Intel posts bug explanation

The fact that a wrong answer can result is potentially somewhat more serious than the glitch as first stated publicly by Robert Collins on his Web site Intel Secrets. Collins went no further, essentially, than asserting that a data "overflow" condition could occur which could then possibly crash a system.

Intel has stated clearly that it will not issue a recall of the Pentium II or Pentium Pro processors but, instead, has described some software work-arounds at Workaround. Computer hardware and software companies are saying that the bug is minor, for now.

The flaw was depicted initially by engineer Robert Collins. Intel's explanation is now posted at Intel.

Intel is reacting much differently to the problem this time than it did a few years back when a professor at Lynchburg College in Virginia discovered a bug in the Pentium processor. Intel, at that time, was adamant in asserting that the bug was minor or inconsequential. This time Intel will step out of the way and leave it to software and hardware companies to define the seriousness of the problem.

The problem can occur when very large "floating point" numbers are converted into "integer" numbers, something which occurs in software programs.

"Erroneous results [can occur] when the [number] is so large that it will not fit into the target data size," Intel said in its technical evaluation of the problem.

The fact that erroneous results occur supports a claim by Martin Atkinson-Barr, a physicist who works as a computer consultant and writes software programs, that the problem is more serious than originally characterized by Collins.

The glitch, as stated earlier this week by Atkinson-Barr, can "propogate" invalid data resulting in a wrong answer, an Intel spokesperson confirmed.

However, most analysts are stating at this time that the bug is minor. "Most common PC applications (e.g.,word processors, data bases, most spreadsheets, most games) do not use floating-point math and thus cannot be affected by this bug," said Linley Gwennap, editor-in-chief of the Microprocessor Report, in a written statement on the glitch. He added that most PC operating sytems such as Windows 95, Windows, NT, and OS/2 will not be affected by the bug.

Software and hardware developers also chimed in. "After initial investigation we believe our core operating systems and applications are unaffected by this processor erratum", said Paul Maritz, Group Vice President Platforms and Applications for Microsoft . "Our investigation continues, and based on the results of our findings, we will implement workarounds in the next scheduled release of any software that is affected by this processor erratum," he added.

"IBM is working closely with Intel to analyze the problem and the potential impact on IBM software. Our analysis and testing on several major IBM software offerings has shown this software to be unaffected by this issue," IBM said.

NEWS.COM first reported the flaw on May 2, stating that Collins had discovered a "major" floating-point bug in the Pentium Pro and Pentium II processors.

Atkinson-Barr said "programs that have [numbers] with values that fall within the bug range and make an attempt to convert them...will give the wrong answer and will not report any error. Fortran, C, Pascal, Algol--anything written in such a language--would be prone to this bug."

Collins had reported that the bug is not "computational" but related to a data "overflow" problem, a glitch that could cause a system to crash but wouldn't give a wrong answer pretending to be a right answer.

Based on one of the points made by Atkinson-Barr, Collins has now made a correction to his posting on the Intel Secrets Web site. "He was right. I changed it," Collins said.

The change is now highlighted in red in a section of the report called "The Nature of the Bug." Collins has dubbed the Pentium Pro and Pentium II bug the Dan-0411 bug.

Intel is an investor in CNET: The Computer Network.