Hi all; Reply to the "There can only be one" article...
It's been repeatedly stated that this article has not been answered on SI. Well, maybe I did ignore it. But, goaded to action, and a little bored, here's an analysis of this article:
Here's a link to the article I'm referencing:
community.metamarkets.com
First, I agree with the premise that "there can only be one", which is the subject of the first eight paragraphs. This has always been true in the memory industry, and most of the memory makers, and all of the memory users, agree with it. But the author, due to his lack of understanding of the memory industry, needlessly clouds his argument with comparisons to other industries. This is confusing to the reader because there are plenty of comparisons, (that he didn't make), to industries where it wasn't the case that "there can only be one". The whole thing reeks of how the reasoning that once allowed the Bible, for instance, to simultaneously damn and praise slavery. So I don't have a problem with the conclusion of those paragraphs, just with the sloppy reasoning.
The author's conclusion from the first eight paragraphs is that "the standard that creates the SUPERIOR PRODUCT will become the next generation standard." [Paragraph 9] I agree with this, but from an engineering point of view, the "superior product" is the one that is cheapest that achieves the desired performance. I should explain this, as this statement is not understood by most of you losers.
If the desired performance is a "family sedan", then it doesn't make sense to put a 6.0 liter engine under the hood, if (as is the case) a smaller engine is cheaper and achieves the "desired" performance. That is not to say that there is no market for cars with 6.0 liter engines. There is a market for such cars, (AFAIK), but it is not the "family sedan" market. As engineers, we do not usually get to define the performance. Instead, we are told what performance we are to achieve, and we then go about finding the "cheapest" way of achieving that performance. And even the word "cheap" isn't obvious. The total cost of an engineering decision is incredibly complex. Cost involves far, far more than just the price of the parts. "Cost" even involves such things as how many possible second sources are available for the part, or the parts it connect to.
Probably the item that mom and pop are least aware of is that high performance memories can be built from pretty much any type of memory, providing one is willing to ignore the "cost" issue. Plain old SDRAM or even EDO DRAM can (and is) used to make very high bandwidth memories simply by making the memory very wide. This means that it is not possible to conclude that a given memory type is going to be required for future processors simply because it provides higher bandwidth per chip (or per pin). On the other hand, the latency of a memory system cannot be improved by adding chips, it can only be degraded. This is the area where Rambus is most deficient, it provides considerably worse latencies than DDR.
Re: (10) "In this regard, [i.e. superior product] , RDRAM is clearly the hands down winner." The author provides no reasoning for this, and since he's a lawyer with no design experience, this is not surprising. Since every chipset maker is now working on DDR chipsets, while only one is working on RDRAM one would assume the reverse. In fact, Nintendo used RDRAM in their earlier game console, but dropped it in their next version. Similarly, PixelFusion (now ClearpSpeed) used RDRAM in the Fuzion 150, but dropped it from their newer line. Also, the Sun MAGIC graphics chip appears to have "disappeared", while Rambus' only pure graphics solution that ever shipped has now been entirely replaced and supplanted by DDR based graphics boards from Nvidia, ATI &c.
Let's go over this again. On the one side, we have a lawyer who's likely never looked at a Gerber file saying that RDRAM is clearly superior, but not telling us how. On the other side, we have dozens of companies that have designed DDR into products, several of which already had experience with RDRAM. Who you gonna believe? Mr Lawyer or Mr. Market? Just where is that RDRAM chipset for AMD anyway? And why isn't ALi working on an RDRAM chipset for either Intel or AMD? Should these engineers go start arguing before judges and this lawyer go get a job designing circuits? Clearly, either this guy is a fool, or most of the engineering community doesn't know how to do their jobs.
Re: (10) "In this regard, RDRAM is clearly the hands down winner. Not only is RDRAM a technologically superior solution, and starts out with this advantage (as Windows and VHS did not), but it also has the largest value chain working to assure that RDRAM systems will be the best PRODUCT for the desktop."
I love this phrase "value chain". This is probably code words for "I'm a fool, and I'm going to make a simplistic analysis of a complicated system. I really should leave this sort of reasoning alone, but lets look at it on its face. Intel, AMD, SiS, VIA, ALi, IBM, ServerWorks, Nanya and God knows who else are all working on chipsets for DDR. This includes every chipset manufacturer, no exceptions. In addition, every memory maker, no exceptions, is working on DDR SDRAM. The "value chain" working on DDR couldn't be larger, unless perhaps, the space aliens land in Silicon Valley and start promoting something else. Everybody is supporting DDR, how could it be better than that? With RDRAM, on the contrary, only one of the chipset makers is supporting the stuff, and they're also supporting DDR, while only a few of the memory makers are making RDRAM, while they are all producing DDR.
