The Twin Stigmata of Rambus

Format for Printing

Format for printing

Request Reprints


By ptnewell
September 7, 2001

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

As fall arrived, the press joined Wall Street analysts in proclaiming that the Great Memory War was over. DDR has won, or at least will have won, in just a few months. RDRAM is expected to survive mainly as a minor niche market. The plummeting stock price should prove how the wealthy (and therefore surely the wise) are making their bets. Staying invested in Rambus is thus places obstinacy above prudence...
The above, of course, is an accurate summary of the situation of Rambus circa fall 1999. Press reports made it clear that the DDR wave was imminent (1). RDRAM shipments were no more than about 200,000/month (I will use 128-Mbit equivalents throughout), and would never amount to much more. And yet�

Come fall of 2000, RDRAM growth had confounded critics, skyrocketing by more than a factor of 10, up to about 4 million/month. DDR alas, had failed to meet the predictions of Semico (not to mention Tom's Hardware and Anand's Hardware), and mysteriously failed to mount a serious challenge. Still, the press, the Wall Street analysts, and the message board pundits propounding DDR were hardly chastised. In the fall of 2000, slightly less than one year ago, Spooner on ZDNET wrote an article proclaiming DDR the "comeback kid". He wrote "All the analysts agree that DDR will become the new standard in 2001." If so, all the analysts were wrong. With only a few months to go in the year, RDRAM is outselling DDR by as much as 10 to 1. This is based on Hyundai's recent boast to have 45% of the market with about 1.6 million (128 Mbit/month) DDR shipments, versus Samsung's claims to have about 50-60% of the RDRAM market with 15-20/million (128 Mbit equivalents) shipped. Only two major American OEMs ever adopted ANY DDR platform. These two are Hewlett-Packard and Compaq, who are offering models based on the AMD 760 chipset with a 6-layer Gigabyte motherboard. These models do not have a price advantage compared with Dell P4/RDRAM platforms. It has been almost 8 months since any DDR model has been announced by any American OEM. Sis, Ali, and Via have cheaper DDR chipsets, but, for whatever reason, no American OEM has touched any one of them.

There must be something about leaves falling that promotes prognostication madness. Another fall is upon us, and once again the Rambus sky is purportedly falling. DDR (as always) will enter the mainstream over the next few months. Apparently all those OEMs turning cold shoulders to DDR platforms over the last year or more have just been playing the coquette, while secretly longing to marry the RDRAM alternative. Or at any rate, so the usual pundits tell us. To be fair, the arguments are not precisely the same as, say, two years ago. Then the price/performance ratio of the only RDRAM platform (the i820) can safely be described as abysmal. (In fact all the DDR/P3 platforms have also failed to materially improve the P3's performance). Recently, the DDR enthusiasts cheered loudly when Via announced its P4X266 chipset would outperform Intel's i850 (P4/RDRAM), at a lower cost. When the P4X266 was "unveiled" (which is to say, various hardware sites were permitted just a few hours each with the system exactly as Via provided it), it turned out to under perform the i850 by as much as 10% (averaging 4% on overall measures), using the more expensive PC2400 DDR Via provided. DDR fans cheered again, unconsciously revealing how deeply cynical even they have become about Via's claims. Thus over the last two years the claims for DDR have been reduced to, "For not too much more, it does not run all that much slower", which hardly seems like the battle cry of a standard poised for victory (6).

Yet the Brookdale i845 remains one stigma marring the Rambus future. The same analysts who were wrong in the fall of 1999, and who were wrong again last fall, still see an imminent end to the RDRAM ramp up. "RDRAM is a sunset industry", Ashok Kumar (Jeffrey Piper) proclaims, marveling that a non-Intel RDRAM/P4 chipset is in the works (from ATI). Morgan Stanley's Mark Edelstone, less melodramatically and thus less quotably asserts that the Brookdale will halt the RDRAM upswing. Those who, in the fall of 2001, believe that the RDRAM rise is abating, and that DDR is surging, are not as wrong as they were in the fall of 1999 and again in the fall of 2000. To be sure, they are wrong, even astonishingly wrong. But it is almost mathematically impossible for them to be as wrong as they were last fall or the one before. Nearing 10% of the DRAM market by bits (and probably already 1/3 by dollar value) RDRAM cannot possibly increase by MORE than the factor of 10 it achieved two years ago, and yet again over the last year. One more such year would give it 100% of the market, and even I doubt the RDRAM transition will be quite that fast.

