Intel® Pentium® Processor
Statistical Analysis of Floating Point Flaw: Intel White Paper - Section 2

Section 2: Introduction
This document describes a subtle flaw inside the Floating Point Unit on certain steppings of the Pentium® processor, and characterizes the significance of this flaw for end-users of applications that are likely to be run on a computer based on the Pentium processor. The flaw in the Pentium processor causes a slight reduction in precision for certain operations in very rare cases.

Intel discovered this flaw as part of its ongoing product development and testing work. So far seven trillion computer cycles worth of testing has been conducted on the Pentium processor as part of this ongoing program. Subsequent to the discovery of the flaw, a detailed investigation was conducted to characterize both the flaw as well as its impact on the end-user base. This document summarizes the outcome of the investigation.

The overall document has 9 sections. This section 2 is an introduction. Section 3 gives a high level summary of the flaw as it manifests itself inside the processor. Section 4 explains in detail how the divide algorithm is implemented and what led to the slight flaw in early steppings. Section 5 describes the general statistical evaluation framework suitable to analyze the frequency of occurrence and the significance of the flaw on the end-user application base. Section 6 analyzes the impact of the flaw on this software base. Section 7 presents the conclusions of the characterization study. Section 8 and Section 9 contain Acknowledgments and Bibliography of References.

This applies to:

FDIV Replacement Program

Solution ID: CS-013006
Last Modified: 10-Feb-2015
Date Created: 08-Jul-2004
Back to Top