Lance, if you invest in this stock, the real question is: What will be the next mainstream memory in the PC platform in 1999 and beyond? Since the stock has a market cap of over a billion dollars with only $40 million in revenue, quite a few people are already betting on Rambus.
The more you read, the more likely you are to conclude that they are right. It seems very clear that Intel is betting that way, and the momentum is already enormous. If you do a some searches under Yahoo for various combinations of Intel, Rambus, Merced, Katmai,etc. , you can find some interesting reading.
Remember the DRAM market is the tail, and Intel(and a few others) are the dog.
A few links follow, none of them particularly recent.
techweb.cmp.com
Intel predicts Rambus DRAM will be a PC standard by 1999
By David Lammers
SEOUL, South Korea -- Intel Corp. is expected to predict by year's end that the second-generation Rambus DRAM (RDRAM) will be the main-memory chip of choice for 1999. The microprocessor giant dropped a strong hint to that effect to the world's top 10 DRAM vendors last week as a contingent of Intel and Rambus technologists traveled Japan and South Korea.
The decision would appear to resolve the competition among the RDRAM, the next-generation synchronous DRAM (SDRAM) and the proposed SyncLink DRAM for a place in the 1999 PC. That system, powered by Intel's Merced CPUs and supporting multiple very high-speed ports into main memory, will have memory-bandwidth requirements far beyond the abilities of current extended-data-out (EDO) DRAM and SDRAM modules.
Intel has yet to make a definitive statement on the issue. But with the company's apparent commitment to RDRAM, it appears that the other technologies will be largely restricted to non-PC applications: from consumer entertainment to communications. Those are not small markets, but an inside track for PC main memory could give RDRAM such economies of scale that it would be difficult for the other alternatives to compete.
Senior semiconductor executives at Samsung Electronics, LG Semicon, NEC and Toshiba -- four companies that have been particularly active in RDRAM development -- said Intel has presented a technology road map that calls for 66-MHz synchronous DRAMs (SDRAMs) to be in widespread use in 1997, working alongside Intel's Pentium Pro and MMX multimedia-extension-enabled Klamath processors.
search.zdnet.com
Memory chip makers are up in arms about recent deals between Rambus and Intel, claiming that Intel is making another run at cornering the memory market, as it had attempted in late 1995.
What's happening is that Intel is locking vendors out of the specification development in such a way that it can bring high-performance Rambus-based systems on the market before its competitors get hold of the specifications. Rambus is essentially command-driven smart memory that coordinates with the CPU more cooperatively than old-fashioned "dumb" memory. It's argued that this type of memory design is necessary in high-speed machines. But some see Intel's plan more as a marketing strategy than a technology. It is believed that Intel can synchronize certain microprocessor chip designs with new memory technologies in such a way that by changing certain specifications on the fly, this technology can ruin the performance of Cyrix, AMD, and even Motorola-based systems while improving its own. In other words, change parameters and protocols at a critical moment after other processor houses have committed to an earlier design.
Currently Intel controls the microprocessor market, the support chip set market, and the motherboard market. It now sees the memory market as another one in which it should profit. But Intel has no reason to go into manufacturing memory if it can just get a piece of the memory makers' pie. It will simply charge a royalty to memory makers to use the Rambus technology. And that royalty may be as high as 6 percent of sales--a healthy chunk. Furthermore, it's rumored that just to get into the Rambus bed, a memory maker might have to put up a $3 million to $6 million "tribute."
This whole deal essentially takes much of the profit and gives it to Intel/Rambus. Intel likes the idea of licensing as much of the computer as it can. You have to wonder what ever happened to "open systems." The PC is becoming more like a Nintendo. The Slot 1 technology needed for the Pentium II, for example, will also be licensed to siphon profits away from motherboard makers that compete with Intel. You can see why the K6 is so appealing to smaller players with its Socket 7 compatibility.
This Rambus situation is the worst, though, according to sources inside the memory business. To counter this threat, an ad hoc specification for a Rambus-like architecture is being developed in a hurry. This is called Synclink DRAM (SLDRAM), and information can be found at www.scizzl.com. This site opened originally to promote what is called the "The Scalable Coherent Interface (Local Area MultiProcessor)." According to the site, this "is effectively a combination computer backplane bus, processor memory bus, I/O bus, high-performance switch, packet switch, ring, mesh, local area network, optical network, parallel bus, serial bus, information sharing and information communication system that provides distributed directory-based cache coherency for a global shared-memory model and uses electrical or fiber optic point-to-point unidirectional cables of various widths." What a mouthful. This alone will keep you reading for hours.
The SLDRAM Consortium includes all the memory makers as well as Apple, IBM, Motorola, and most other semiconductor makers except Intel. Whether this crowd can get anywhere remains to be seen. Anyone who watched the VESA-bus folk get decimated by Intel's PCI bus will understand Intel's momentum. PCI blew away not only VESA but ISA and EISA, too.
techweb.com
The Katmai processor marks a monumental effort that will involve the convergence of new chip sets, DRAM technology, a graphics bus, MMX instruction set, graphics components, and perhaps a faster system bus. At the same time, Intel has stepped up efforts to ensure a base of software tools and applications that will take advantage of the revamped graphics subsystem.
"There's a tremendous amount of work with how the CPU will interact with memory, chip set, and graphics devices," said Jonathan Khazam, marketing director for graphics programs at Intel in Santa Clara, Calif. "We must get the most value out of every single component on that platform. And we've got a number of different programs in place with software developers in terms of how to best structure their code to take full advantage of the platform."
By the time Katmai rolls out, Intel expects to be supporting 800-megahertz, 64-megabit Direct RDRAM, which is scheduled to start shipping in volume in early 1999. Intel has worked closely with Rambus and DRAM vendors for more than a year to coordinate the transition to the Rambus memory. But the high price premium for the newer protocol-based DRAM recently forced Intel to adopt a new module format that could accomodate both Direct RDRAM and 100-MHz SDRAM. Even so, Intel said it still plans to introduce a chip set that will support Direct RDRAM in 1999.
techweb.com
So far, Intel is standing firm, saying its future chip sets will support only Rambus memory. However, companies such as Micron and Via Technologies are developing chip sets for SLDRAM. Via already has a chip set supporting DDR DRAM and plans to introduce a chip set for Direct RDRAM as well.
chipanalyst.com@2874942tjpyzk/mpr/merced/v11_14.html
At the Forum, Intel's Fred Pollack confirmed that Merced is on target for shipments in 1999, using a 0.18-micron process; we expect it to appear in systems around the middle of that year. |
