On RMBS.. you might find this recent Electronic News article interesting:
"From Electronic News--May 3, 1999
Prying the Lid Off of Rambus Designs
By Peter Brown
San Jose, Calif.-- Even as debate over Rambus' DR-DRAMs heats up, OEMs are just
now getting their first look into system-level design challenges associated with the
adoption of the high speed memory interface technology.
Two key issues stand out for PC OEMs deciding if and when to adopt next generation
direct Rambus DRAM (D-RDRAM)--cost and performance. On paper, these issues are
nothing new; Rambus costs more and delivers more performance.
But there will be additional costs at the system level associated with Rambus, specifically
in terms of modules and extensions to PCB boards, according to Steven Cullen, principle
analyst at Cahners In-Stat Group. "If PC OEMs are willing to pay more for a faster
processor, then more than likely they are willing to pay to get a better DRAM," Cullen
said. "It may cost more initially but if it gives you a performance edge, why not?"
The situation is much more complex in reality, however, given the variety of PC market
sub-segment, price points and profit margins that now exist. Most PC OEMs will have to
make significant trade-offs to balance price and performance when moving to Rambus
initially and perhaps for the life of the technology, according to reports from many
industry analysts.
Besides the price premium for the chips and the royalties that DRAM vendors will have
to pay, it now appears there will be significant costs that will be associated with
integrating Rambus into computers. These costs will be balanced against potential margin
gains to help OEMs decide where the technology should be used.
"You're not going to notice much of a performance boost if any initially; but Rambus, by
nature, offers a roadmap to much higher speeds than conventional SDRAM can and that
is really the issue," said Mike Feibus, principal at Mercury Research. "Intel has a
long-term view right now. Unfortunately, computer makers have to make every quarter."
"The only place Rambus will be playing is in Pentium III systems priced at $2,500 and
up. That's the only place where the extra costs of the memory can be hidden," said Bill
Gervasi, vice-chairman of the JEDEC memory timing subcommittee and technology
analyst at Transmeta Corp, a semiconductor start-up in Santa Clara.
Those in the Rambus camp clearly deny the notion that Rambus is not for all PCs.
Rambus will find its way into value-priced PCs - $1,000 or less - by the Christmas
selling season in 2000, asserted Paul Otellini, executive vice president, general manager
of the Intel Architecture Business group.
The notion that Rambus is for everybody was seconded by John Todd, senior vice
president and chief financial officer of Gateway, who added, “We certainly have plans
for Rambus across a whole range of segments.” That included the value segment about a
year from now, he said.
Intel Corp. and Rambus Inc. both have stated the costs for D-RDRAM will be minimal
at first and will decline over time with die shrinks and yield improvements. They have not
said much about system level costs for fear of scaring away OEMs that are already
complaining about higher memory costs, analysts said. Coupled with the gag clauses
Rambus many licensees were forced to sign, the lack of information could create more
fear, uncertainty and doubt in the PC market, analysts contend.
As more PC OEMs move further to value-priced PCs, the high end is shrinking
substantially, meaning OEMs will be looking at ways to increase margins at the low level
rather than concentrating on the $2,500 and up market. Just a slight increase in price
could influence OEMs to go one way or the other at the PC fork in the road, said
In-Stat's Cullen.
Additional cost tradeoffs could come from the chip scale package Rambus intends to use
and because of a larger die size, that some reports indicate could be 15 to 30 percent
larger than today's PC-100 SDRAM.
Jeff Mailloux, DRAM marketing manager at Micron Electronics Inc., said Micron's
128-megabit D-RDRAM is 25 percent larger than the die of its SDRAM parts.
According to Transmeta's Gervasi, The Rambus logic does not scale well as the rest of
the die is shrunk, actually becoming larger as a percentage of the total die size as the rest
of the design gets smaller. That assertion is denied by Rambus, however.
If all these added costs are real, why then would high bandwidth memory that costs more
in a system be desirable given the PC market dynamics of today? The easy answer is
performance, but like so many other issues related to Rambus, the easy answer is not
always the correct answer. Rambus is boasting it will be able to achieve a 1.6-gigabyte
per second peak performance. However, issues now have been raised that applications
today may not be able to take advantage of this high bandwidth and some analysts
believe new versions of standard SDRAM will offer almost twice as much throughput.
“I'll tell you, there's a lot of Maalox being sold right now in Research Triangle Park and
Houston,” said Mercury's Feibus. “The PC makers all basically bought into Rambus.
What they're upset about is the short term. I think they are all resigned to go Rambus
eventually. They are not all resigned to cripple their high end systems for the fourth
quarter or not ship them for the long-term Rambus goal. I don't think they're willing to fly
the flag that high. Some of them do have backup scenarios with alternative chipsets for
133MHz Pentium 3 and PC133 SDRAM. And that is for the immediate term. They are
not willing to forego shipments in the short term to fly the flag,” Feibus added.
Rambus did a very good job selling their technology and explaining that it really was
going to buy some competitive advantage on the performance side, said Brian Elliot,
director of OEM sales at module maker Wintec. The reality about the cost and benefits
of Rambus, as well as the timing of its introduction, has hit some OEMs hard. “There is a
lot more pain within the computer industry than even what the media has been able to
pick up,” he added.
“There will be some applications that need Rambus. Not now but a year from now,” said
Mark Ellsberry, VP of marketing at Hyundai Electronics America Inc. Applications such
as 3-D graphics, picture phones and downloading large bundles of information from the
Internet will need high bandwidth memory, Ellsberry noted. These applications could
come as early as next year.
“How else could we make progress without having the hardware already in place,” said
Ellsberry. “Software companies are waiting for this bandwidth so they can start
introducing their high bandwidth needy software.”
It will be hard for OEMs to sort out just what choice provides the greater bandwidth
among proposed solutions. According to one analysis, double data rate (DDR)
SDRAM, the primary competitor to Rambus, features a 2.1GB to 3.6GB per second
peak throughput on a memory module, but at lower clock rate, with overall lower system
costs because such devices to handle the heat generated by the faster clock of the
Rambus devices will not be needed with SDRAM. Again, Rambus disputes this
assessment.
The performance of both DDR and D-RDRAM will be put to the test in the coming
months as both begin to emerge in systems. The comparison between the two will call
into question other performance issues such as latency, peak versus sustained
performance as well as how the DRAM will be able to perform in low power modes.
“Both 2.1GB per second for DDR and 1.6GB per second for D-RDRAM are both
possible, but you need to have the chipsets and infrastructure developed,” said Chee Ho,
director of product marketing at Infineon Technologies. “If you look today, one has that
infrastructure developed, Rambus. The other does not.”
Performance will also influence power consumption that looks to be higher than today's
SDRAM, requiring OEMs to include a heat spreader for the memory to work at full
speed. This also means an additional cost to systems manufacturers of at much as $15,
according to Gervasi.
In the coming weeks, Electronic News will try to sort the details over these specific
technical issues associated with Rambus PC design to help OEMs understand the cycle
over which they can absorb the costs or look for less expensive alternatives. |
