Chapter 6

Numerical considerations

Abstract

This chapter gives an overview of the floating-point numerical considerations in parallel programming. It introduces the concept of precision, which is determined by the format of the representation. It further introduces accuracy, which is determined by the hardware and software implementation of arithmetic operations. This is followed by some important CUDA features that can affect the speed and accuracy of an application. It then explains why parallel execution can result in different results than sequential execution. It also explains why sorting is used when it is critical to have very accurate results. Finally, it uses pivoting in the Gaussian Elimination method to illustrate the challenges for ...

Get Programming Massively Parallel Processors, 3rd Edition now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.