close this bookVolume 4: No. 47
View the documentFunding news and conferences
View the documentCareer advice
View the documentPentium bug update
View the documentAI languages
View the documentJob opportunities
View the documentSoftware development
View the documentEntrepreneurship
View the documentInternet news and services
View the documentChristmas services

Someone at Stanford reports that the FDIV bug often affects "bruised integers." For instance, a P-90 believes that 4.999999/14.999999 < 4.999999/15, due to an error in the 5th decimal place of the 1st calculation. [Greg Gasser (applogic@xnet.com), comp.sys.intel, 12/8/94.]

Cleve Moler, Tim Coe, and others are working with Intel to create a minimal software patch for the P-90 bug. Developers will be given code to check for divisor bit patterns that might cause errors. About 1 of every 200 divisions will require scaling both operands by 15/16. [moler@mathworks.com, comp.arch, 12/5/94.] (Or you can do every division both ways and keep the larger result.)

Pentiums free of the FDIV bug may not be available in quantity until the 2nd quarter of '95. [Jaikumar Vijayan, CW, 12/12/94, p. 1.] (Sales of Pentium-based machines have not slowed despite massive press coverage of the bug. People want the extra speed for routine tasks.)

Intel says it has yet to be notified of any harm other than Thomas Nicely's four months of bug hunting. The chief damage is to Intel, which spent $150M this year advertising the reliability of Intel chips. [Dean Takahashi, SJM, 12/12/94, 1D.]

jwelter@ibm.net reports having to spend 400 man hours reprocessing all collected data for a 0.25m aerial survey of Trinidad and Tobago. Fortunately, the $1M data-logging effort was not compromised -- that portion took 10 people, 2 aircraft, and 2 months. Clients include the Canadian government and the British Columbia provincial government. [comp.sys.intel, 12/3/94. Bill Park.]

IBM has halted its shipments of Pentium-based PCs, and will replace customers' processors on request. It estimates that a financial spreadsheet update of 15 minutes/day could encounter an error every 24 days -- or 20 errors/day for a company running 500 PCs or spreadsheets. "We believe no one should have to wonder about the integrity of data calculated on IBM PCs." [Reuter, 12/12/94.] (Intel's estimates are for a 4th-decimal error; IBM's are probably for any error at all. Note that IBM is selling an alternative chip, the PowerPC.)