Greetings fellow linkage people,
November 24, 1994.
In today's New York Times financial section an article appeared indicating
that Intel Pentium chips had an error in the floating point unit (f.p.u.). The
error caused rounding of numbers during some division steps such that only 5
instead of 16 significant numbers were accurate. This potentially presents a
serious problem for the types of computation necessary for linkage analysis.
Intel has known about the error since June, and have now started manufacturing
corrected Pentium chips.
Intel made some lame statement that it would affect average users only 9
times in a billion. The basis for such probability statement is unclear. _HOWEVER_
they also indicated that they would be willing to "work with scientists and
engineers if their applications were likely to be affected". The full implication
of the statement is unclear -- presumably one can obtain a corrected chip if it
is evident that one's results are likely to be affected i.e. wrong.
Just thought everyone should be aware of the potential problem. Perhaps
this is the source of some of the recent inconsistancies on DOS machines.
--
Gerard Tromp, Ph.D.
Research Assistant Professor Vox: 215-955-4487
Department of Biochemistry 215-955-4488
and Molecular Biology Fax: 215-955-5393
Thomas Jefferson University
233 South Tenth Street, Room 328 E-mail: tromp at sanger.bcm.tju.edu
Philadelphia, PA 19107
U.S.A.
