A fantastic post on Rambus by ptnewell on Yahoo!
Throughout the great memory war it has paid to follow concrete acts by the participants, rather than lend much credence to their rhetorical flourishes. For example Hyundai announced a conversion to the RDRAM (Rambus/Intel) side in February 2000, saying that they "wanted to be the world leader in RDRAM production." In reality they produced almost no RDRAM, and stayed firmly in the Team-DDR (Micron/AMD/Via) camp. Likewise AMD announced plans for an eventual RDRAM chipset, and even ran a series of advertisements for engineers with Rambus experience. Whether this was a deceptive feint (the ads were so numerous they could not have been overlooked by any party serious about the issue) or a case of genuine self-doubt and ass-covering is unclear. AMD’s concrete actions have been clearly supportive of DDR, not RDRAM.
In this light, it is worth examining the behavior of both Intel and Rambus since the putative weakening of their ties in early June of 2000. At that time, Intel announced intentions to support a PC133 chipset for the Pentium 4, and to seriously investigate DDR "at certain price entry levels". A crescendo of articles had appeared throughout May suggesting a coordinated anti-trust action against Rambus was likely. In defending against anti-trust claims, a certain degree of separation between Intel and Rambus is desirable. A monopoly is legal; however using monopoly power in one field to establish a monopoly in a second is not (e.g., using a monopoly in operating systems to force a monopoly in web browsers). In fact Micron’s eventual filing against Rambus complained that (paragraph 39) "As a result of Intel’s announcements, DRAM manufacturers were effectively forced to license the RDRAM technology from Rambus. Micron licensed this technology from Rambus in 1997." Rambus and Intel would find their financial interests best served if RDRAM is seen to win in the marketplace, rather than by fiat.
Consider the concrete actions by Intel in the five months since the supposed breakup with Rambus:
(1) Intel has not announced any DDR standard. Prior to using PC066, PC100, and PC133, Intel announced standards to resolve incompatibilities between vendors (the JEDEC "standard" in each case proving insufficiently standard). Such discrepancies between various rival manufacturer’s versions of DDR have been widely reported. If Intel were seriously trying to bring a DDR chipset to market, they would need to announce a DDR standard probably six months prior to entering the OEM verification process (giving Dell, Gateway, etc. sample versions). Conversely, if Intel hoped DDR would fail, the last thing it would do is announce a standard.
(2) Intel has announced a $70 rebate on the use of RDRAM in the Pentium 4. Since price has always been the biggest complaint, nothing could more thoroughly undermine establishing a DDR volume market. There is absolutely nothing Intel could do to be more strongly supportive of RDRAM.
(3) Intel has postponed the use of DDR in servers in favor of PC133, and apparently RDRAM. Although DDR is still supposed to appear in servers eventually, the date has been pushed back indefinitely. Once again, this undermines the likelihood of a volume DDR market appearing.
(4) Oweowepd translated an article from the Chinese version of Yahoo (posted Oct. 27) in which Intel’s CEO Barrett stated that certain third party licenses had already been granted for RDRAM chipsets for the Pentium 4. However Via, which wants to build a DDR chipset, has not been licensed. In fact Intel has apparently been stalling Via, according to other reports. Again, anti-trust concerns (Via is a chipset competitor) may force Intel to license Via eventually, but this stalling strongly suggests Intel’s real hopes.
(5) Intel’s Paul Otelli, in a conference call Wednesday, November 1, seemed to push the time of any Intel produced DDR chipset into the very distant future. He stated that the Brookdale (PC133 Pentium 4 chipset) would not appear until the second half of next year, and that Intel would consider a DDR chipset only following that – apparently into 2002. Even then the precise wording was significant: "after DDR becomes mainstream". Indeed. And AMD will produce a RDRAM chipset after it becomes mainstream. But Intel appears to be doing everything practical to prevent DDR from ever becoming mainstream.
(6) Even the introduction of a PC133 Pentium 4 chipset, AFTER 8-10 months of RDRAM exclusivity, will make it harder to introduce DDR/P4 chipset, since it would have to squeeze in between RDRAM on the performance end and PC133 on the inexpensive end. RDRAM cannot claim the low-end of the market by the end of 2001, but PC133 might help Intel fend off DDR from doing so.
The financial self-interest of Intel would clearly be best served if the DDR effort fizzled and Via and AMD were forced to play catch-up with Intel. Conversely, Intel’s "you go first" approach to DDR is incomprehensible unless they believe AMD and Via are driving into a dead-end. Intel would not risk playing catch-up unless it was genuinely convinced that the DDR noise and stability problems it has so often pointed out before (and which are prominent on Intel’s current web site) were real. Certainly the failed launches of DDR chipsets by Acer, Via, and most recently by AMD gives reason to believe Intel is correct.
Intel and Rambus may be playing good cop/bad cop, working toward the same goal in fact if not in word. There is reason to believe that Intel is ambushing AMD and Via, with the Pentium 4 likely to have higher performance, a faster ramp up, and more aggressive pricing than first suggested. Rambus is acting as the "bad cop", suing three key members of Team DDR (Micron, Infineon, and Hyundai), while threatening others (AMD). Indeed, the first chipset targets chosen by Rambus (AMD and Transmeta) hardly suggest a genuine break with Intel. TXN and MOT would be fatter targets, since they use huge numbers of memory controllers. DSPs may be the most lucrative Rambus target of all, since even though they use small amounts of memory, each DSP needs a logic controller (with 3-5% royalties instead of 1-2%). By contrast, PCs use large amounts of memory, but ship fewer units, hence use fewer logic controllers. Rambus, like Intel, is acting as if they still believe the war against DDR can be won.
Rambus has been ruthless in its legal tactics, to an extent no manufacturing company could ever risk. Micron, which thought to intimidate Rambus by coordinating three simultaneous lawsuits (it no doubt wanted more) against Rambus must have been astonished at the response. Rambus has initiated law suits in France, Germany, England, and most recently, Italy. These legal actions have reportedly involved actual seizures. Unlike any other IP battle, Rambus has gone after secondary players; e.g. filing a lawsuit against Sega for use of Hitachi memory in its Dreamcast. (Hitachi since settled with Rambus). Likewise, Hyundai has complained about difficulties because major customers have demanded indemnification against Rambus lawsuits.
The good cop is required to disavow the bad cop’s actions. Barrett has done so, with his comments on Rambus as a "toll taker" (apparently he was shocked to learn that a pure IP company was collecting royalty). Yet Intel’s concrete actions, as outlined above continue to be strongly pro-Rambus. Likewise, Rambus, in targeting AMD first among all the scores (or even >100 companies) that use memory hardly evinces a genuine break with Intel. With the aggressive legal tactics Rambus employs, it could be very difficult for any OEM to actually ship AMD DDR boxes. Intel is probably not as surprised or upset about this as Barrett’s remarks might suggest.
Instead of scrutinizing the nuances of each remark by each participant, it might be most instructive to pay attention to the financial best interests and concrete actions of the players. Intel would clearly be best served if DDR flopped, and AMD and Via were forced to adapt to a new set of rules where Intel had the lead. Indeed, that is a standard gorilla tactic when rivals (in CPUs and chipset design) narrow the gap. In that light, Intel’s concrete actions (offering a rebate on RDRAM use, refusing to specify a DDR standard, refusing to introduce even a DDR server chipset until after DDR "becomes mainstream") speak louder than words. Likewise, the apparent separation between Intel and Rambus may be useful to both. Rambus avoids anti-trust considerations while gaining the freedom to use legal tactics Intel cannot endorse – but may be quietly enjoying. |
