More on Intel's system on a chip. (It's nice to be home ;-))
eetimes.com
Intel ties upcoming PC-on-a-chip to Rambus
By Mark Carroll and Will Wade
EE Times
(05/28/99, 2:12 p.m. EDT)
TAIPEI, Taiwan — Intel Corp. will take a crack at putting a PC-on-a-chip
late next year. The so-called Timna processor, manufactured with
0.18-micron technology, includes a Pentium II-class processor, a graphics
controller, 128 kbytes of L2 cache and — the item that's raising the most
eyebrows — a Direct Rambus memory controller.
If Intel follows through with its plan, Timna could drive Rambus DRAMs
into the mainstream and stake out a new territory in the low-cost PC space.
But motherboard makers here, already struggling to integrate Direct
Rambus into high-end Pentium III designs this year, are worried about the
hurdles involved in bringing the new memories into volume, low-end
systems.
At the same time, Timna is reigniting concerns about Intel's extending its
prowess beyond CPUs to stake claims in graphics and memories.
With Timna, Intel is signaling that it is rethinking the pieces of the PC
architecture. The processor includes what has typically been referred to as
the north-bridge functions of a PC core-logic chip set. As for the
south-bridge components, "Intel is now talking about an I/O hub when they
talk about core logic, [because] for Timna CPUs, all the core logic does is
control I/O devices, mass-storage devices and so on," said one
motherboard R&D engineer here. (The engineer, like many other sources
here, requested anonymity in light of the fact that the Timna product has not
yet been formally announced. Intel declined to comment for this report.)
"First, they control the CPU market. [Next,] the Timna is basically a
PC-on-a-chip," said the marketing manager for another Taiwanese
motherboard maker. "Now, they want control of the memory market as
well."
Some of Taiwan's motherboard makers think this increased integration will
not succeed in the market. "For specific OEM applications, the Timna may
be a good solution," said another R&D engineer. "Historically though, the
market has not had wide acceptance of a set graphics function. Many
consumers want the ability to upgrade their video cards."
Indeed, Intel has a troubled history with integrated parts. Its efforts to build
in the 1980s all-in-one processors for the notebook market was eventually
abandoned, in part because the integrated chips failed to keep up with
rapidly changing features demanded by OEMs.
"I think the Timna is an awesomely bad idea," said Peter Glaskowsky,
senior graphics analyst with MicroDesign Resources (Sebastopol, Calif.).
Integrating so many functions, even at the 0.18-micron level, will mean
large die size and high manufacturing costs, he said, but aiming the part at
the low-cost market means low margins.
More important, the rapidly shifting graphics market demands frequent
updates, but it would be prohibitively expensive to revise Timna because of
the MPU component. "What you end up with is a part that is too big, and is
lacking in competitive features in both the graphics accelerator and the
processor cores, and those are all bad things," said Glaskowsky. The only
place such a device might succeed, he said, is the very lowest end of the
PC market. Even with a mediocre graphics core, the performance gained
by shrinking line widths could make Timna competitive with other low-end
parts.
PC makers here also said they fear the problems they are having with
Rambus on this year's high-performance desktops will now be visited on
their low-end systems as well. "The Timna raises the bar as far as what it
would take to have alternative technologies from RDRAM," another
engineer said. "If Intel's value-end PC only supports RDRAM next year,
that makes it awfully hard to fight."
Despite the board-design issues, "It makes sense for the Timna to use
RDRAM," said Dean McCarron, principal of market watcher Mercury
Research (Scottsdale, Ariz.). Using integrated RDRAM in the low-cost PC
can lower overall memory costs, since it takes less memory to deliver
adequate performance.
"One of the reasons Intel went with RDRAM is that the memory controller
is only 30 pins per channel," said one major Taiwanese motherboard
manufacturer. "That's much easier to move onto the CPU silicon than the
standard SDRAM memory controller."