The other stigma haunting Rambus investors is, of course, the legal battles. For the last reported quarter, royalties were $20 million. The RDRAM market was put at $500 million. Based on the early quarters of 2000, when Rambus was reporting only RDRAM royalties, it is possible to infer that total percentage taken by Rambus is somewhere around 2.25% (probably divided as 1.5% on the RDRAM itself, and the rest coming from royalties on the controllers, RACs). Thus RDRAM-only associated royalties for January-March 2001 amounted to about $11 million or so. The point I am coming to here is that the SDRAM royalties (~$9 million) must have barely exceeded legal fees (about $8.5 million). With the collapse of SDRAM prices since March, it is almost certain that the legal fees now exceed SDRAM/DDR royalties. Thus Rambus net earnings are now LESS than if it had never claimed SDRAM, and thus had no court fights. Without claims to SDRAM and DDR, Rambus would be almost alone among technology companies in having rising earnings (indeed, given the delay in receiving royalty payments, rising earnings would have been assured through the January 2002 report, regardless of what happens in the DRAM market in the near future). Instead, the cost of the court fights coupled with the collapse in memory prices are driving Rambus close to the break-even point even as their core product surges.

Which still leaves Rambus better off than nearly all other technology companies. Many, including Micron, Hynix, Infineon (to name a favorite few) seem to post larger losses every quarter. Others such as Cisco, Gateway, and Compaq claim small pro forma profits, although they have large losses under the Generally Accepted Accounting Principles (GAAP) of the accounting standards board FASB. (Some companies such as Micron seem to have large "one time" charges EVERY quarter, keeping their pro forma losses merely ugly instead of horrendous).

The fortunes of Rambus investors have sharply diverged from the fortunes of the company they invested in. Suppose you were transported back to early 2000, knowing all about the earnings trajectory of every technology stock and nothing about stock price movements. In the spring of 2000 one could have bought few technology companies with earnings that have held up as well as Rambus. Their FASB profits over the trailing 12 months are substantially larger than their pro forma profits, amounting to about 0.72/share (or $2.88/share pre-split, for old-timers), dwarfing analysts' predictions from early last year. How many tech companies can say as much? One could probably not have found any other technology companies with such rapid market share growth (more than a factor of 10) ahead over the next year. And the "wild" predictions of Rambus supporters that companies like Micron and Hynix that did not support RDRAM would lose money as a result have amply come to pass. Toshiba, and, especially, Samsung, have managed to remain profitable far longer than their competitors, primarily because of the booming RDRAM market (2). (RDRAM prices seem to have recently finally dropped below the point where they are profitable for Toshiba, however, at least on their older 0.20 micron fabrication lines. At any rate, Toshiba is now losing money also). Yet you would have been far better off investing in companies such as Micron, even knowing they were poised to lose money, than in Rambus, with its sky-rocketing earnings ahead! The twin stigmata of the legal battles and the Brookdale have depressed investors, and continue to loom large on the horizon. "There are no catalysts for Rambus share prices," Edelstone advises his clients.

I will spend little time on the legal issues. Any intelligent and fair minded observer who still thought Robert Payne judicial after he declared that "write request" does not mean a request to write, and that "bus" does not mean "bus" must surely have changed their mind when, halfway through the Richmond trial, he freighted these terms with ADDITIONAL requirements, imbuing them with meanings not present in his Markman ruling (3). Robert Payne has of course, capped his intellectually indefensible string of rulings with the proclamation that any disagreement with his own thoroughly frivolous views is "frivolous", and hence ordered Rambus to pay Infineon $7.5 million for filing a "frivolous" lawsuit. Even putting aside the two Italian court-appointed experts, or the independent views cited in reference (3), patent lawyers at more than a half-dozen companies have concluded that Rambus's claims are valid, and these companies have signed with Rambus. Are all these scores of lawyers then spending their companies' money "frivolously"?

