<!—#set var="article_header" value="Winds of Change —
Platform 2000 Conference in San Jose» —>
Introduction
For the second time this year Bert McComas, founder and principal analyst of InQuest Market Research, hosted the Platform Conference 2000 at the Silicon Valley Conference Center in San Jose, California. Again, McComas and his team had put together an informative and interesting agenda covering the latest developments in the memory, motherboard and processor markets.
According to Sherry Garber and Bob Merritt from Semico Research Corp. the DRAM market has arrived at a new inflection point. The classic DRAM migration model — memory developed for mainframes slowly making its way down to the PC — does not apply anymore. Now different types of DRAM (FPM/EDO, SDRAM, DDR SDRAM, RDRAM) are targeted at different market sectors, and each sector has different demands. While the mainframes, servers and workstations still drive the highest memory density, cost is a secondary issue. The desktop PC market on the other hand, is very cost-sensitive, and for notebooks memory should not only be as inexpensive as possible, but also consume very little power. And then there are of course also the non-computer segments like the communications industry and consumer electronics.
Especially in the consumer electronics market cost per megabit is critical, which kind of makes me wonder why Sony’s chose Rambus DRAM for the Playstation. But then different laws of economics apply for the Playstation. While other consumer electronics vendors must make money selling the hardware, Sony can afford to work with very small profit margins for the hardware, because the real money is in the software, i.e. the games for the Playstation.
DRAM Consumption
One major factor for the success of a DRAM type is the performance and availability of the proper chipset. Today we have several suppliers of chipsets, and the Intel/Rambus situation demonstrated how critical the time-to-market factor is. Intel’s 440BX/PC100 chipset is still in high demand, but it is slowly phasing out of production. The slow ramp of the Intel/Rambus memory system, and Intel’s initial decision not to support PC133, left the window of opportunity wide open for VIA and ALi. Their chipsets pretty much pushed Rambus out of the low-end PC market. ALi for example just launched their new M1647 DDR-SDRAM chipset for AMD’s Athlon processor. According to the latest rumors another company that plans to come out with DDR-SDRAM chipsets is graphics chips manufacturer nVidia.
Over the next 4 years we should see quite a shift in the DRAM consumption. While in 1999 about 50 percent of the DRAM shipments went into PCs (desktops and notebooks), this number will drop to about 27 percent in 2004, according to the Semico forecast. By then the communications industry (including settop-boxes, switches and routers) will be the biggest consumer with 24 percent, compared to only 9 percent in 1999.
In 1999 the PC industry (notebooks and desktops) was the largest DRAM consumer.
By the year 2004 the DRAM consumption of the communications industry will increase to 24%.
DRAM Market Forecast
The DRAM revenue forecast for 2000 is $31.1 Billion, up $10 Billion from last year. Semico expects another $15 Billion increase for 2001, but thinks that the market will stagnate in 2002 due to an overcapacity situation. When DRAM manufacturers switch their production lines from 8″ wafers to 12″ wafers during 2001 the DRAM supply will increase dramatically and flood the market.
As far as the DRAM types are concerned Semico still believes that SDRAM will be the winner in 2002. Apparently over 50 % of all memory shipments are going to be SDRAM components, followed by DDR SDRAM (about 18 %), EDO DRAM (about 2 %) and RDRAM (less than 2 %).
According to Semico Research Rambus DRAM (RDRAM) plays no significant role in the memory market in 2002. This analysis is based on current Rambus designs for high-end PCs, workstations and servers.
Semico based the low numbers for RDRAM on the current cost structure for Rambus memory. The company must lower the prices or come up with a less expensive solution to succeed in the PC market. I think that even though the price is important it is not the main issue here but performance. If RDRAM was indeed superior to SDRAM — which is not the case and has been proven in several benchmark tests by now — people might not mind paying more. But why would anybody buy a memory component that offers less performance for a higher price? In my opinion the Semico forecast is right on target.
Things should definitely continue to be interesting in the memory and chipset market. In mid August Intel holds its’ semi-annual developer forum (IDF) in San Jose. I’m curious to see what surprises await us there…
