If you think predicting the direction of the DRAM market is tough, then try to forecast when the Rambus memory will ramp up. Timing will be everything with the next-generation Direct RDRAM. That's for everyone from the chip makers, module suppliers, tester and packaging vendors to OEMs and Intel Corp., which has spent many millions of dollars in an all-out attempt to make Rambus the standard wideband chip.
The problem is that trying to guess the Direct RDRAM timeline is like throwing darts at a calendar.
Making a mistake could be very costly to any of these folks. Device makers have to commit to a costly equipping of their production lines with a bevy of new front-end processing tools, high-speed testers, special tooling and handling equipment, and chip-scale packaging gear.
But if chip makers launch full-scale production too soon, the new market will begin with a costly inventory glut. Starting too late, however, means they could miss locking-in early major customers.
Equipment vendors are caught in the middle. They must have special Direct RDRAM tools ready to go when the memory suppliers suddenly decide to jump into the new market. But if the new wideband DRAM keeps getting pushed back, this market will start looking like 300-mm deja vu all over again with the tidal wave of red ink.
Unfortunately, the Direct RDRAM crystal ball has gotten cloudier in recent days. Intel was supposed to be making its big launch this spring, bringing out its 500-MHz Pentium III processors, a 133-MHz processor bus, and the Camino chip set. At the same time, a gaggle of memory producers would be ready to supply OEMs production quantities of Direct RDRAM.
But Intel was forced to acknowledge in late February that the Camino and Direct RDRAM products had been delayed till the end of the third quarter. And it could take longer.
There are several reasons for this setback. Intel reportedly is late with the Camino chip set. As for the RDRAM, only a smattering of samples so far have made it to market. And production quantities from most suppliers aren't in the picture until late fall at the earliest.
Finally, precious little applications software has been written yet to take advantage of Direct RDRAM performance. These programs definitely will be necessary to stimulate demand for the new RDRAM-equipped machines.
Perhaps the hardest thing for anyone to divine here is the street price the new chip will carry at its launch. Informed speculation varies widely - from a 15-to-20% premium all the way up to an 80% price penalty over present SDRAMs.
Figuring out that initial price will be very important since it - not any amount of Intel hype or pressure - will determine how fast that Direct RDRAM will penetrate the memory market.
A high premium will slow things down - relegating the new wideband memory chip to modest quantities in the high performance workstation niche market. This would not make it easy for suppliers to crank up production to drive down prices so the part could move into the mainstream desktop PC market.
Despite its many backers, Direct RDRAM has a competing wideband chip, double data rate SDRAM, that's growing stronger by the day.
But Intel has put all its memory chips on Direct Rambus. Its new Camino chip set, called 820, supports only Direct RDRAM, although Intel is said to be developing an alternative 815 chip set that would enable its new Pentium-III processors to work with PC100 SDRAMs.
And there may be other chip sets that will make life more difficult for Rambus. Three Taiwan-based chip set makers -- VIA Technology, Silicon Integrated Systems, and Acer Laboratory -- claim they will have Camino-equivalent chip sets shortly that will link Pentium IIIs with either PC100 or PC133 SDRAMs.
The danger here to Direct RDRAM is that a large installed base of high-speed SDRAM memory with Pentium processors could easily be upgraded to DDR-SDRAMs at a potentially lower cost. That could cut heavily into the potential memory market eyed by Direct Rambus backers.
There will be even more competition for Rambus when Advanced Micro Devices Inc. launches its next-generation K7 MPU this summer. While the K7 will be able to link to Direct Rambus, AMD will also offer an interface to SDRAM, and eventually to DDR-SDRAM memory.
The latest slippage in Direct Rambus is now causing some analysts to consider revising their forecasts. Sherry Garber, analyst for Semico Research Inc. in Phoenix, may pare back her Direct RDRAM estimates.
She originally had estimated that 30 million Direct Rambus chips would be shipped this year, or less than a quarter of the 127 million DDR-SDRAMs sold in 1999. Next year, she had predicted that 300 million Direct RDRAMs and 589 million DDR chips would ship.
Intel has been trying hard this winter to jump-start Direct Rambus with a flurry of cash infusions to memory makers to help them pay for major investments they will have to make in equipment and production lines. The chip giant invested $500 million in Micron Technology and $100 million in Samsung Electronics, as well as offered to make major investments in Toshiba, NEC, and Mitsubishi.
But some DRAM makers don't want to move too quickly. Executives from these vendors fear they could get into trouble quickly if they move too fast. If they ramp up Direct RDRAM production lines before the market picks up, they will end up with excess capacity at the very beginning of the product life cycle.
That could propel Direct Rambus directly into a price war. While the resulting cut-throat pricing could spur demand for the new memory chip, it could also drop producers into a sea of red ink. That's not exactly the way that vendors want to establish the Direct RDRAM market.
The high-cost of moving into Direct RDRAM may also force some DRAM producers that are now under heavy financial pressure to delay making the big investment in Direct RDRAM - at least in the early stages when shipments of the new chip aren't expected to be large enough to pay back the investment.
That could help to speed up the segmentation that's beginning to occur in the DRAM market. There would be a mixture of Direct Rambus players, those pushing Double Data Rate, suppliers going after niche markets, and a few making older trailing edge DRAMs where a market still exists.
For now, everyone is waiting in the Direct Rambus guessing game. Memory makers are waiting on large scale OEM commitments, OEMs are waiting for the outlook in pricing and supply, equipment vendors are waiting on tool orders, and everyone is waiting for Intel's next move.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.