Surely reductio ad absurdum has been reached. I still hope and believe the theft of Rambus's intellectual property will not succeed. I am aware of those putative attorney's insisting that the Appeals court will defer to Payne. The legal requirements for a de nova Markman examination (that is, the legal requirement that the Appeals court NOT defer to Payne) is deprecated by such legal pundits. Although I am somewhat aghast at such expressions of casual contempt for the law, I am hardly ready to concede such advanced stage of corruption exists in our courts. The Appeals court will likely do what the law requires, and reconsider the Markman rulings without deference to Payne's mistakes.

Let us return instead to the first stigma on Rambus shares, the Brookdale. Dropping the metaphor, the issue is whether Rambus is inordinately dependent on Intel's goodwill, and whether that sanction may be granted or withdrawn capriciously. It is not uncommon to hear it said that "Intel made Rambus." Balderdash. Only a relatively small portion of the $500 million in RDRAM sales that occurred January-March of 2001 were associated with the measly 800,000 Pentium 4s that sold in that quarter. Rambus's biggest patron (to date) is not Intel but Sony. The Sony Playstation 2 has used more RDRAM than the Pentium 4 (at least in quarters reported thus far). Sony HDTVs use RDRAM, as do certain other Sony devices. Only 70% of the DRAM market is associated with computers; while 30% is associated with consumer goods. (Intel CPUs couple to only 80% of that, and Intel provides chipsets for perhaps 80% of THAT. Thus Intel accounts for perhaps 0.7 x 0.8 x 0.8 =45% of the DRAM market, and probably less). It is in the consumer goods sector 30% that RDRAM has shown significant strength, which, fortunately for shareholders, is the portion of the market that is actually growing. The use of DRAM in such items as set top boxes, HDTVs, and the like is a relatively new but fast growing phenomenon. It is not improbable that in 5 years or a little more, the percentages will be reversed, and 70% of the DRAM market will be in consumer goods, and only 30% associated with computers. (This is the appropriate place to confess that the 5 or 10 to 1 ratio of RDRAM to DDR sales cited above overstates their respective roles in the PC market. RDRAM is probably ahead of DDR by less than 2-1 in the PC market. Still, it clearly is ahead even there (4), confounding industry and pundit predictions). It is well worth pointing out that RDRAM is breaking into areas traditionally reserved for expensive SRAM, especially in the communication industry. Brocade Silkworms and Vitesse IQPrisms are just a few instances. Personally I doubt the volume of these communication market uses will ever rival the consumer goods sectors, but at least for some investors the very word "communications" is magic, and often inspires irrational valuations.

My thesis then is not only that Rambus is valued as though it had no claim to SDRAM or DDR, or indeed valued LOWER for having such claims because its earnings are currently lower than they would be with an RDRAM-only approach, but it is valued LOWER than if it had no relationship with Intel. If one considers companies such as PMC-Sierra, with relatively small revenues, worse growth rate histories, and less breadth and depth to the usage of their products, it is hard to see how a Rambus devoted purely to the communications and consumer goods usage of RDRAM would not be valued higher than it presently is. I understand that investors fear adverse decisions by the courts and by Intel, but surely it is a peculiar situation where Rambus is valued lower than if these decisions were known, and had gone as badly as is possible.

However Intel does not have a realistic option of avoiding a heavy reliance on RDRAM, at least for the next few years. Intel has its own skeptics. Let us put aside the issue of their rivalry with AMD. Since both Intel and AMD buy their chip making manufacturing prowess from Applied Materials, I expect Intel will be rational enough to use their vast financial resources to outspend AMD whenever necessary and maintain a production edge. I presume even Craig Barrett understands enough to price cut to the point where AMD bleeds until it no longer poses a serious threat. To sum up Intel's real and perennial problem, I defer again to the always quotable and seldom correct Ashok Kumar: "No one needs a 2 GHz processor." If no one needs faster processors, then no one need pay a premium for the higher grades. The result would be that CPUs become a commodity. Intel would be in essentially the same boat as Hynix or Infineon. As Warren Buffett has stated, the only thing worse than being in the commodity business is being in a commodity business with high capital costs.

