From: becker@super.org (Donald J. Becker) Subject: Re: To cache or not to cache. Which machine to buy. Date: Wed, 10 Feb 1993 22:46:40 GMT
In article <HAGAN.93Feb8220332@freya.cs.umass.edu> hagan@freya.cs.umass.edu (Craig I. Hagan) writes:
>no, it makes perfect sense from a business, and an ENGINEERING point of
>view. the yield of the chip (how many of them are good -- fully functional)
>depends roughly on the size of the die (how big it is) (and other things... but)
Correct, the yield depends upon the die size to first approx. And the
_current_ 486SX chips reportedly use both a finer-line process and
omit the FPU to reduce the die size.
>the 48{6,7}sx was, IMHO, the mad brainchild of one of the engineers, as
>there will be a large amount 486 chips that have either flawed
>math units, or flawed integer units, and would otherwise have to be
>thrown out. I can just see the designer now thinking "why not sell
>them as a math unit less chip, and give me a raise?"
Wrong -- but many people are spreading this misinformation. The
_original_ 486SX and 487SX were fully functional, fully tested 486DX
dies that were delivered in the same ceramic package but bonded out
differently. In 486SX case the FPU was disabled, and for the 487SX a
few pins were swapped. When the 487SX is plugged in the 486SX is
disabled; not even the integer unit is used (and in some motherboard
the 486SX may then be removed).
A die with a flaw in the FPU is likely have other, less obvious,
flaws and dissipate more power. Since heat dissipation is a problem
with the 486DX, such a die would be discarded rather than packaged and
sold as a 486SX.
The current generation 486SX chips are a different story. Now that
the marketing test revealed hungry buyers, the 486SX parts are
reportedly produced with a simpler FPU-less mask on a different
finer-width fab line. They can be packaged in much cheaper plastic
packages since they don't have to dissipate as much power. Thus the
486SX now makes technical sense, where before it was purely market
differentiation.
-- Donald Becker becker@super.org Supercomputing Research Center 17100 Science Drive, Bowie MD 21114 301-805-7482