LinuxSA Mailing list archives

  From: Mike Andrew <mikero@norfolk.nf>
  To  : Linux SA <linuxsa@linuxsa.org.au>
  Date: Tue, 27 Jun 2000 09:23:20 +1130

Re: More space required

On Tue, 27 Jun 2000, Daryl Tester wrote:
> And to think, some people think PC hardware isn't at all kludgy.  :-)

In fact, there are a few near fatal design errors

the severely limited number of interrupts.

This was a disastrous decision by Intel during the 8085 (nothing to do with
PC's) where they , unbelievebly, specified interrupts active high. So bad in
fact, that but for the pc, Zilog would have wiped them out. (considering the
brain dead 64k page mode we suffered with for over a decade, too bad they
survived). (The poor design of the 8085 is not too difficult to understand,
Exxon bought out the entire original 8080 design team, leaving Intel
floundering with Turkeys. Computing has been set back 30 years due to that
single event. The hardware and cpu designs we have now are attrocious.)

Look at the ridiculous design of the Pentium II, a circuit board smothered in
plastic with a bloody great fan. And, idiots that we are, we pay premiums for
ones with bigger, noiser fans, because they run faster. I think that's called
sharp end technology.

The end result was active high interrupts required controllers for each line,
enter the 8 port (XT) followed by the 15 port (AT) interrupt architecture.
Ditto DMA. Zilog used a vectored single low interrupt line for 256, minimum,
devices. The choice of the 8086 as the flagship of the PC was so poor that it
was  ignored. Intel at one point shot themselves, and Msoft in the foot by
changing interrupt 21 to mean something else entirely. It was only the
convenience of the slotted peripherals (pinched from Apple ][ ) that brought
oem vendors on stream and made the pc a uniform platform for their product.

The INS8051 serial uart. Put out by Nat Semi originally. Even they were
embarrassed by it. The uart is so poorly designed that it rarely figured
in uart reference books. The initial 'Osborne" bibles refused to list it. 20
years later, we are still hamstrung by limited, fixed baud rates to 112k, with
control registers that do everything except tell you the state of the line and
modem.

Mode changes from real to virtual to whatever. So poorly thought out that it
requires / required the keyboard to reset the cpu. Like that's real high end
technology right?

The so called 8086 'expanded mode' (1meg address space) has been a bane. The
decision to use the top half of the memory space for peripherals could only be
made by a cretin, we have suffered non-linear memory ever since.  This again,
was due to an original flaw in the reset structure of the cpu that made code
necessary at the start of memory (motorola works from the top down). The whole
idea of 16k chunks of expanded, not extended, memory was laughable, Something
the industry had to take seriously for most of the 90's. We all paid a small
fortune for Lotus 123 and Autocad if they could use this EMS crap. 

If you explained this method to your grandkids they would not believe it
possible to so badly engineer something.

The ide controller is fatal. It imposes raw cpu signals on a cable up to 1
meter long. Fry your HD, and you are most likely to fry the cpu. I should know
<grin>

The epp/ecr parallel offerings now so popular, actually STOP the cpu when
running. No further comment necessary.

The overall architecture of the 80x86 design was so bad, that nixes of
most flavours ignored it. Preferring instead the linear address space provided
by motorola 6802/9/000/040 and cpu32. Most embedded micro controllers running
linux use the cpu32 engine in preference to the still silly offerings by intel.

Short answer? No, the good guy rarely wins, it's the ones who have the
marketing hype that win the day. I've no fite with anyone needing to make a
buck, but crap wrapped up in pink plastic, is stil crap. Everyone who has
become obsessed with Gigahertz specifications as a statement of how good
their hardware is, needs a reaiity check. Imroving speed by reducing micron
measurements is easy compared to making good design in the first place. (aka
harvard architecture). Look at the possiblilities of multi cpu, all we have are
dual-celerons, that sometimes work. No matter, we'll just tweak the speed a bit
and wait a lot faster.

Oh, and one final slap, the AGP bus (a total misnomer since it can only be used
on one card) is a stupid as the MMX technology that went before it. Both were
inspired to lift sales on old product, neither improved anything on the pc. In
fact, linux cannot detect the difference between PCI and AGP video cards, they
are, after all, the same bus underneath.


 -- 
http://linux.nf/stepbystep.htm + mirrors StepByStep submissions:
mikero@norfolk.nf

-- 
LinuxSA WWW: http://www.linuxsa.org.au/  IRC: #linuxsa on irc.linux.org.au
To unsubscribe from the LinuxSA list:
  mail linuxsa-request@linuxsa.org.au with "unsubscribe" as the subject