JC,
Here's another article that linked off the EETimes article you posted (it's a long one, but very good). It covers both sides of the issue fairly well I think--at least it presents the arguments from both sides that we've been arguing here. I've highlighted both the positive and the negative:
-----------------------------------
Analysis: Intel positioned to accept PC133 SDRAMs as Rambus prequel
By David Lammers and Mark Carroll
EE Times
(07/22/99, 10:07 a.m. EDT)
PORTLAND, Ore. — With the introduction of Rambus-based desktops only about 70 days away, Intel Corp. acknowledged this week that the price of the Rambus technology remains too high for general adoption, and signaled its intent to support PC133 SDRAMs as an alternative and complementary memory architecture by the first half of 2000.
That scenario will create a memory technology wrestling match, played out next year in a marketplace where Intel exerts considerable sway, pitting the faster, more expensive Rambus DRAMs against the much cheaper 133-MHz synchronous DRAMs. With personal computers rapidly becoming a consumer item for which cost is a prime consideration, some analysts question whether Intel will be able to establish the Rambus technology in the mainstream PC market.
Intel's decision to "proactively evaluate" a chip set that would support the 133-MHz SDRAMs will be explained further at the Intel Developer Forum (IDF), planned for Aug. 31-Sept. 2 in Palm Springs, Calif., said Peter MacWilliams, an Intel fellow and director of platform architecture for the company. MacWilliams said Intel could quickly create a chip set for the Pentium III that would support 133-MHz SDRAMs, but said "there is a lot of noise but very little benefit" to system performance in adding PC133 support. A PC100 SDRAM with a CAS latency of two clock cycles would provide the same performance as a PC133 solution with a CAS 3 access time, he said.
Others argue that the PC133 DRAMs are a good match to Pentium III-based systems with a 133-MHz bus architecture, and come in at a cost which the mainstream PC market can absorb. Intel vice president Pat Gelsinger said he envisions the PC133 memories selling well in the smaller commercial and small-office, home-office accounts, but not in the consumer marketplace.
At IDF, Intel will disclose benchmarks which Gelsinger said will validate Intel's support for the Rambus technology.
MacWilliams said the Rambus technology "looks to be solid" for the planned September introduction of commercial systems based on the Direct RDRAMs. Intel has ironed the bugs out of its 820 "Camino" chip set and will be in "sufficient volumes" by the third quarter, he said.
"The third-quarter introduction is still on track," MacWilliams said. "It is still a go. The key ingredients are all there, the clock chips, the connectors, the RIMMs [Rambus-in-line memory modules]. We have five suppliers of the 72-Mbit RDRAMs, and five for the 128/144-Mbit RDRAMs. The key issue is that we have production versions of the chip set," he said.
A temporary transition to a PC133 generation is a distraction, MacWilliams said. Getting the industry to standardize on a single PC133 dual-in-line memory module (DIMM) will take time and resources, he said. And the performance of a PC133 SDRAM with a 3-3-3 architecture (CAS, refresh, and RAS) could be inferior to a PC100 part with a 2-2-2 specification, something that could be supported by today's BX chip set architecture, he said.
Intel's existing 440BX and 440GX chip sets both support CL2 (CAS latency of two clock cycles). Only recently have many OEMs begun to consider the PC100 SDRAMs with the faster CL2 spec. One scenario for the commercial desktops used in large businesses is to stick with the familiar PC100 SDRAMs for Pentium III-based systems, but move to the CL2 version SDRAMs to obtain more system performance.
While Intel preps the developer community for its dual-track approach, Via Technologies Inc. announced its PC133 chip set, dubbed the Apollo Pro133. This chip set supports both memory and front-side-bus signaling at 133 MHz. Intel is set to release its 0.18-micron Coppermine Pentium III CPU late this year, with support for a 133-MHz processor and memory bus.
"PC133 is not only the smartest upgrade choice for board-level designers, system assemblers or end-users seeking a performance boost, it's the only choice," said Wen-chi Chen, president and chief executive officer of Via. "Apollo Pro133 is kick-starting a new and exciting platform evolution that makes sense in today's performance-hungry but cost-sensitive PC marketplace. This standard will have a profound impact on the industry well into next year."
Via's Pro133 consists of the VT82C693A north bridge chip in a 492-pin BGA and the VT82C596B south bridge chip in a 324-pin BGA. It is priced at $29 per unit in OEM quantities.
Taiwan's motherboard vendors are not yet very excited about PC133. "We are currently shipping one PC133 board," said one project manager. "Demand is low for now, as there is no performance increase until Coppermine becomes available in November."
PC133 does have a large cost advantage over RDRAM, however. "We now offer a PC133 128-Mbyte DIMM at $143 per unit," said Robin Chang, associate director of product marketing for Acer Technology Inc. "In comparison, our 144-Mbyte RIMM module costs about $840."
But, with consumers receptive to the appeal of faster-megahertz memories, and cost considerations ever more important, sources said Intel almost surely will take its chip sets to the PC133 speed. The sheer cost difference between the SDRAM and Rambus solutions continues to plague Intel's decision, made in December 1996, to support the Rambus technology.
MacWilliams said the recent sharp decline in the price of the PC100 SDRAMs — albeit with a small bounceback over the past 10 days — has widened the price gap between the SDRAMs and Rambus DRAMs. "The DRAM vendors are losing lots of money on SDRAMs," MacWilliams said. "As prices go down they are less willing to track the same aggressive pricing with RDRAMs."
