News and Analysis for the Strategic Decision Maker
Search tips Advanced search
Search | News | Careers | Classifieds | Events | About Us | Subscription | E-Wiz
Home | Latest Issue Stock Index | E-InSITE | Registration
_VIEWPOINT: January 17, 2000
A Look at DRAMs in 2000 (Part 2)
Jim Handy and George Iwanyc
Dataquest does, in fact, show significant double-data rate (DDR) shipments in the near future. Why is that? We believe that DDR will be the technology of choice in some key applications, the main one being graphics accelerators. Graphics accelerators have two virtues that support the use of extremely fast SDRAMs as well as DDR point-to-point connections.
Jim Handy and George Iwanyc are senior analysts with Dataquest/Gartner Group, a market research firm based in San Jose. The first part of this Viewpoint appeared in the Jan. 10 issue of Electronic News.
Another application that is interested in DDR is servers. There are pluses and minuses to using DDR in this application. The minus is that the widespread use of modules adds a lot of uncertainty to the amount of loading on the address and data buses. This would indicate that the DDR parts used in flexible memory systems will need to be conservatively derated, perhaps not being able to harness all the speed that DDR has to offer.
The plus is that the server, by nature of its being a large-memory machine, must have a large number of DRAM chips inside. The more chips inside the box, the more bandwidth can be achieved through such tried-and-true means as bus interleaving and bus widening. Large memory arrays have the option of supporting 1,024-bit wide bus loads from DRAM straight into a similar-sized cache line on an off-chip L2 or L3 cache. Try to do this with the nominal eight or nine DRAM chips in a typical PC's main memory! RDRAM was conceived to maximize bandwidth in low-chip count systems, and can actually add complexity to large memory systems. We anticipate seeing very large memory arrays continuing to be built with non-Rambus technologies.
What about other systems? Assuming, then, that RDRAM will take a large slice of the PC market in 2001 and beyond and given that the PC market now consumes, and will continue to consume, between 75 percent and 80 percent of all DRAM sold, one has to conclude that RDRAM will become the focus of all DRAM manufacturers' cost-reduction efforts. This same phenomenon has happened in the past. At its inception, SDRAM had a larger die size, was more costly to test, and yielded worse than the equivalent EDO chips. Look at the market now. EDO DRAMs routinely sell at the same price as their SDRAM counterparts, and sometimes command a higher price, depending more on the availability of parts on the market than on the speed of the technology. Now, if the RDRAM receives all the attention of the designers in the die-shrink group, and if the RDRAM constitutes the bulk of the market, then it is inevitable that RDRAM pricing should eventually drop below SDRAM pricing. This will then have a self-propelling effect of forcing RDRAM to be used in applications that don't need all that bandwidth, but that need to use the cheapest DRAM available. This means that anything from hard disk drives right down to the lowly digital telephone answering machine are likely to be found using RDRAMs in future years.
For 2000, though, we see a more gentle progression. RDRAMs will penetrate to reach 13 percent of the year's market over the course of the year, the bulk of this number shipping toward the end of the year. DDR will not achieve such a high number and we expect to see the difference in penetration between the two technologies increase rather than decrease. Although it is likely that DDR's acceptance in graphics subsystems will ramp faster than will RDRAM (from a perspective of the percent of all systems penetrated) we do not anticipate that overall DDR shipments will ever come near those of RDRAMs.
_
_
Copyright © 1999 Cahners Business Information, A Division of Reed Elsevier, Inc.
|
