Memory In Motion

"RAM technology changes so fast that the memory industry has done a lot to confuse their customers," says Michael Lavoie, account executive at Kittery, Maine-based IT contractor GreenPages. "I think it's changing faster than the market really wants to bear right now."

Good news abounds for the weary. With an emerging standard in the mainstream and Rambus changing its market strategy, memory purchasing is not likely to get any more dangerous. Totally new formats are still more than a year away, but speed and capacity continue to improve within the currently supported specs. And one thing definitely isn't changing: No matter how you look at it, RAM is cheap and getting cheaper.

Heading To the Mainstream
Although there are certainly blue-chip companies such as Micron Technology and Samsung, by and large the mainstream memory market is a commodity business working to standards laid out by the Joint Electron Device Engineering Council (JEDEC). JEDEC's latest technology direction is DDR-II memory, the second generation of the Double Data Rate spec currently found in modern desktops, notebooks and servers. At the top of the DDR-I heap is DDR400, running on a 400-MHz bus and delivering up to 3.2 GBps of memory access. Although most systems currently ship with DDR333 at best, and high-performance RAM is already available in the 434-MHz band, the 400-MHz series is supported by most third-party chipsets and is on the table for Intel adoption as well. It is expected to make it into the mainstream sometime this year.

Intel famously tried to establish the narrow-pipe, super-high-speed Rambus RDRAM format as the industry standard by mandating it for early Pentium 4 designs, but that plan ultimately fell apart. Consumer resistance to the high-priced modules and pressure from alternative chipset manufacturers that built P4 architectures with DDR support have now convinced Intel to write RDRAM out of its P4 road map entirely.

id
unit-1659132512259
type
Sponsored post

That doesn't mean Rambus has given up its mandate as a high-end memory technology firm. RDRAM is currently running at 1.2 KHz and 4.8 GBps in a dual-channel architecture. SiS has picked up where Intel left off and even has a quad-channel memory controller design in its upcoming SiS659 chipset. Meanwhile, Rambus has turned the bulk of its efforts toward a new technology known as Yellowstone, targeted for embedded devices such as network equipment and game consoles, where the price of RAM is not easily distinguishable to the end user, and performance, rather than cost, can be the primary driver.

DDR In Control
Still, DDR is clearly in control of the everyday market. With sales of old-style SIMMs and slower SDRAM modules slowing, and RDRAM pushed further to the fringe, RAM vendor Kingston Technology says it expects DDR to represent the outright majority of its sales this year, up from just 27 percent in 2002.

While DDR-I will be the story this year, the DDR-II spec awaiting final ratification promises faster speeds with substantially lower power consumption than the current generation. Required power will drop from 2.5 volts to just 1.8 volts, enabling cooler operation at better speeds. Because of the change in architecture, however, DDR-II does mean a new module and socket configuration, with the same size as existing DDR-I but with new sockets and keying. JEDEC's rough road maps project DDR-II out to a 667-MHz, 5.4-GBps implementation, although market realities make the precise lifespan and evolution of DDR-II an open question. JEDEC did not originally have DDR400 in its sights as a DDR-I technology, but the push for faster memory convinced manufacturers to produce it.

That push continues, even beyond what the mainstream memory-module producers target. In the performance graphics market, both ATI and nVidia have been commissioning leading-edge, extremely high-speed memory ICs for years, and both work directly with memory makers to produce high-speed chips more suited to their applications. While ATI has published its own interpretation of the DDR-II spec back to the community, there is no word yet on adoption by the formal standards body.

Meanwhile, IT customers are taking full advantage of rock-bottom memory prices. "Memory has dropped enough that it's kind of an afterthought,people aren't afraid to put two gigs of memory into a system anymore," GreenPages' Lavoie says.

Because of massive competitive pressures, Desi Rhoden, JEDEC memory chairman and CEO of industry consortium Advanced Memory International, says some lower-tier RAM manufacturers have been recycling subspec chips back into the mainstream memory supply, threatening customer performance and VAR integrity. JEDEC plans to establish a testing and certification process to provide quality assurances at a higher level. At present, the chipset and motherboard manufacturers along with system OEMs typically issue the brand-and-model memory endorsements for their platforms. The tricky part will be establishing a sufficiently low barrier-to-entry in the certification process to keep the honest lower-cost providers in the game.

"Users like to get the cheapest they can get, understandably, but need somewhere to go to verify the quality," Rhoden says.

Trends Ahead
As long as the modules work, however, some are content to let the wheels of the industry turn. Louis DiMeglio, vice president of the solutions group at Purchase, N.Y.-based data center hosting specialist Interliant, says watching the day-to-day evolution of memory is relatively less important than monitoring the CPU and disk market.

"Speed is interesting but doesn't really turn us on that much," he says. "Most of the time if we see bottlenecks, they're on the disk side."

However, Interliant is watching trends, such as the so-called "RAID-RAM," which offers hot-swappable redundant memory modules. "One of our bigger causes of downtime with customers is failure of components like memory, so the idea that we could have a failed memory chip and not have an outage is attractive to us," DiMeglio says.

Despite the availability of DDR-II grade chips, mainstream DIMM customers will likely have to wait until at least late 2004 before the technology emerges on high-end desktop and server motherboards. Performance gaps will likely be met by eager chipmakers trying to squeeze more life and a few more megahertz out of the DDR-I spec, while chipset makers increasingly adopt multichannel memory controllers, which can provide a modest speed gain for low cost.

Meanwhile, as you wait for the latest developments from JEDEC and Rambus, don't forget to stock up on the oldies but goodies, because they're about to go out of style. "If they're using EDO [SIMMs], they either better buy up what they need now or transition over," warns Craig Tilmont, Kingston's product marketing manager.

Jason Compton ([email protected]) is a freelance technology writer based in Evanston, Ill.