Intel initiates Pentium replacements

Electronic News, Jan 2, 1995 by Jim DeTar

SANTA CLARA, CALIF. - INTEL essentially put an end to the controversy surrounding the Pentium floating point unit (FPU) flaw by offering late last month to replace the microprocesors on-demand with no questions asked. At the same time at least one systems company, Intergraph Computer Systems, said it will begin shipping systems with Intel's new, bug-free version of the Pentium starting today.

Intel said it will exchange the processor for any owner of a Pentium processor-based system who is concerned about the device's FPU flaw. The company was widely criticized for offering to replace the processors only on the basis of need - as determined by Intel - rather than on request. The company was lauded by several system OEM buyers for finally making the move.

"The past few weeks have been deeply troubling," said Dr. Andrew S. Grove, Intel's president and CEO. "What we view as an extremely minor technical problem has taken on a life of its own. Our previous policy was to talk with users to determine whether their needs required replacement of the processor. To some people, this policy seemed arrogant and uncaring. We apologize. We were motivated by a belief that replacement is simply unnecessary for most people. We still feel that way, but we are changing our policy because we want there to be no doubt that we stand behind this product."

Intel will take a reserve against calendar 4Q94 earnings to cover costs associated with the replacement program.

Intergraph, meanwhile, said it will start shipping personal workstations and servers at this time containing the fixed version of the Pentium MPU. In two weeks, the company will begin implementing a program to replace Pentiums in systems previously shipped. IBM had already begun replacing defective Pentiums with the fixed versions, but also had stopped Pentium-based systems shipments and created an intra-industry dispute (EN, Dec. 19, 1994).

Wade Patterson, VP of Intergraph Computer Systems, said "Though no customer has reported any errors, we are moving as rapidly as possible to ensure that our customers can rely on the results of their computer systems."

Nevertheless, even as Intel moved to take the defective chips off the market, a Stanford University professor released results of a study which he claims back up IBM'S report that the Pentium glitch is apt to show up much more frequently in certain applications than Intel is willing to admit.

Professor Vaughan Pratt, a Stanford computer scientist, said he has made a detailed, independent study of the arithmetic bug in the Pentium and his findings indicate it may be more prevalent than Intel indicated, although he cautioned that the error rate is dependent on the application being used.

"The bottom line," said Prof Pratt, "is that the error rate is extraordinarily dependent on the application. It can range from as low as one in 40 billion for mathematical operations using perfectly random numbers, to one in a few thousand for applications that are particularly vulnerable to the bug."

Intel claims the flaw can produce errors in floating point divide operations once every 9 billion random number pairs, or about once in every 27,000 years; IBM recently estimated that a person could experience such errors as frequently as once every 24 days.

"Intel's figure is based on the assumption that a user is feeding the computer with random data," said Prof. Platt. "My approach, like that of IBM, assumes that you are running random programs. In the latter case you can encounter concentrations of high error rates." According to his calculations, a Pentium operating at 90MHz performs a maximum of 2 million divisions per second. In this case, the one in 40 billion rate corresponds to one error every six hours, while the one in a few thousand rate translates to one error every 10 milliseconds or so. These figures must be scaled, however, according to the proportion of time spent by any given application on division.

Because of the tremendous variability of the error rate that the bug can produce, only detailed studies can determine what the error rate is likely to be for specific applications, the computer scientist said. For Internet users interested in additional information, Prof. Pratt is publishing the results of his analyses on the Internet (boole.stanford.edu) under file (/pub/FDIV/README).

Meanwhile, Intel is developing a software "patch" to allow users to fix their existing Pentiums in-system if they choose to do so. The software fix will be bundled and sold with retail software packages, Intel said. The group putting together the software solution consists of Cleve Moler, chairman of Mathworks; Tim Coe, a chip designer at Vitesse Semiconductor; Peter Tang, a computer scientist at Argonne National Laboratories, and Terje Mathisen of Norsk Hydro in Norway. This patch may quickly become irrelevant though, if Intel is able to ramp the fixed versions of the Pentium rapidly enough to meet demand for replacements as well as new systems, which remains to be seen.