||BETTER, QUICKER, FASTER
IN THIS THREE PART SERIES PETER HAYES LOOKS AT THE
TECHNOLOGY AND POLITICS THAT GOVERN ADVANCED COMPUTER
PROCESSING SPEED. TODAY - IN PART TWO - HE LOOKS AT THE
KEY TECHNIQUES THAT HELP COMPUTER CPU'S GO FASTER.
It's very difficult to underestimate the role Intel has
played within the modern world. Barring the Ford Motor
Company, no single commercial organisation can claim to
have changed the industrial landscape quite so much. As
inventors of the Integrated Circuit (IC) their place in
hi-tech history is assured - whatever the future might
However like any king, leader or dictator there are
always those that want to overthrow the status quo. Today
rival companies such as IBM, Cyrix, AMD are breathing
down the Silicon Valley giant's neck; producing central
processor units (CPUs) that are (in the view of many key
industry voices) either comparable to Intel's models or
else better priced.
The next few years seem certain be one of the interesting
periods in the whole history of affordable computing as
the playfield becomes more and more level.
In the last part of the series we looked at the various
elements that dictate the speed that a computer can
operate at. While I cannot simply repeat all that
information here, it is worth emphasising again that the
Central Processing Unit is not a complete computer in
itself, merely the most important and most talked about
Equally worth repeating is that hardware is nothing
without software. If software happened to very badly
written - for example using horrendous inefficient maths
techniques - this, alone, could nullify the advantages of
a "more advanced" processor.
Journalists are often impressed by cutting edge products,
at whatever price, but the public is becoming more and
more seduced by price. The rise of the sub 1,000 dollar
computer has played right into the hands of Intel's chief
rivals who thrive on providing far more
bang-for-your-buck. The $64,000 question - for Intel -
has to be how to respond.
Sadly Intel has not had happy experiences producing.
cut-price (non state-of-the-art) chips, a fact that they
even acknowledge in interviews. However if they simply
ignore this growing market they will certainly see a drop
in their market share, and if they do take part they are
going to have to find a way to cut costs if present
profit margins are to be maintained.
The one policy change that they have set out is to become
a more rounded high technology company, even getting
involved in the toy and easy-to-use computer market.
While a step away from the main thrust of this article,
this is a clear indication that the CPU market is not the
license to print money it once was.
What Intel must secretly hope for today is that new
breakthroughs take place that can only be performed by
the very latest set of chips: General speech recognition,
for example. But once again the snag is clear, Intel are
still predominantly a hardware company and therefore tied
to the breakthroughs of the software sector.
While Intel have started to struggle, IBM have started to
find their feet in this tricky and politic-led field. In
fact they are perhaps the first chip company to explain
their products clearly and in detail - including their
"next generation" 64-bit Power PC (TM)
The overall "big idea" is to streamline the
delivery of the target software by dividing the
mathematical workload. Their most prominent function is
to have separate units for floating point mathematics
called FPU's - by far the difficult maths that a
processor has to deal with.
This is the headline complication that would-be low level
programmers or hardware designers have to grapple with: A
CPU - which is essentially a bells-and-whistles
calculator - takes a vastly different amounts of time
dealing with the different maths functions. In fact it is
not uncommon for programmers to avoid some maths
techniques (such as floating-point) in order to gain
faster processing times.
While the full function of the PowerPC processor can only
be understood by designers or experts, the overall
flowchart is clear and straightforward. The input
instructions are delivered and modified through three
separate in built caches, and when the maths has been
split into categories (and then performed), the final
results are shipped-out through yet another special
(In abstract terms, this is a little like an efficient
restaurant kitchen that has two doors for the waiters -
"in" and"out" - and chefs that divide
the dish components between themselves. The better chefs
providing the more complicated parts.)
While not wishing to complicate matters too much further,
chip designers do not live in a vacuum. They are trying
to provide chips that run software and they have to
analyse software (and therefore software design) to find
what the common wait-states and hold-ups actually are.
Naturally they know the perennial hold-ups and the
maths-based problems, but new and revolutionary software
might create new and unique ones. In short, designers
have to know (or at least guess) as to what the software
the chip will be applied to.
Our PowerPC chip, for example, will not be much good in
say a pocket calculator (although it could be used in
one), and will far more likely be used to control a
graphic work-station or as part of an Internet server.
Equally the designer has to know the limits, and
consider, the supporting chip market. If the P3 was used
in a graphical application computer its output would be
tied to the co-processors (such as a graphic board) that
it would supply.
If the processed data is mostly screen information (as it
would be on a dedicated games computer) the chip only
needs to provide the amount of changes need for every
screen frame - what would be the use of a computer that
provided screen changes faster than they could actually
Once again we have to take a step back and realise that
while there is always applications that cry out for
speed, raw speed is not a benefit without end. In fact in
certain cases optimal speed may have already been
reached: Processors that power television controls and
microwaves for example.
Next time - in the final part of this series - we will
look forward to the future of computing and some of the
revolutionary new ideas for processing information.