SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Rambus (RMBS) News Only -- Ignore unavailable to you. Want to Upgrade?


To: MileHigh who wrote (79)12/5/1998 10:20:00 AM
From: MileHigh  Respond to of 236
 
December 07, 1998, Issue: 1138
Section: News
--------------------------------------------------------------------------------
Intel simplifies transition to new MPUs, chipsets
Mark Hachman and Sandy Chen

Intel Corp. has backed up its pledge to simplify its chipset schedule while speed-ing product introductions through improvements in manufacturing, according to the company's customers.

Intel's "transition management" program is designed to reduce the testing and other qualifications necessary to convert to a new product, especially a chipset. That strategy, as well as a hurried pace of processor introductions, was spelled out in confidential Intel product roadmaps obtained by EBN.

In effect, Intel's customers are asking the company to do more with less. For business PCs, Intel now must produce chipsets that support a larger number of available processors, without sacrificing any features. At the same time, it must also anticipate the needs of consumers, who have traditional-ly required the latest technology almost before it is produced.

Intel first tipped off analysts that it was rewriting its chipset roadmap at its November analysts meeting in San Francisco. "There was one thing that didn't resonate so well, and that was how fast we move things in terms of the architecture of the platform," said Paul Otellini, executive vice president and general manager of Intel's Architecture Business Group, at last month's meeting.

"Many of the [changes] require a qualification cycle at an end-user site, and the qualification cycles are lengthy and expensive, and the IT departments would rather not deal with them," Otellini added.

In response, Intel has lengthened the time between the introductions of chipsets for business PCs from three to six months, and now to 18 months, said Tim Teckman, director of marketing for performance desktops at Intel in Hillsboro, Ore.

For example, the first mainstream chipset for the Pentium II, the 440FX, began shipping in volume in May 1997 and was superseded by the 440LX just four months later.

The successor to the 440LX, known as the 440BX, was rolled out last April and should last into 2000, according to Intel's latest roadmap; that's a life span of more than 20 months. The 440BX is expected to underwrite a range of microprocessors, from the 350-MHz to the 450-MHz Pentium II, as well as the 450- and 500-MHz Katmai processors.

The 440BX gradually will be replaced by the Intel 820 chipset-or Camino-still scheduled for introduction in June 1999. However, sources believe there may be two versions of the Camino, one each for 100- and 133-MHz system buses. Also in June, the low-cost Basic PC will include the Intel 810 chipset-or Whitney-which connects to Intel's 370-pin Celeron chip, Intel's customers said.

"Our goal is one major transition a year, with minor upgrades as the technology and business solutions dictate," Teckman said in an interview with EBN last week. But he acknowledged that anticipating the correct features customers will require over the life of the chipset, as well as managing the more rapid transitions in consumer space, will be a challenge.

In addition to extending the life of its chipsets, Intel has pledged to increase the flow of information to its customers, who have been occasionally surprised by unexpected features, Teckman said.

Meanwhile, the company must satisfy designers of consumer PCs, who want the latest technology yesterday. "For the most part, consumers simply want faster clock rates on their processors," said Nathan Brookwood, an analyst at Insight 64, Saratoga, Calif. "Every so often that will require a platform change-[for example], the transition to Rambus [DRAM]."

Brookwood said he expects Intel to simplify its consumer chipset roadmap as well, but with continued innovation and integration in the Basic PC, which the confidential documents define as costing less than $1,100.

While Intel has pared down its chipset roadmap, the company has also made manufacturing improvements. In wafer starts per week, Intel plans to commit more capacity to the 0.18-micron equipment currently in development than it has to the 0.25- and 0.35-micron processes now in production, said Sunlin Chou, vice president and general manager of Intel's Technology and Manufacturing Group, at last month's analysts meeting.

By stepping up its manufacturing-process development, Intel will cut expenses by amortizing its costs across a greater number of dice per wafer. The process acceleration will also advance several CPU-upgrade introduction dates by a quarter.

