Details emerge on Pentium Pro bug

Robert Collins, the publisher of the Intel Secrets Web site, this morning released a detailed explanation of what he is calling a bug, as well as software to test the Pentium Pro and Pentium II for the bug.

Intel is conceding today that the Collins report identifies a "real issue."

"When you see things that are this technically explored they tend to have validity," said Howard High, an Intel spokesperson.

However, the seriousness of the bug is far from clear at this point. Some students at the Massachusetts Institute of Technology (MIT) who looked at Collins' report claim that it is "extremely minor."

Both Intel and Collins are being extremely cagey about characterizing the nature of the problem.

Collins, a fomer Texas Instruments engineer, says that he does not know if it could have any real impact for a typical PC user. "I don't know the answer to that," he said.

He added, however, that bugs are often reported by Intel as erratum and made public and that this is not really any different than past bugs which are typically not reported in the press.

Collins says the bug is related to operations where the chip converts floating point numbers into integer numbers. Floating point operations are typically used in scientific and engineering software programs.

Collins' report explains that floating point and integer numbers are stored in different formats inside a processor. The chip may choose to convert a floating point number into an integer number but if it doesn't fit in the smaller integer format, an error can occur.

If that happens, the processor is supposed to warn the computer's software as mandated by IEEE Floating Point Standards. The problem here, according to Collins, is that the chip is not setting up a warning flag when the error occurs.

In an attempt to describe the nature of the bug, Collins compared it to a widely reported error that occured during an Arian 5 rocket launch. In that case, less than a minute into the launch, a computer on board suffered from a similar floating point irregularity that prompted it to dump its memory. "Unfortunately, this memory dump was interpreted by the rocket as instructions to its rocket nozzles. Result--boom!" Collins said.

But some observers at MIT claim this example is somewhat sensational and is not related to the problem.

The bug is also reminiscent of, but not the same as, the infamous floating point bug in the original Pentium chip. That bug, reported two years ago, resulted in very rare instances in inaccurate computations and sparked an angry backlash from users against Intel for its handling of the problem.

High said Intel engineers are now studying the Collins posting and will "fully characterize" the problem by the end of the week. If there indeed is a bug, Intel will label it as a Pentium Pro "erratum."

"It appears that the processor is not operating within the processor specification [because of the purported bug]," High said. Intel calls anything that does not operate within the specification an erratum.

Collins says he found the bug while using his own testing software on Pentium Pro and Pentium II processors after an anonoymous discoverer tipped him off to the problem. Collins has dubbed the bug "Dan-0411" after its original discoverer and the date he received the tip, April 11.

Collins also tested Intel 486, Pentium, Pentium with MMX, and AMD K6 processors but didn't find the bug in any of these other chips.

Intel is an investor in CNET: The Computer Network.