Intel doesn't learn.....
Date: Thu, 22 Dec 94 01:02:36 PST
Subject: Intel doesn't learn.....
[Okay, my amusement (and amazement) over the Intel idiocy has nearly been used
up. I'll stop soon; I promise. This note, however, gives a bit of needed
From: email@example.com (Mike O'Dell)
Intel has announced they will replace the busted Pentium chips,
but in all the copy and such released, they *still* don't understand.
They assert that they're doing it because they're forced to and that
the problem *really isn't that bad.*
The VAX 11/780 had been out for nearly a year when DEC announced they were
replacing the FPUs in all machines. Why??? The GUARD BITS kept off
then end of the floating-point accumulator were not always being set
correctly for double-precision values. Not that bits in the real
program-visible values were wrong, but the frigging GUARD BITS caused
rounding errors in chained computations. But they knew that would
result in floating point calculations of impared accuracy,
so they fixed them.
When the original HP-35 calculator was found to have busted
transcendentals, they mailed out boxes to return them in so HP could
change the ROMs (soldered to the boards, incidently) to fix them.
When the original 360s were delivered, there was lots of grief over
the dynamic range of the floating point, so IBM *changed the
representation* to the base-16 notation it uses now, and then fixed
all the machines in the field.
Lots of machines have bugs and problems. THe vax 750 had horrible
VM-related microcode bugs, as you well know. Lots of PDP-11 models
had unspeakable grot in dark corners of supervisor operation.
the 68K family took 3 major revs to get even close to a working MMU.
the AT&T 32K required inserting a no-op between instructions
because of pipeline locking bugs. and the list goes on and on.
But they didn't get WRONG ANSWERS for arithmetic calculations.
I know of no other case where a main-stream machine got wrong answers and
the manufacturer didn't move heaven and earth to fix them.
Intel still just doesn't understand that lots of things can be broken
but fixed by the compiler. but if you do MATHS and get wrong answers,
you don't know what you can trust. and the only thing a computer has
going for it IS trust. lose that, lose the game.
© 1994 Peter Langston