Intel stumbles
by Alan Zisman
(c) 1999. First
published in Canadian Computer Wholesaler, November, 1999
Computers keep getting faster and faster, right?
So how come a computer built with a 600 MHz CPU
doesn?t feel 50% faster
than one built with a 400 KHz processor?
The problem is that not all parts of the computer
improve at the same
rate?hard drives and memory have gotten faster, but they don?t always
keep
pace with the improvements in processor speed. In particular, the speed
with which the processor communicates with the RAM hasn?t kept pace?and
if the processor can?t get the information it needs, no matter how fast
it is, it will be forced to hang around and wait.
My 1996 Pentium-166 communicated with its RAM over a
66 MHz bus. My
1998 Pentium II-400 uses a 100 MHz bus?50% faster, while the CPU is
running
233% faster. And that new P-III-600 will still be using that 100 MHz
bus.
Intel wants to speed this up. But it?s having problems
getting the industry
to do things its way.
While increases in CPU speed gets all the publicity,
the key here is
chipsets?the companion chips that support the CPU on the motherboard.
Intel?s
820 chipset (code named ?Camino?), scheduled for release at the end of
September, would have been the first from the company to support a 133
MHz bus and Rambus memory (aka RDRAM or DRDRAM?Direct Rambus DRAM),
replacing
the company?s current 440BX design, with its 100 MHz bus and SDRAM
memory.
Until recently, Intel, a major manufacturer of chipsets as well as
CPUs,
had suggested that all its future designs would support Rambus memory
only.
Despite Intel?s clout, events just haven?t gone its way this time.
While Rambus promises better performance than SDRAM,
it is also more
expensive?estimates range from 20% to about double the cost.
Price-sensitive
OEMs understandably shy away from implementing changes that will
increase
their product?s cost. RAM manufacturers, as well, are reporting
problems
getting high enough yield of Rambus memory chips, and there are
concerns
that RDRAM from different manufacturers may not be compatible. There
have
even been suggestions that RDRAM is susceptible to hard-to-detect
errors.
Word on the street is that Intel has been forced to
scale back its projections
for the new chipset?while initially, they expected it would soon
account
for 20-25% of their production, resistance to the product from OEMs has
forced them to drop that estimate to 5%, according to Ashok Kumar, an
analyst
with U.S. Bancorp Piper Jaffray, quoted in a September 21st report on
CNET.COM.
As well Intel has backed down from its plans for a
Rambus-only future,
announcing in early September that they would also support 133 MHz
SDRAM
in a future a chipset design
Taiwanese manufacturer Via Technologies is already
offering a 133MHz
SDRAM-compatible chipset, the Apollo Pro133, while Intel?s comparable
product
is not yet ready?even in advance of an official standard for SDRAM-133.
Computer manufacturer, Micron, for one, has chosen to go with Via?s
chipset
for its upcoming high-end models, citing savings of US$200-300 on the
ultimate
price tag. Rambus memory will be available in 600, 700, and 800 MHz
versions,
with some estimates suggesting that systems using the 820 chipset and
128
megs of 800 MHz Rambus memory could end up as much as US$500 more than
otherwise comparable SDRAM-equipped systems.
Intel admits that many users would, in fact, see
little difference between
models using SDRAM instead of the higher-performance, more expensive
Rambus
variety, but that on more intensive computing tasks, such as image
editing,
the differences would become apparent. At the August Intel?s Developer
Forum, InQuest Market Research?s Bert McComas reported, according to a
September 16th CNET.COM report, that Rambus systems were, in fact, 25%
slower than SDRAM systems running standard Office 2000 tasks?due to
increased
memory latency involved in RDRAM. His analysis has been questioned, but
there seems no doubt that many customers would see little immediate
benefits
from switching from SDRAM to Rambus systems?certainly not US$200-500
worth.
While choosing to ignore Intel?s 820 chipset, Micron
reports more interest
in next year?s 840 model (aka ?Carmel?)?by offering two channels of
communication
between the CPU and the memory bus, it promises to double the
bandwidth,
and significantly improve performance. However, Micron (which is a
major
RAM manufacturer as well as a computer OEM) is reportedly also working
on 166 MHz DDR SDRAM, which by doubling the current bandwidth could
theoretically
outperform even 800 MHz RDRAM
Ironically, while Rambus was being touted by Intel as
the ultimate performance
enhancer for expensive, high-end PCs, the one place the company?s
products
have found widespread acceptance is with inexpensive game systems.
Nintendo?s
popular Nintendo-64 systems have been built around Rambus memory since
debuting several years ago.
As if OEM reluctance to adopt the 820 wasn?t enough,
close to the September
release date, a new problem arose with the 820 designs. Intel was
forced
to announce the chipset could only reliably support 2 memory slots? a
3-slot
motherboard could lose data, even if the third slot is empty. This
problem
dropped the maximum allowable RAM by 50% to 512 megabytes (Future,
higher
capacity Rambus modules would raise this to 1 gig).
At the time of the announcement, somewhere between
100,000 and a million
motherboards had already been produced with three slots, and will need
to be scrapped. The problem was discovered in testing production-line
motherboards.
Perhaps as a result, on September 24th, Intel decided
to delay the announced
September 27th release of the 820 chipset indefinitely. This was the
second
delay?last February, the company pushed back the release from June to
September.
As well as supporting Rambus, the 820 chipset also
supports 4x AGP video
and the ATA66 drive interface, both standards that are double what is
supported
on the current 440BX chipset?there are concerns that a lack of support
for the 820 will delay widespread implementation of these two
performance-enhancing
standards. Intel will continue supporting the 440BX for the immediate
future,
building support for the older design into its upcoming processor
releases,
as well as pushing it for notebook designs. It?s been suggested that
440BX
models will remain in production at least through the end of next year.
As Microsoft has also learned, Intel has discovered
that just because
you?re the industry leader doesn?t mean you always get things your way.