Intel To SLDRAM: No Way.(Intel less interested in sampling synchronou
From ELECTRONIC NEWS (1991), May 25th, 1998
COPYRIGHT 1998 Information Access Company
San Jose, Calif.--With the announcement by Micron Technology last month
that it is sampling the memory market's first synchronous link RAM (SLDRAM)
device (EN, April 20), it looked almost as if a high-performance memory
slobberknocker between SLDRAM and direct Rambus DRAM (DRDRAM) was about to
take place. Don't get too excited, Intel said.
The reason Intel chose Rambus for its next-generation DRAM technology is
well documented. Rambus was a proven technology, had a higher performance,
and SLDRAM and double data rate DRAM (DDR DRAM) were too immature and in
their infancy at the time to judge if they would be suitable candidates.
Intel said recently it is sticking to its guns and is with DRDRAM all the
way for the next generation of PC main memory.
So with Intel laying down the law that DRDRAM would be the PC memory of the
future, these other technologies should have packed up and disappeared,
right? Not so fast. Less than a month ago, Micron's SLDRAM device was
introduced clocking in at 400MHz performance and based on a 0.25-micron
manufacturing process. In addition, Micron said the 64-megabit part is
planned to begin production in the 2H98 about the same time as the first
DRDRAM device is scheduled to be rolled out. However, the knock on SLDRAM
is that the performance is far less than the 1.6-gigabit performance per
channel per device that DRDRAM is boasting. "The Micron announcement is
insignificant and it's ridiculous to count on
one part as a viable technology when its infrastructure is so chaotic,"
said Subodh Toprani, VP of logic products division at Rambus. "There is no
way a device from two or three vendors has come close to being accepted as
the main memory of choice. Why would the PC industry go through the changes
needed to accept another DRAM technology when it can't be mass marketed and
barely hits the performance of PC-100 DRAM? It is not a viable alternative
for the PC space."
Mr. Toprani said the limitations on SLDRAM are that the device needs
software and repeaters after only eight devices while the Rambus DRAM can
go up to 32 devices per line before needing a buffer. Given this, the
performance constraints of SLDRAM and Intel's confident backing, DRDRAM is
set to be the main memory of choice, for now at least.
A Switch To SLDRAM?
"We are one of the cheerleaders for SLDRAM," said Mark Ellsberry, VP of
marketing at Hyundai Electronics America (HEA). "We are pursuing SLDRAM
vigorously but we also think that direct Rambus is going to be the main
memory of choice starting in 1999 and going through 2001. Beyond that
timeframe we will have to make some changes to enhance performance and
those changes might be to modify Rambus or switch to SLDRAM." Mr. Ellsberry
said HEA will have SLDRAM parts by that time and could very
well be a viable option in three to four years. He added that most DRAM
memory makers are involved in both technologies and should have parts for
both high performance DRAM by that time as well.
However, according to Pete MacWilliams, Intel fellow and director of
platform architecture for Intel Architecture Labs, Intel has no plans in
its future roadmap to support SLDRAM at all and plans to support PC-100
DRAM until DRDRAM makes its entrance. Also, Mr. MacWilliams disagrees that
SDRAM might have a presence two or three years down the road. "I think that
if sync link doesn't win out of the shoot, it doesn't have a place," he
said. "If in the first few generations (SLDRAM) isn't competing then they
won't have a chance. Not to say this is a bad technology it's just that
Rambus has such an advantage in timing, support, and performance." Mr.
MacWilliams said assuming the Rambus technology can be ramped up in
time and have the technology performing like it should, DRDRAM would be
good for 4 to 5 years and at that time something new will have captured the
attention of the microprocessor giant.
Mr. Toprani said Rambus plans to be at close to 4 to 5-gigabits per second
by 2005, maintaining its leadership in the high speed memory arena for
years to come. Meaning, Rambus plans to dominate the DRAM market with its
extensive roadmap to future technology and interfaces.
What About Cost?
The argument about cost is a serious one, mainly because current DRAMs are
so cheap and most consumers are no longer willing to pay expensive prices
for memory. SLDRAM boasts that it will be cost effective compared to
Rambus, however, the Rambus camp, of course, has different opinions. One of
the knacks against Rambus is that the technology is proprietary
compared to the open architectures of DDR and SLDRAM. This means that
Rambus is getting royalties on its technology and having to go to the
company itself every time an upgrade is released or a new device is
announced.
Intel admits the royalties from Rambus on DRDRAM will amount to 2 percent
from each vendor and the technology also has a die penalty that will amount
to a larger price but this should be peanuts compared to the performance
benefit, said Mr. MacWilliams. Also, compared with the added components
needed to build a SLDRAM systems, DRDRAM ends up being less expensive, he
added.
"The royalty impact is limited by our agreement with Rambus to 2 percent
and probably some DRAM guys can negotiate that down over time. The die size
is also larger which will lead to some higher costs," said Mr. MacWilliams.
"However, if you look at sync link and the components needed to get a
system up and running, the cost far exceeds that of the incremental costs
of royalties and large dies. Factor in the higher performance and cost is
no longer an issue."
Mr. Toprani agreed with Intel's assessment that the other needs and
components required to run a SLDRAM system exceed the price that DRDRAM
ends up being in the long run despite the other fees. He added that the die
size of the DRDRAM will be dramatically decreased in the next few years as
the technology moves to deeper submicron processes, eliminating the extra
cost.
However, HEA said that there are enough cost concerns and technical issues
with Rambus technology that the company needed to pursue a secondary
technology to make sure it made its bases covered, said Mr. Ellsberry. "We
are moving to a high speed interface and we are still a year away and a
lot can happen in that time frame. SLDRAM for us is a backup measure," he
added. "However, one of the places we will be focusing SLDRAM is on the
high end server and workstation market where we see a weakness in DRDRAM.
Workstation OEMs are already talking to us about SLDRAM in high end
workstations and servers. Once we get success in that segment it may
demonstrate its good for PC main memory.
"Intel does not want to change the architecture on motherboards very often
and picked one, defined that as the standard starting in 1999," said M |
