Pentium is unreliable for floating-point calculations

BRANDAUER CARL M brandy at benji.Colorado.EDU
Thu Dec 1 11:24:18 EST 1994


jjb at watson.ibm.com (John Barton) writes:

...
>   Just so I don't miss out in the incredible net traffic on this one...
>I wonder what percentage of the world's numerical software is
>more accurate than this erroneous chip?  One part in 10**6?

I would guess all but the most simple forms of Basic are better than one
part in 10^6. If you are willing to wait, dc on UNIX and Mathematica can
perform nearly unlimited precision calculations.

Cheers - Carl




More information about the Bio-soft mailing list