Intel thus is always planning and pushing new uses for faster and more capable PCs. Nearly all the possibilities they have identified are high bandwidth. From teleconferencing to videophone calls over the Internet, from ever more realistic games to video editing, from streaming entertainment to extensive multi-tasking, the mass market uses for premium priced PCs all seem to be high-bandwidth uses. RDRAM quite simply has 3-4 times the bandwidth of its DDR rival. (1066 MHz RDRAM which is here now, but currently used only by Vitesse, versus 266 MHz DDR, or perhaps 333 MHz DDR if one wishes to stretch a point to unproven products). Of course PC133 could be used to get twice the bandwidth of 266 MHz DDR, by using four times the data path. In fact using a 256 bit data path, PC066 can be said to have the same bandwidth as 266 MHz DDR (on a PC standard 64 bit data bus). Indeed, if one is willing to use enough motherboard traces, any amount of bandwidth can be obtained with any memory, no matter how slow.

Of course it is nonsense to say that PC066 is therefore the technological equal of 266 MHz DDR, just as it is nonsense to argue that the NForce, with its 128 bit path, is technologically comparable to a 32 bit PC1066 RDRAM design (both with 4.2 GB/s bandwidth). Actually the comparison with PC066 is too kind to the NForce. Because RDRAM is serial, only one copy of the 32 bit data path is needed. Because DDR is not, a duplicate copy of the entire data path is needed for each DIMM. A two-DIMM implementation of the NForce thus requires 256 bits of data traces on a PC-board (each of which must be precisely the same length, because of the speed of light delays associated with high speeds). I doubt whether the NForce design will prove economically practical, as such schemes have traditionally been reserved for the much more expensive server market. Regardless, it should be readily apparent that the Rambus high-speed serial approach leaves FAR more headroom for increased bandwidth, essentially on demand. When Intel builds CPUs fast enough, they can easily double the bandwidth by going to a 64 bit path, and STILL only use a fraction of the PC-board real estate the NForce requires. DDR poises roughly the same threat to RDRAM that the Clippers do to the Lakers (and are posting pretty much similar results in the marketplace). The elegant and cheap simplicity of high speed serial RDRAM is what has made it so successful in the consumer electronics market. As Intel introduces ever more complex CPUs, hogging ever more PC-board real estate but simultaneously demanding ever more bandwidth, they must inevitably turn to the same solution as Sony or Panasonic: RDRAM.

Max Planck, the physicist who first discovered quantum physics, said that a scientific revolution never converts its critics. Instead the old guard opposition eventually dies off. It is unfortunate that the companies who bet wrongly on DDR and against RDRAM, and who thereby suffered huge losses over the last year, have not gone bankrupt. The normal consequence of such a disastrous error have been averted or at least postponed by endless rounds of Wall Street-engineered new infusions of cash from the capital markets. For example, Hynix has once again just last week arranged (through Salomon Smith Barney) the issuance of GDR (basically stock shares, traded globally) for a new infusion of cash. As long as Micron, Hynix and Infineon, the same three companies who fought to prevent SDRAM from ever replacing EDO (5), can continue to subsidize commercial losses with new investor money, the DRAM war will continue. Most such cash infusions however result in share ownership dilution. The peculiar subsidization of unprofitable DRAM by investor losses cannot continue indefinitely. Neither can the divergence between the impressive successes of Rambus the company and the mounting losses of its shareholders.

(1) Micron Volume DDR production just two months away (in 10/99)

(2) Samsung estimates RDRAM at 12% of the DRAM market, and only profitable DRAM

(3) As well expressed in this independent legal analysis

(4) DDR sales collapse, RDRAM gaining acceptance and momentum (5/15/2001)

     DDR module orders vanish as of start of Q2 (01) (But rebound predicted in '02)

(5) Micron, Hyundai, and Infineon pushing "Burst EDO" instead of SDRAM in 1995

(6) Yet Via is still able to snow the credulous: It has reissued its DDR/Athlon chipset, with the primary change simply that it was provided to reviewers loaded with PC2400. Anand and others swallowed the bait, not realizing that when coupled with cheap DDR the new platform will quite likely have the same cheap results as before.


TMF Money Advisor
Got money questions? Your answers are just a phone call away! TMF Money Advisor puts you in touch with an objective Financial Planner whenever you need it.