It looks like DDR support is truly beginning to pick up steam this month. Micron will be very shortly announcing a ‘DDR Toolkit’ tailored for the design community and decision makers in the industry. The toolbox will essentially consist of data sheets and design guides for the Samurai chipset and Micron DDR modules. In addition, they will be offering the manufacturers the opportunity to order reference DDR modules and motherboards to assist in their design and testing. The representative I spoke with confirmed that the viability of the platform has been proven, as far as Micron is concerned, for both 100MHz and 133MHz DDR.
One possible disappointment for users is that the additional timing and control pins necessary to implement this memory will make it impossible to allow either standard SDRAM or DDR SDRAM to use the same socket. DDR modules will be 184 pin DIMMs, and the recommendation to manufacturers is that motherboards should have either SDRAM or DDR SDRAM support, but not both. Of course, it is very likely that manufacturers will implement a design with both types of sockets, similar to what they did to allow users to keep their EDO SIMMs and later upgrade to SDRAM when prices dropped.
Initially, prices are expected to be higher than the equivalent sized SDRAM modules, as all new technology generally is. However, the Micron spokesperson claimed that, unlike DRDRAM, there are no fundamental reasons why DDR should carry a premium over SDRAM. There are no die size penalties, and testing is no more costly than SDRAM is today. The only issue is the additional pin count. Of course, nobody in the industry is willing to make any predictions about pricing, but if DDR is accepted as quickly as SDRAM was in 1996, we should see prices drop to within a few percent of SDRAM within six to nine months after introduction.
As indicated in previous Industry Update columns, VIA will apparently be demonstrating DDR motherboards from four major manufacturers at Computex in Taiwan this June with a special pavilion focused on DDR. The Pro266 chipset is sampling right now, and is scheduled to be available in Q3. In addition, SiS, ALi and AMD all plan on having DDR capable chipsets for the Athlon available by late Q3 or early Q4.
It also appears that all major memory manufacturers are currently sampling DDR memory chips, and all claim that there have been no yield or manufacturing issues to date. Since Intel is still pushing DRDRAM very hard, none of the manufacturers is willing to make any public statements about market share or ramping schedules for DDR, but all seem to believe there is much wider industry support than for DRDRAM.
Samsung reduced prices for DRDRAM by over 30% in April, allowing module prices to drop a significant amount. Dell, who gets the best prices in the industry, is now offering 128MB upgrades for under $500 – which is still more than three times the cost of SDRAM.
In addition, a spokesperson for NEC informed me that they are now shipping between 700,000 and 1 million DRDRAM chips per month, with PC800 making up over 40% of that volume. I was also told that yields for DRDRAM were above 90% for all grades (PC600 and up), but could get no figures for PC800 specifically. The current production of DRDRAM is less than 5% of total capacity, but they also believe that DRDRAM will likely only capture 6% of the overall market in 2000. Recent industry articles have indicated that NEC is cutting back on DRDRAM production, however their spokesperson did not make any mention of that last week while discussing this subject.
Though NEC has begun shipping in volume, and Samsung has been getting improved yields,I personally do not believe that this price drop is because of a significant breakthrough in production. Instead, I believe that this is directly the result of Intel relaxing their specification to allow more chips to qualify.
Once we see DDR chipsets and modules actually available (less than 6 months, by most accounts) we should see the real battle begin. If DRDRAM has not dropped in price significantly (like, at least another 50%), it will most likely be dead, for all practical purposes, on the PC platform. There are, of course, other possible applications, but the PC platform is the one that Intel (and probably Rambus) cares most about at this time.
In my conversation with NEC this past week, I asked about VCSDRAM plans. Right now, VCSDRAM is shipping ‘in volume’, though it doesn’t seem to be readily available to end users. However, when I asked if NEC planned on implementing VC in their DDR chips, they indicated that they definitely were planning to do this. Once we see a chipset that actually implements VC memory properly, that could make for some interesting performance comparisons. I’ll try to get an update on this issue in the near future…
Be the first to discuss this article!