomputers. Although many microprocessors that were designed in the 1980s started to have floating-point coprocessors, their floating-point arithmetic speed was extremely slow, about three orders of magnitude slower than that of mainframes and supercomputers. With advances in microprocessor technology, many microprocessors that were designed in the 1990s, such as Intel Pentium III and AMD Athlon, started to have high-performance floating-point capabilities that rivaled those of supercomputers. High-speed floating-point arithmetic is a standard feature of microprocessors and GPUs today. Floating-point representation ...