<!–#set var="article_header" value="The DRAM Challenge –
Vendors Gather at DDR-SDRAM Summit” –>
Introduction
So what is the deal with DDR-SDRAM? Is it going to kill Rambus RDRAM or will there be a friendly coexistence? Well, as usual, it depends on whom you are asking. At the DDR-SDRAM Summit in San Jose, the official launch of the SDR successor, all major chipset, board and graphics chips vendors united behind the new technology. Except Intel and Rambus, of course.
Everybody was there: VIA Technologies, Micron Technology, ALi, AMD, ServerWorks, Transmeta, Infineon, ATI, Nvidia, Samsung, Hyundai MicroElectronics, AMI2, and Mitsubishi. And everybody emphasized their support for DDR-SDRAM, which will bring a major increase in system performance for PC and workstation/server platforms.
In the second half of this year we are going to see quite a few DDR-SDRAM products. AMD plans to introduce their new 760 chipset supporting DDR memory (PC1600/2100). According to AMD, DDR is the most cost-effective solution for PCs and offers the significant benefit of matching processor and memory bandwidth.
VIA Technologies will introduce their first discrete DDR chipset Apollo Pro 2000 for desktop and mobile systems with AMD or Intel processors (single and multiprocessor version). Competitor ALi intends to ship an Aladdin Pro chipset and claims to be the first vendor to offer AMD and Intel DDR chipsets with Direct Intel License for Slot-1 and Socket-370 for the volume PC market. In his presentation, Senior Director of Strategic Marketing, Fred Leung pointed out that ALi’s Taiwan channel customers demand DDR-SDRAM and PC OEMs are very interested as well. He also mentioned that for example board manufacturers Aopen, Biostar and Iwill are going to build motherboards that support DDR.
Other Market Uses
Other markets are migrating to DDR-SDRAM as well: networking (server, router, switches), set-top boxes, HDTV, and of course the graphics market. Graphics processor company Nvidia is one of the largest consumers of DDR will continue to push the technology in its graphics chips confirmed Nvidia’s Product Manager Bryn Young at the Summit.
The server industry is probably the most memory-hungry market out there. In fact, servers are driving the memory market today: 42 percent of all memory goes into servers, compared to 32 percent in commercial desktops and 19 percent in consumer desktops. For 2001 the forecast for the average memory in 2-way systems is 1.9 GByte, in 4-way systems 3.5 GByte.
The Politics behind DDR-SDRAM vs. RAMBUS
It seems the industry has made their decision: DDR-SDRAM is here to stay. So why is Intel still supporting Rambus? It cannot be the supposedly superior performance of RDRAM. I will not reiterate those benchmark results here, but rather go into the politics behind the scenes. When I talked to attendees at the summit, I gathered some interesting insights.
First of all, Intel is probably the only company out there that does not need to pay royalties to Rambus. Rumor has it that Intel cut a deal with Rambus: Give us the Rambus technology for free and in return we help you to make money of the other companies, i.e. chipset and board vendors, PC manufacturers. And as the major PC processor and chipset company Intel has considerable leverage. It can force vendors to use Rambus memory by threatening to withhold information about design changes, so the vendors cannot get their products out in time for a new Intel chip generation. And it would not be the first time for Intel to use this kind of tactics.
Will Rambus Dominate the Market?
This might also be one of the reasons why the analyst community was convinced (and partially still is) that Rambus memory is going to dominate the market. Maybe they believe that because of its sheer market dominance Intel will succeed in steering everybody in the Rambus direction. George Iwanyc, Senior Analyst at Dataquest, for example still predicts that the RDRAM interface will dominate the market with 55% in 2003, DDR only gets about 20%. Semico Research on the other hand, believes that in 2004 RDRAM is going to have a 0.1% market share while DDR runs away with 57.4%.
Pretty confusing, if you ask me. It just proofs that things are not always as straightforward as they seem. Right now it looks as if Rambus has lost the battle, however. As somebody at the Summit put it: ‘The board and chipset manufacturers will only use Rambus memory if the OEMs force it down their throats.’ Nobody likes paying royalties for an inferior product if they have a better option that happens to be free. The other big concern of memory makers is that manufacturing RDRAM requires major changes in their production. The chip is bigger, meaning fewer chips per wafer resulting in lower yields. DDR memory on the other hand requires little change.
Intel’s Way Out?
Maybe Intel really wants drop Rambus but is still looking for the easiest (meaning least expensive) way out. As an insider, who frequently attended the JEDEC meetings, told me at the Summit, Intel has a reputation for letting others fight its battles. It happened at one of the JEDEC (Joint Electronic Device Engineering Circle) meetings for example, that a member company pushed for a new interface specification. After intensive grilling by the other members as to what the need for this new specification is anyway, it came out that Intel wanted it. All the while Intel sat back and said nothing. Maybe Intel is banking on the DDR proponents to push Rambus out of the market. After everything is said and done, Intel can quietly switch to DDR…
Ignoring Rambus
During the official part of the DDR-SDRAM Summit nobody talked about Rambus or Intel. When one question from the audience touched the subject, it was dismissed and left unanswered. As an insider told me later, the consensus in the industry is not to say anything about Rambus but rather ignore the company. This way neither Rambus nor Intel are forced to react, and the notion is that Rambus will just disappear. The plan might work if the rest of the industry continues to pull in the same direction.
Ultimately, it is all about what the user wants and what he or she is willing to pay for it. Rambus memory is expensive. Intel could hide the costs of RDRAM in a PC system by giving the OEMs a discount on Intel processors if they use Rambus memory. This way the OEMs can offer systems with RDRAM at the same price as systems with DDR memory. But once the end user wants to upgrade the main memory, he or she must pay the full price of RDRAM. But we’ll have to wait and see what happens when the first DDR products hit the market. My advice for users, who are looking into buying a new system, is to stay tuned and wait for the DDR benchmarks before making a decision.