news
Date: 04/17 00:23 EST
Intel supports 700-MHz Rambus as Double-Data-Rate backers ready 266-MHz parts -- Rambus slows as DDR gains in DRAM race
Apr. 16, 1999 (Electronic Engineering Times - CMP via COMTEX) -- Santa
Clara, Calif. - Intel Corp. continued last week to back off on its
speed claims for Direct Rambus, leaving systems OEMs skeptical about
the value of the proprietary technology. At the same time, the
relatively quiet multivendor Double Data Rate (DDR) DRAM effort showed
new signs of aggressiveness, boasting 133-MHz DDR parts and dropping
hints of higher speeds. Overall, the message for the personal computer
main-memory market seemed to be that DDR DRAMs will soon approach the
peak throughput, and in fact beat the latency of the Rambus devices.
Intel, which initially insisted that Direct Rambus channels would
operate at 800 MHz-which is actually marketing-speak for a 400-MHz
synchronous bus transferring data on both rising and falling clock
edges-said its Camino chip set also will support a 700-MHz spec, one
that lets the DRAM manufacturers use a higher proportion of their
yields. An Intel spokesman said internal discussions are under way
about support for a 500-MHz speed bin.
In today's uncertain environment,where Intel's decision to support
the Rambus architecture is hotly debated, the 700-MHz speed grade is
seen as another indication that Intel and Rambus won't be able to
deliver the technology on time and at prices that desktop PC buyers
will tolerate. Others argue that the jump in performance Rambus
promises-originally to 1.6-Gbytes/second peak bandwidth per channel-is
the only way to entice buyers to purchase high-end systems and enable a
new class of bandwidth-intensive applications.
The spokesman said Intel had no plans to delay further the
third-quarter commercial introduction of Rambus-based desktops,
following the decision at the Intel Developer Forum to postpone the
launch by three months, until September.
Through most of last year, Intel and Rambus touted 800-MHz
performance. But with poor yields of those chips, Intel decided to
support a 600-MHz spec for less-expensive desktops.
Steve Cullen, a memory analyst at market research firm In-Stat Corp.
(Phoenix), said the 700-MHz speed "adds to the confusion" over Rambus
and said "the performance vs. the price is still questionable."
But Cullen, based in the Boston area, said the 700-MHz speed grade
will let DRAM manufacturers price their chips more reasonably. Prices
for 800-MHz parts would have been even higher, he added.
Nearly all Rambus DRAMs shipping this year will have a density of 128
Mbits, for desktops with 128 Mbytes of main memory on a single Rambus
In-line Memory Module, or RIMM. In-Stat predicts that at the chip level
the premium for Direct Rambus DRAMs (R-DRAMs) will be 20 to 30 percent,
perhaps more at the module level.
But Jim Sogas, a Hitachi America director in charge of its DRAM
business unit, said the premium will be "60 to 80 percent, nearly two
times [what it is now]. It looks pretty bad."
With DRAM makers struggling for a profit, none will price R-DRAMs
below cost, he said.
Higher price, however, may not be the only discouraging factor. Sogas
said overall performance of the lower Rambus speed grade would be no
better than today's PC100 synchronous memories. Others have said that
even with the fastest DRAMs, the difference in memory bandwidth would
be visible only in specific situations. Consequently, Sogas said,
Hitachi is pursuing the market for PC-133 SDRAMs, now shipping to
server manufacturers, and will follow with PC-266 DDR SDRAMs.
Meanwhile, the DDR camp is preparing to move for the first time in
years to a new I/O signaling standard. The first-generation DDR parts
will use the LV-TTL interface at 3.3 V. A second generation, or DDR-2,
will switch to Stub Series Terminated Logic (SSTL), a low-voltage-swing
technology already in very fast SRAMs. The signaling technology
promises both a cut in power and increase in operating frequencies in
the messy, plug-in module world of PC motherboards.
Sogas also raised questions about the heat density of the Rambus
parts. In many systems, seven of the eight chips on the module will
have to be kept in "nap" mode, while one chip actively feeds data,
increasing latency even further when another chip is accessed. The
modules will require a large heat spreader to handle thermal concerns.
Sogas predicted that Intel will reconsider its Rambus decision.
An Intel spokesman said the company is sticking by its plan for a
third-quarter introduction, though he declined to say if the Camino
chip set was sampling. Support for a 133-MHz bus between the CPU and
the system controller chip is seen as critical to the Rambus-generation
rollout, and one cause for the three-month delay was the difficulty in
supporting the faster processor bus with the Camino chip set.
Meanwhile, many DRAM vendors who have publicly supported Direct
Rambus while privately complaining of shaky technology and runaway
costs appear to smell blood. Sogas said an ad hoc group will announce
the DDR-2 specification in a few weeks. "We don't want DDR-2 to be seen
as just another Jedec option. The DRAM industry wants to coalesce on
this standard," Sogas said.
While Intel struggles to ready its Camino chip set, other companies
are working on core logic to support the PC-133 SDRAM. Via Technologies
Corp. leads Taiwan-based companies that support PC-133 SDRAMs in PCs.
After that, the transition to DDR versions would be followed by a
switch to the DDR-2 spec and the SSTL interface. VLSI Technologies,
Opti Inc., Micron Technology Inc., Acer Laboratories and FIC also are
said to be working on core-logic solutions.
The DDR camp has not forgotten infrastructure requirements. Tony
Ochoa, product marketing manager at Micro Linear Corp. (San Jose), said
his company is developing active termination ICs for DDR-1 and DDR-2
systems.
-0-
By: David Lammers
Copyright 1999 CMP Media Inc.
|