New 333- and 366-MHz Dixon mobile chips, which are sold under the Celeron brand name, will be released in 1999's second and third quarters, respectively, and will include 128 Kbytes of on-chip cache. Mobile Coppermine chips, at 400-, 500-, and "500+"-MHz speeds, will be available in September 1999, according to the roadmap.

On the desktop, a 366-MHz Celeron will be launched Jan. 6, a quarter earlier than first projected, while 400- and 433-MHz Celeron roll-outs will be moved to the first and second quarters, respectively. A 600-MHz Coppermine chip has also been moved up three months, to late in the third quarter.

Additionally, Intel has added a 533-MHz speed grade to its Katmai desktop lineup, which is slated for a second-quarter introduction. In the workstation arena, Intel has added a 550-MHz chip to its Tanner family and a 667-MHz device to its Cascades processor line, both of which will be introduced in the second half of 1999.




To: MileHigh who wrote (79)12/5/1998 10:29:00 AM
From: MileHigh  Respond to of 236
 
DDR tugs tiger's tail -- Rambus rival will broaden memory-IC options
Andrew MacLellan

Silicon Valley- Despite Intel Corp.'s well-orchestrated campaign to guide OEMs toward a single next-generation memory architecture, a group of component suppliers last week made it clear that Direct Rambus DRAM won't be the only debutante at next year's coming-out party.

Forging a path parallel to the one blazed by Intel and its architectural partner, Rambus Inc., nearly a dozen DRAM vendors said they will ramp double-data-rate SDRAM into volume production in 1999, giving OEMs a competitive high-bandwidth memory alternative.

The manufacturers, all of which also happen to be Direct RDRAM licensees, will bring DDR into volume production at about the same time that Rambus memory is scheduled to penetrate the upper reaches of the PC market.

"This was an effort to prove to people that the DRAM industry can get together and agree on a common spec," said Jim Sogas, director of DRAM marketing at Hitachi Semiconductor (America) Inc., Brisbane, Calif. "We expect DDR to go from near-invisible to the forefront of people's thinking in the next six months or so."

Such a united front is uncommon in the competitive DRAM industry, and is all the more remarkable because processor giant Intel has so far refused to design support for DDR into its chipset roadmap.

"Intel is going to do what they're going to do," Sogas said. "But I don't think they can ignore what the rest of the industry is doing."

Although Hitachi said 2.1-Gbyte/s DDR bandwidth is possible now, Rambus said the claim is premature and will be bettered by subsequent versions of Direct RDRAM.

"Historically, we've shown a performance increase of 15% to 20% on an annual basis," said Subodh Toprani, vice president and general manager of Rambus' logic products division, Mountain View, Calif. "I can't say that we'll continue at that pace indefinitely, but we're not going to remain static."

Hitachi and other suppliers said they expect DDR to succeed first in high-end server applications, where longer design cycles evoke a more cautious technology-upgrade track. Because DDR is a close cousin to existing EDO DRAM and PC-100 SDRAM, it presents fewer technical challenges, according to suppliers.

"We think DDR will find a welcome market in the server-applications space," said Lane Mason, director of graphics/memory product strategy for IBM Microelectronics in Burlington, Vt. "That includes PC servers driven by Intel, but also goes up to high-end servers from HP, IBM, and Compaq."

So far, server divisions within both IBM and Mountain View-based Silicon Graphics Inc. have endorsed DDR in future high-end platforms.

"After thorough consideration, we believe DDR offers better bandwidth per module and lower latency than [Direct] RDRAM," said Richard Bahr, vice president of SGI's Engineering, Server, and Supercomputing business unit, in a statement issued last week. "Moreover, DDR offers lower intrinsic costs both in the DRAM itself and in the ease of integration with our current server architecture."

Toprani noted that Compaq Computer Corp. has already committed its server line to Direct RDRAM. He said any support for DDR among high-end computer companies is rash and stems from early commitments to the technology. "My belief is that subsequent to this product cycle, they will come back to Rambus because it has better performance and better price," he said.

Other DRAM suppliers, however, maintain that DDR will also compete effectively in the mainstream PC arena, a claim that could bring DDR into more immediate and widespread competition with Direct RDRAM.

