phil@sci.UUCP (Phil Kaufman) (08/22/87)
I have been watching with interest all of the discussion on the
net regarding the history of the PC and its use of Intel
microprocessors. I have also enjoyed the various semi-religious
arguments about computer architectures and related topics.
I can't speak for IBM as to why they chose the 8088 for the first
PC, but I can give you some very informed historical information
mixed with a lot of personal biases. I was in charge of much of
the strategic planning for Intel in the late 70's and was General
Manager of the Microprocessor Operation in the early 80's.
Consequently, I do know at least a little about what went on.
The architecture of the 8088/8086 was frozen in the middle of
1976! At that time all of the available useful software for
personal computer use resided either on Apple II 6502 or on
8080/8085/Z80 under CP/M. No one but Apple really saw any future
for 6502s. It was critically important for Intel both to leverage
the software base that existed in the 8080/8085/Z80 arena and the
familiarity of designers with the prior existing software and
hardware. It was also important to extend the capabilities of
microprocessors dramatically if the underlying semiconductor
technology was to be fully exploited into large market growth.
There was absolutely no desire to build yet another nice clean
plain vanilla architecture just because it could be done. From
these considerations came the 8088/8086 and its progeny.
The 8088 was created because 8-bit systems were quite adequate
for many applications, most all peripherals were 8-bits, and
8-bit systems had a cost advantage. However, it was made
absolutely compatible with the 8086 because clearly the world was
moving towards 16 bits - 32 bits was still in the far future. (The
8088 was exactly an 8086 with the addition of a byte multiplexer.
It cost more than an 8086 to build and was sold for much less
that an 8086 in order to develop the market.)
One of the things that has always, and still does, distinguish
Intel from its competitors is that Intel builds chip sets that
make computers while others build microprocessors. This is a much
more critical distinction than might be apparent. On the day that
the 8088/8086 was first offered a complete set of chips was
available to build an entire computer. To do so required that the
8088/8086 bus architecture be made reasonably compatible with
many of the existing 8085 peripheral chips. Motorola had, at
best, a CPU chip! Anyone wanting a well integrated low end
computer, i.e. a personal computer, had to chose Intel. Only
the bigger more expensive 'workstations' could afford the lack of
integration of any other choice.
The development of the 8087 floating point processor was an
integral part of the Intel plan. This effort began in 1974 and
included the investment of hundreds of thousands of dollars on
consulting from the world's best numerical analysis people and
the development of what is now the IEEE standard. No one else
comes close even today. The availability of the 8087, even if a
particular product didn't need it, was a key factor in many
choices to use the 8088/8086.
Another thing that distinguished Intel was the focus on software
and the tools necessary to develop computers and applications.
Intel had an operating system, development systems, In Circuit
Emulators, evaluation boards, etc. No one else came close.
Intel spent enormous energy getting 8-bit software converted to
the 8088/8086. That isn't a lot of software by today's standards
but it was the majority of what was available and its existence
effected a lot of decisions. It was the 8-bit software that gave
the 8088/8086 a kick-start and left all others in the dust.
( IBM would have really used CP/M for the PC if DRI hadn't
refused to give them a good OEM deal. Microsoft was smart enough
to understand who set standards and to effectively give an
operating system to IBM and make their money off of all of the
other vendors who followed IBM. )
An aside on computer architectures: Intel deliberately elected
not to build a "micro-PDP11 or micro-VAX" as Motorola and others
did. It was felt that to do so would only be a half step towards
the real future and that a bolder more risky step should be made.
Thus, many tens of millions of dollars were invested in the 432.
Remember the 432? It was a multiprocessor object oriented total
departure from any prior architecture machine. And, it was a
flop. It was a high risk, potentially high reward strategy that
didn't pan out.
Nice clean architectures are esthetically pleasing. No one, not
even Intel, thought that the 8088/8086 had a "nice" architecture
- just that it was good for the times. National had a much nicer
architecture than Motorola and you see where it got them.
So, when it came time to decide on a new microprocessor for a new
product people had a choice of the whole solution from Intel
(with an ugly architecture) or a microprocessor chip (with a nice
clean architecture) from several others. Which would you have
chosen to make your product successful? Which reason actually
swayed IBM I don't know. But, I don't think there was any other
real choice available.
The story went on with the 80286. Again, the focus was on solving
the whole problem and recognizing the realities in the world in
terms of both existing software and new needs. The 80286 was so
compatible that to this day few people have written any 80286
unique software, electing instead to spread their development
efforts over the 8088/8086/80186/80286 all at once by writing to
the lowest common denominator. Too bad, a lot of great software
could have been done that never will be.
Both the 80286 and the 80386 recognize the issue of memory
management. You simply can't have a lot of memory and modern
software without memory management. Only if memory management is
on chip is it both fast and standard for all computers built.
Look at the 68XXX. Every builder invented his own memory
management and no two systems are really compatible even though
the have the same CPU chip.
I think there are several lessons in all of this history, and
opinion. People by and large buy solutions to problems not
technical delights. Of the over 5 million PCs sold, I'd wager
that under 1 percent of the users know what the CPU instruction
set is. So, who really cares if it is elegant? Just us few folks
that have to write the low level software - and we don't buy many
machines. In fact, we'll write software for anything if the
result is that we can sell our software to a large installed base.
So, the PC world is dominated by Intel and it is going to stay
that way for a long time to come.
Phil Kaufman