Re: (11-16) "To fully illustrate what I mean by better product and the power of a value chain, there is no better example of this than the just released Via KT266 chipset. This is the state of the art in DDR PC systems. It is the system that DDR supporters have been pointing at for months as the first truly great DDR system for the PC. Below are the relevant excerpts from a review of the system by Anandtech
...
Compare this to any chipset or system that has been released with RDRAM; all such systems having been put on the market through Rambus' large value chain of Intel and Dell and Samsung and Sony and the enormous value chain these companies control to drive their markets. What you get is not some not-yet-ready for the hobbyist product like the above Via offering, but real products on a large scale."
Since the AMD 761 is the chipset that is used by HP and Compaq in their DDR computers, the author naturally avoids talking about it, and instead looks at a "just released" chipset from VIA, calling it "state of the art". Yeah, that's fair, LOL!!!
Paragraphs 11 through 16 compare the KT266 with the RDRAM systems released by Intel. This is an apples to oranges comparison. The KT266 was sold (at the time of the excerpted Anandtech article) almost entirely to do it yourselfers, while the RDRAM systems were largely sold to big business. Big business and "bleeding edge" DIYs have very little in common, and this should not be a surprise to any of us. A more reasonable comparison would be between the Athlon DDR systems from HP and Compaq, and the RDRAM systems from the same companies, but even that wouldn't be fair because the Athlon DDR systems are not being sold to business. Intel pretty much has a lock on the business side of things, and the first decent DDR machines for business are likely to be from Intel. There are no DDR machines being sold to business at this time, so making comparisons between current DDR systems and current RDRAM systems is kind of stupid.
As an investor, your duty is to analyze what the future will bring, not make straw-man arguments about what the present contains. That Intel is going to have DDR machines is obvious. A more rational pro Rambus argument would be to hold that when Intel does get DDR, it's not going to be sold into business, and therefore that part of the market is reserved for Rambus. But making that conclusion based on a comparison between Intel business machines and AMD DIY machines is silly.
Intel, who originally stated that PC133 wouldn't be a good idea, is now selling it into the business sector. I believe that it should be obvious to all that they will do the same with DDR. Certainly Samsung believes that unbuffered DDR is suitable for business computers, unless you believe that DIY types are buying Alphas:
samsungelectronics.com
Re: (17) "Samsung produced 5 million Rambus DRAM chips in January but increased production to 10 million last month. It will again raise output to 15 million next month and 20 million in the fall, which will result in a total production of 150 million to 160 million units this year." There are a whole bunch of problems with this analysis. (a) It fails to mention what total world or even Samsung production will be. Instead, it looks at just one memory makers RDRAM production. World-wide memory production is b i g, and Rambus memory is just a tiny fraction. (b) It fails to analyze what the effect of the i845 and the numerous other alternative chipsets for the P4 will be. Industry analysts believe that by the end of the year, most P4s will go out with either SDRAM or DDR. (c) It fails to note that Samsung has made many previous predictions that it failed to achieve, and in addition, that Samsung is fully supporting DDR. For instance, Samsung was predicting that RDRAM would be 50% of the 2001 market as recently as April 1999:
eetimes.com
Also see #reply-15947382 #reply-15962808 #reply-15905562 #reply-15791326
Re: (17) "DDR simply does not have the power of Rambus behind it insuring the standardization and quality of RDRAM as it comes out of the fab. Neither does DDR have the power of Intel, Sony, and DELL, and their suppliers, making sure that systems incorporating RDRAM are professional, smooth performing and stable mature products." The industry has been shipping SDRAM for years, and it didn't have the "power of Rambus" behind it, so this argument seems a bit hopeful. As far as Intel, Sony and Dell, go, the largest of these, Intel, is supporting DDR, so DDR does have the power of Intel behind it. Sony's use of RDRAM is small, and Dell is just a box maker. When Intel provides Dell with a DDR solution, Dell will use it, simple as that. In fact, Dell already sells DDR based graphics boards. This whole paragraph is continuing the fallacy of comparing a hobbiest type DDR system with a business type RDRAM system, and concluding that this is the situation that will always obtain. Intel already cried "uncle", and pretty much every motherboard maker demonstrated their DDR solution for the P4 at Comdex:
i845 (Brookdale) DDR
Nearly every motherboard manufacturer we talked to had two i845 motherboards at the show: a PC133 i845 on display, and a DDR i845 underneath the table that Intel wouldn’t allow them to show.