While the additional cost of making an RDRAM is estimated by Intel to be from 20 percent to 30 percent higher than SDRAM at most DRAM vendors (one vendor said its cost adder was 50 percent), the prices charged for RDRAMs in the early days of the marketplace are four to five times as much as for SDRAMs. And while DRAM vendors try to charge as much as possible for RDRAMs, the collapse in the SDRAM price has made the difference between the two memory types particularly glaring, MacWilliams said.
DRAM vendors are charging a stiff premium for RDRAMs running at 400 MHz, which delivers 800 Mbits per second by reading data from both the rising and falling edge of the clock. Estimates for the price of a 128-Mbyte RIMM vary considerably. Intel said such a module would cost about $200. A DIMM of similar density, populated with 64-Mbit PC100 SDRAMs, would cost about $80.
One small module maker in Taiwan said its 128-Mbyte RIMM sells for $840; Intel officials dismissed that price as an aberration. Hyundai Electronics America said it is currently selling 128-Mbyte RIMMs "in the high $300-to-$400 range," said Farhad Tabrizi, the company's strategic memory marketing manager.
The price of the 128-Mbyte RIMM should drop to $250 by the fourth quarter, at which time the 128-Mbyte PC133 DIMM will sell for about $100 to $125, Tabrizi said. The 100 percent premium will reflect the higher cost of the RDRAM package, the more expensive testers, and the larger die size of the RDRAMs, he said.
One Taiwanese engineer said the current high price of RDRAM is a transitory trend. "The pricing for RDRAM now is just ridiculous," said the sales engineer. "We recently paid $400 for a 400-MHz 128-Mbyte RIMM sample. Personally, I think the DRAM vendors are trying to make excessive profits on RDRAM now to counter the current price fall of SDRAM. The current price of RDRAM doesn't reflect the real price when volume production begins," the engineer said.
The number of DRAM vendors serious about Rambus technology is expected to increase later this year.
More RDRAM suppliers coming
Hyundai and LG Semicon, both Rambus supporters, are currently in the process of merging their semiconductor operations. Toshiba Corp., an early backer of Rambus technology, will supply much of its RDRAM production to Sony Corp. for use in the Playstation 2 video game machine. NEC Corp. and Samsung Electronics are other major suppliers today. But four more DRAM vendors will enter the RDRAM market in the next quarter and will be validated by Intel and Rambus, said Subodh Toprani, vice president of Rambus Inc. (Mountain View, Calif.).
Avo Kanadjian, vice president of memory marketing at Samsung Semiconductor Inc. (San Jose, Calif.), said, "as far as Samsung is concerned, we remain unchanged in our pricing and volume plans. What has happened is that the reference point — the price of the SDRAM — has changed. That unfortunate gap has made the Rambus premium look twice as large."
Samsung expects the RDRAMs, primarily at the 128-Mbit density, to have a 50 percent premium over the equivalent PC133 solution. But that equation assumes that the SDRAM price will recover to the $8 range for a 64-Mbit SDRAM, he said.
For the near term, Tabrizi said the PC133 DRAMs fit well in the mainstream, low-cost PC market, where a fast processor, a 133-MHz system bus, and 133-MHz SDRAMs provide adequate performance for today's mainstream applications.
Next year, Hyundai estimates that only about 7 to 17 percent of all DRAMs shipped will be RDRAMs, hampered by continuing high costs for packaging and testing.
"One reason for the cost difference is that the Rambus DRAMs require a new infrastructure at the back end," Tabrizi said. "To produce one-million RDRAMs a month requires an investment of $10-to-$20 million. Hyundai currently makes about 40 million 64-Mbit DRAMs per month, and to convert our capacity to the Rambus memories would cost a minimum of $400 million and perhaps as much as $800 million. On the other hand, in the case of PC133 we already have the testers in place.
But there is no doubt that, with Intel behind it, Rambus will become established. Hyundai is behind it."
Sherry Garber, a senior vice president at Semico Research Corp. (Phoenix), said recent events have led her to question whether the Rambus architecture will ever really get off the ground. The cost of manufacturing an RDRAM remains too high, and the market is shifting toward commodity desktops at a very rapid pace, she said. That makes Rambus a niche market even into 2001.
Garber said Semico's latest estimates call for 182.9 million Rambus memory units to be shipped next year, nearly all at the 128-Mbit density. That will account for only 6.2 percent of total DRAM units shipped. In 2001, according to Semico, that ratio will increase to only 7.8 percent.
By then the 256-Mbit generation will start to become a factor. "SDRAM manufacturers already are coming down the learning curve on 256-Mbit DRAMs for the server market, and Rambus won't have a 256-Mbit density part for quite some time" Garber said. "Can they be ready with a 256-Mbit RDRAM by 2002? That's the fear among the RDRAM vendors."
A PC maker's choice of memory technology is increasingly driven by marketing, said Steven Przyblski, principal consultant at Verdande Group (San Jose, Calif.). "The PC133 solution offers very little, in some cases no, performance improvement," he said. "Its only significant value is that it gives the marketing people something to use to argue that they have a leg up over the competition."
Based on that pressure from the PC OEM marketing community, Intel is very likely to support a PC133 solution in 2000, Przyblski said. That decision, in turn, "might have some impact on the speed of the RDRAM ramp-up and the overall migration to the Rambus technology."
But Przyblski said he still believes Intel will succeed in establishing RDRAMs in the mainstream. "It is more a question of when, rather than whether," he said. |
