October 19, 1998
Processor performance battles: AMD's K7 vs. Intel's Katmai
By Bob O'Donnell
As a columnist, there are times when I wish I had a simple time machine and could take
an oh-so-brief glimpse into the future to see whether current developments really turn out
to be as important as they first appear. After last week's Microprocessor Forum -- where
all the major chip industry participants laid out their future product plans -- I'm
particularly desirous of such a device.
Between the pre-show announcements and news from the event itself, there was an
absolute wealth of information on where processors -- and the personal computers they
drive -- are headed. The problem is, although companies such as Intel, AMD, Cyrix, Centaur
Technologies, and others all described their own visions of the future, no one really
knows what the collective future of the competitive market will bring.
Probably the most interesting battle that is shaping up for 1999 is between AMD's K7
processor -- which is expected to be introduced in the first half of 1999 -- and Intel's
Katmai New Instructions-enhanced Pentium II line, which is due in the first quarter. Both
chips are expected to debut at speeds of around 500 MHz and will offer external Level 2
cache on the same card as the processor -- a design similar to the current generation of
Pentium IIs. Intel's Katmai chip, however, is really only an incremental improvement over
the existing Pentium II architecture -- adding support for 70 additional instructions that
are designed to improve floating-point dependent applications such as 3-D games and video
editing. Like MMX technology, Katmai helps only applications that specifically support the
additional instructions. Given how few applications use MMX now -- nearly two years after
its debut -- it's not certain that the Katmai New Instructions (which are sometimes
referred to as MMX2, though that is not the name Intel has given them) will make a very
large impact either.
AMD's K7, on the other hand, is a fairly significant improvement over the company's
existing K6 and K6-2 (and even forthcoming K6-3) line of processors. A new internal
architecture that is designed to make the chip more efficient as well as support for a
200-MHz system bus based on the Alpha EV6 bus are just two of the many enhancements that
should make the K7 directly competitive on performance with Intel's offerings at the time.
And given AMD's pledge to price its processors 25 percent below Intel's products, a
successful launch of the K7 could even given AMD a respectable lead in the all-important
price/performance ratio.
As impressive as the specs may be, however, AMD faces a number of potential problems,
not the least of which is its own manufacturing history. The company claims that it is
past these issues, but its past record does not instill a great deal of confidence in this
prediction. Technically, AMD also has some issues to overcome. First of all, the K7 will
abandon the existing Socket 7 architecture for a new Slot A connector, which the company
claims is "mechanically compatible" with Intel's Slot 1. Dig a little deeper,
however, and you'll find that it isn't electrically compatible, which means you can't use
it on Slot 1 motherboards. The obvious question, then, is why did AMD even bother making
it mechanically compatible? It seems to me that it is just going to lead people to
mistakenly try to plug K7s into Slot 1-based systems and potentially screw up both.
Directly related to this issue is the need for K7-specific chip sets and motherboards,
as well as support from BIOS manufacturers. AMD seems to have tackled all these issues by
announcing that both it and third parties will provide chip sets that support the new K7
technologies, and that motherboard and BIOS companies have already pledged support.
Getting all these different pieces to fall into place at the right time in sufficient
volume, however, still won't be easy.
But if everything does come together, AMD will be able to offer a complete solution for
system vendors, which is bound to make it an attractive alternative to Intel's complete
solutions. (If it can just hook up with the Linux movement, it could even offer K-Linux
machines to battle Wintel.)
AMD has an enormous challenge in trying to take on Intel in the performance arena, but
if my instincts about the future do come true, the second half of 1999 will be an
interesting period for PC buyers to watch.
©
Copyright 1998, by InfoWorld Publishing Corp., a
subsidiary of IDG Communications, Inc. Reprinted from InfoWorld,
155 Bovet Road, San Mateo, CA 94402. Further reproduction is prohibited.