anandtech.com
Paragraph 18 quotes an October 1999 article from Micron in order to suggest that in 2002, DDR266 (i.e. PC2100) won't be stable. I guess, for a lawyer, this sort of makes sense, but in technology, precedent doesn't make law, LOL. These were early production from Micron, of course they had bugs to work out. Here's a link to the quoted article:
eetimes.com
Re: (19) " Things are so bad that Micron may actually have to produce its own chips. Can you imagine any portion of the Rambus value chain scrapping and scraping as Micron is in trying to get its DDR systems out the door?" Since every chipset maker decided to support DDR, Micron ran away from the business and is sticking to memory. Everyone is supporting DDR, it's too competitive for Micron to enter.
Re: (19) "It is also interesting how the stability problem, 18 months later, is still not really under control." This is either a bald faced lie, or a mistake due to the author conflating different issues. This is hardly surprising, to mom and pop, all problems appear to be the same one. Every design has lots of problems that have to be ironed out, and Rambus systems are not exceptions. In fact, isn't it RDRAM that had to give up on the third RIMM slot?
Re: (19) "Quite simply, there is just not coordinated industry support to help build DDR systems into great products. The products currently being built demonstrate this unequivocally. It is doubtful if such a value chain will exist anytime soon either." It's kind of hard to believe that all those companies (including Intel) aren't going to get DDR to work. I wonder why they're working on it. Must be because Shaw knows engineering better than they do, LOL!!!
Re: (19) "RDRAM is the more profitable product to produce at this time. Several DRAMs have already stated their goals are to move towards more higher value added products like RDRAM and away from SDRAM products. Where excess profits exist, vendors will move into rapidly to take advantage of the opportunity. Even Infineon, with jury selection starting today, has recently become RDRAM certified. The more RDRAM being produced the more the entire industry invests in producing RDRAM products."
The argument suggests that the higher profitability of RDRAM will cause larger production of it, therefore helping RDRAM gain a larger portion of the marketplace. But the larger production of RDRAM will cause lower prices, and therefore lower profits. That will cause smaller production of RDRAM. Confused? You should be, (the argument sucks), but the memory industry isn't, and neither are their customers, LOL!!! Predicting what the next mainstream memory is going to be is too complicated for mom and pop, but you can figure it out (though a bit late) if you simply watch the actions (not the press releases) of the participants:
All the memory makers are supporting DDR, while several of them have no RDRAM at all. Advantage DDR. All the chipset makers are supporting DDR, while only one is supporting RDRAM. Advantage DDR. It's pretty clear that industry has concluded that DDR is going to be around, most of the companies involved with memory have ignored RDRAM completely, and, well, since "there can only be one", I guess that means that the industry has concluded that RDRAM is going to be dead, dead, dead. Pretty simple.
Re: (19) "For any decent value-chain to form around DDR products AMD, VIA, Micron and Hyundai will have to move at least as many mountains as Intel has managed to do over the past two years ..."
Since Intel is supporting DDR, I guess AMD, VIA, Micron and Hyundai don't have to do so, LOL, but lets take a look at this from the point of view of an industry designer. (a) He knows that there can only be one. (b) He knows that RDRAM is only being used by Intel's chipsets and Sony's PS/2, and that if they drop RDRAM, RDRAM would become history. (c) He knows that AMD, VIA, Nanya, Intel, SiS, ALi, ServerWorks, IBM, ATi, etc, etc are using DDR, and therefore DDR has much more immunity to becoming "history". (d) He knows that if he uses a memory type that becomes "history", his product will be more expensive than otherwise, and he might even become "history". (e) He knows that even though Intel has strongly (at least officially) supported RDRAM for years, there have been no new design wins since late 1999, and that RDRAM use is therefore largely going to be limited to Intel and Sony designs.
Now, given the above realizations, which are now all obvious to all design engineers, why in the hell would anyone decide to use RDRAM? The only possible conclusion is that the guy hates his company, LOL.
Re: (19) "... and they will have to do so in the face of the rapidly growing demand of the more profitable to produce RDRAM." The profitability of RDRAM is only of significance to the memory makers. The users (who will determine what memory is used) don't give a fig whether the memory makers are profitable or not, they only care about whether they're going to get their product, and how much that product costs.
The remaining two paragraphs reiterate the points addressed in the above. Basically, Shaw's analysis is the rosy analysis of the genius who finds a three card monte in progress and concludes that since he's a genius, he's smarter than the guy who makes a living with the thing.
-- Carl
P.S. 3-card monte: pagat.com |
