Oh no! It's Intel FDIV time again!
Floating point errors back from the dead to haunt Chipzilla
As if it weren't enough that you can't buy Coppermine Pentium IIIs for love nor money, the latest list of erratanotbugs on the Intel Web site (over 60 and counting) includes one guaranteed to send a shiver down the spines of everyone at Satan Clara: "E56. It is possible that a negative sign bit may be incorrectly applied to the result of an x87 floating point operation if it is closely preceded by a Streaming SIMD Extensions (SSE) multiply operation... ...Implication: If this erratum occurs, the result of an x87 floating point instruction which should be positive will instead be negative." It is apparently possible for a BIOS mod to provide a workaround for the bug, but at the moment, any Coppermine could be doing its sums wrong. This is terribly embarrassing for Chipzilla, which still uses the original Pentium FDIV (floating point divide) fault to put a scare into new recruits and to remind employees of the inherent fallibility of man. It's not entirely clear whether this bug applies to Katmai as well as Coppermine cores. The original Pentiums were made into expensive keyrings after Intel refunded millions of dollars to users, so if you're lucky enough to have a Coppermine, maybe those nice folks at Intel are about to refund you the purchase price... The URL for the site where you can download the PDF is here. Intel adds: "Intel publishes Spec updates, listing Errata (all processors have errata) to end users and manufacturers. This errata was identified and a fix published some time ago." ®