Another advantage is that you can increase bandwidth with a fairly low rise
in pin count. "With one channel at 30 pins, you get 1.6 Gbytes of
bandwidth," said an engineer for another Taiwanese motherboard maker.
"With 60 pins you get 3.2 Gbytes and with 90 pins you get 5.4 Gbytes, and
so on."
While Intel's schedule calls for rolling out Timna in the third quarter of next
year, McCarron said it could take an additional six months for
value-segment PC makers to begin the switch from SDRAMs to Direct
Rambus. Intel may have to delay Timna to early 2001, he said.
"Ultimately, RDRAM will migrate down to this level of the market,"
McCarron added. "Intel's designing Rambus controllers into Timna shows
their assumption that RDRAM prices will migrate downward as well."
As they watch Intel reveal portions of its road map, motherboard makers
are struggling to get their first PCs with Direct Rambus out the door. The
second iteration of the Camino chip set (the 820 B0) — Intel's first core
logic to support Rambus — is due to begin sampling here at the end of
May.
Taiwanese motherboard makers are having somewhat mixed results with
the development of RDRAM boards. The currently high price of RDRAM
is a main stumbling block, but some motherboard makers are also having
problems getting their RDRAM boards to run.
"The current [A1 version of] Camino has problems," said one Taiwanese
engineer. "Now, it's hard for us to say if it's a core-logic problem, a
Rambus memory problem or our board design. Hopefully the next iteration
of Camino will work when we get it at the end of May. It usually takes a
couple of revisions to get a working motherboard. RDRAM boards may
take up to four revisions, though."
Another problem with the Rambus transition is the high cost of
logic-analyzer equipment Rambus requires. In late April, Hewlett-Packard
Co. staged demos here of its HP16700A and HP16702A logic analyzers
geared to address testing issues with Rambus. The systems cost about
$230,000.
"The cost is just too high for our company," said one R&D engineer. "We
currently aren't using a logic analyzer for our RDRAM boards."
"We will let Intel solve the logic-flow problems," another engineer said.
"We will concentrate on EMI and signal-quality issues of the board. For
those problems, we have purchased a high-speed oscilloscope.
"The cost of the RDRAM memory is 50 percent higher than SDRAM," he
went on. "When they have 128-Mbit parts, the cost premium will come
down to 30 to 40 percent."
Mark Ellsberry, vice president of marketing for Hyundai Electronics
America, said Timna will help drive RDRAM further down the ladder and
into the value-PC segment. "I think Intel is very interested in getting
Rambus into as many systems as possible," he said. "Over the next few
years, we will see a gradual change in virtually all PCs, and SDRAM is
going to be replaced by RDRAM. PC133 will stay in low-end systems for
the next one to two years but will be forced out toward the end of 2000,
when the price premium for Rambus memory starts to shrink."
Meanwhile, it is not yet clear where Intel is sourcing the graphics controller
for Timna or exactly what is new about the Socket 370-S, which will first
appear with the processor next year. "It's up in the air which graphics core
they will use," said analyst McCarron.
Intel recently launched the 752 graphics controller, which McCarron
described as a minor debug of its i740. "I would imagine the graphics core
[for Timna] will be something completely different," he said.
Time is on Intel's side. Between now and the time Timna rolls, it could
launch two generations of graphics chips from its Chips and Technologies
unit or even strike a new partnership, McCarron said.
RDRAM operates in a highly serial mode, which works well with burst
access but not so well with random access. This is different from much of
the current graphics technology, so interfacing any existing graphics core
with Rambus would require significant work, he added.
Glaskowsky said the core may use the same technology as Intel's upcoming
Capitola graphics device, another update of the i740 due out later this year.
Given the lengthy validation process for a complex chip like Timna, he said
the company will need to freeze the design at least a year before the ship
date and so has probably already decided on a graphics strategy.
"The best the Timna could be is a next-generation Celeron core with
Capitola graphics," he said. "I give it a less than 50 percent chance that it
will ever be released." |