"We're not attempting to stop what [Rambus proponents] are doing," Sogas said. "What we're hoping to accomplish is to let people know there are cost-effective alternatives. We're not trying to say that Rambus isn't good, but our target is definitely to penetrate the PC. We think we'll achieve that, and may the best DRAM win."

The coexistence of several DRAM variants is nothing new in a market that currently supports fast-page-mode and EDO DRAM, as well as PC-66, PC-100, and PC-133 SDRAM. The emergence of Rambus and DDR is different because both technologies will be introduced at roughly the same time, whereas previous DRAM interfaces were rolled out in consecutive order.

"There's a possibility that just as EDO forestalled the advent of SDRAM, a higher-frequency SDRAM could delay the transition to Rambus," said Jim Handy, an analyst at Dataquest Inc., San Jose. "The chances of this happening are becoming more remote, but there is still a possibility."

What's likely to happen over the next six months, observers said, is a pitched battle of performance metrics, with supporters from each camp poking holes in the claims of the other. And there is evidence that this is already happening.

While Rambus has consistently touted the 1.6-Gbyte/s bandwidth of its 8-bit-wide Direct RDRAM architecture as industry-leading, DDR proponents now claim they can deliver up to 2.1 Gbytes/s using a standard 133-MHz SDRAM DIMM running on a 64-bit-wide bus.

---

The DDR SDRAM Story

Supporters:

Fujitsu Ltd., Hitachi Ltd., Hyundai Electronics Industries Co. Ltd., IBM Microelectronics, LG Semicon Co. Ltd., Micron Technology Inc., Mitsubishi Electric Corp., NEC Corp., Samsung Electronics Co. Ltd., Siemens Components, Toshiba Corp.

Volume production:

Slated for 1999




To: MileHigh who wrote (79)12/5/1998 10:33:00 AM
From: MileHigh  Read Replies (3) | Respond to of 236
 
December 07, 1998, Issue: 820
Section: News
--------------------------------------------------------------------------------
Vendors support DDR SDRAM
Amber Howle

Irvine, Calif. -- Eleven major memory suppliers last week said they were prepared to support Double Data Rate (DDR) SDRAM. Samples of the devices already are shipping to OEMs, and DDR SDRAM modules are expected to hit the channel by the middle of next year.

Vendors officially backing DDR SDRAM are Fujitsu Computer Products of America, Hitachi America Ltd., Hyundai Electronics America, IBM Corp., LG Electronics USA Inc., Micron Technology Inc., Mitsubishi Electronics America Inc., NEC Technologies Inc., Samsung Semiconductor, Siemens Semiconductor and Toshiba America Information Systems Inc.

DDR essentially doubles the rate of data transfer from a DRAM chip to the main processor and is scalable with the current generation of SDRAM. Its major competitor, Direct RDRAM, which was developed by Rambus Inc., Mountain View, Calif., and Intel Corp., Santa Clara, Calif., requires a whole new architecture. Both technologies offer comparable performance, but analysts said DDR most likely will find a niche in the server and workstation segments over the next two years.

"[This announcement] is appealing to workstation and large file-server manufacturers," said Jim Handy, memory analyst at Dataquest, San Jose, Calif. "That's where [DDR SDRAM] is finding its acceptance."

DDR SDRAM is a good fit for large servers and workstations that need a large amount of DRAM and can support wider data buses, said Steve Cullen, DRAM analyst at Cahners In-Stat Group, Scottsdale, Ariz. Direct Rambus technology supports a 16-bit bus transferring data at 800 bits per second per pin. Although DDR only operates at 200 bps, it supports a 64-bit bus, resulting in comparable speed. Any data path less than 32 bits would be a good candidate for Direct RDRAM, while DDR would be suitable for data paths more than 64 bits, Cullen said.

Vendors advocating DDR want to quell that notion. "We spent a lot of money developing it on the expectation that it will become mainstream," said Jim Sogas, director of DRAM product marketing at Hitachi. "There is a perception that it's nichey. We're trying to kill that perception."

Copyright ® 1998 CMP Media Inc.