REALITIES ABOUT RAMBUS
Story Filed: Tuesday, September 11, 2001 11:31 AM EST
Sep 11, 2001 (JAGfn.com via COMTEX) -- From Hager Technology Research www.Fredhager.com
Realities About Rambus By Bill Teel
There is a huge misperception regarding Intel's relationship with Rambus that is spreading like wildfire again, and that is Intel is dropping RDRAM support and replacing it with DDR. Intel is not, nor do they have any plans to drop Rambus RDRAM support now and in the future. If anything, as we may see next week at the Intel Developers Forum, Intel is continuing to expand their support for RDRAM memory and Rambus technology in general. Please read the following link Intel recently sent to their business partners, and general public to help clear up the untruths and misconceptions regarding Rambus and Intel.
intel.com
Those of you who've seen the movie Star Wars are familiar with the Millennium Falcon, the spaceship that Hans Solo and his sidekick Wookie, Chewbacca piloted around the solar system in their various adventures.
The Millennium Falcon wasn't a particularly sleek spacecraft, and it was often prone to miss-starts, breakdowns, random attacks from various rogue factions, and the occasional asteroid that whacked the craft upside the cockpit. But, inevitably, when the machine needed to perform and outrun its adversaries, it did so swiftly and with authority. The ship was faster than anything in the universe, and no one could take it down, because they couldn't keep up.
You're probably scratching your head and thinking, "Rambus has hit a new low that is below the IPO price, and you're telling us stories about the friggin' Millennium Falcon!? Have you lost your minds?"
The answer is no, and while we're not an investment advisory service here at Hager Technology Research, there's no law that says we can't offer a little sage advice. And that is, relax. Turn of the computer, stop watching the ticker, and forget about it for a while. Rambus, like the Millennium Falcon is getting hit from all sides with all kinds of things, but it too will emerge as the faster, stronger, more stable vehicle that will outrun, and outperform all others. And it too may just help save the day at the end for computer users.
We believe the reason Rambus' price has been sliding is because of the misperception by many that Intel is dropping Rambus. Wherever we turn, and many of the people we speak to, continue to believe, despite what Intel has said time and again, that RDRAM will be dropped by Intel. Even the more sophisticated analysts who we thought knew better don't know any better.
We believe Rambus will ultimately be vindicated of this false rumor, and in the long run they will prevail as the dominant main memory for PCs. And, we anticipate the company's business will continue to grow beyond memory, and Rambus, as an investment will offer substantial returns for investors for the next several years. That's our opinion and belief, and why we continue to hold Rambus' stock, and have absolutely no intention of selling.
Just as possession is nine tenths of the law with regards to weapons and narcotics, perception seems to be nine tenths of the law to judge a company's health. And right now, the perception out there is that Intel is going to introduce DDR as a replacement for RDRAM. And RDRAM will become an ever-shrinking aspect of Intel's roadmap. Most of the trade press' content, which one editor told us last week is definitely influenced by their advertisers (go figure), would like the world to believe that Intel is not perfect and has made a big mistake with Rambus. Guess who advertises heavily in many of the trade publications? Take a look sometime.
Then there are the analysts, who mostly relay the information they hear in the trade press to their clients and the mainstream press. And while they may be serving their clients needs now, they seem too entirely focused on the short-term to think about what happens after 2GHz. CNBC, and many investors don't know the difference between a DDR, RDRAM and the WWF, and if the word out there (from analysts ala the trade press) is DDR is going into the P4, it must mean RDRAM is going out.
Since we seem to have assumed the role of last remaining arbiters of RDRAM, we might as well play the role of modern-day memory Cassandra as well. Our vocal chords are a little stressed from explaining granularity to people, but since the trade press repeats some of the same stuff over and over again, we might as well reiterate our case once more.
First, as Intel has stated numerous times, they will be offering a selection of memory choices for the Pentium 4 for the user to decide which is best for their purposes. RDRAM will be available for those who appreciate performance, DDR for those who want the slight performance improvement over SDRAM, and SDRAM for the low end of the spectrum.
A couple of notes on these points: First, Intel's Pentium 4 with SDRAM will have lower latency than their DDR version. (Lower is better.) Latency is the time it takes to start sending information from the chip to the processor. Critics have panned RDRAM for it's high latency, because RDRAM is geared for larger amounts of information to be delivered, e.g. large graphics files, etc. Currently, the RDRAM P4 outperforms the SDRAM P4 in latency, due to the efficiency of the architecture of the entire chipset. The two are still pretty close (regarding latency), but RDRAM clearly outperforms the P4 SDRAM in all other areas. Therefore, we predict the DDR Pentium 4 system will actually perform worse than the Pentium 4 system with SDRAM on some benchmarks because of DDR's higher latency. And these are the benchmarks so many publications have said are really what counts in the office environment, and the reason for saving a few bucks on an SDRAM P4 over an RDRAM P4. It begs the question of what's the point of DDR? Whose tastes will it satisfy, really?
Also, the margin of difference in price between the three versions is continually narrowing, and with the most recent drops in processors prices (upwards of 50hitting this coming weekend, a 2.0 GHz P4 system with RDRAM introduced shortly will likely offer incredible performance for the price.
Second, processors are getting faster. Some people say we don't need more speed, but they have been saying that for years. As processors get faster, higher-speed memory is needed to continue to further the relationship and enable better software to be used, as well as improved performance with Internet interaction. Rambus' RDRAM memory is the only memory in existence that has the headroom to enable Intel to continue to increase their processor speeds, and keep systems performing better as a whole.
Third, and here's that granularity word again, as memory requires more bandwidth, the cost-effectiveness of RDRAM chips become widely apparent due to the fact that at higher chip densities, systems can utilize fewer RDRAM chips than they can DDR chips to achieve the desired performance. RDRAM chips may cost more now, due to the learning curve, transition of equipment, and volume of shipments, but in the future, the value will certainly come from RDRAM chips, because fewer RDRAM chips will have to be purchased. And in the short-term, the new 4-bank RDRAM memory chips, due out in Q1 2002 will narrow the cost of manufacturing difference between RDRAM and DDR to practically nothing.
A lot of people come to us and tell us that not all good technologies make it. And we agree. There are plenty of technologies out there that far outperform their competitors, but because of price, politics, or whatever reasons, they just don't make it. We do not believe this is the case with Rambus. Not only is it a better technology, it is a more logical, cost-effective technology for the long-term. If it were simply only providing a performance advantage that could be achieved with other forms of memory, then we think Intel would have dropped Rambus a long time ago. They haven't, and they won't, because they need Rambus' RDRAM as much as Rambus needs them. It is not as simple as throwing up their hands and saying, "Oh well, this stuff's good, but I guess nobody wants it, so let's move on." The situation currently with Intel and their choice of memory is reminiscent of when the company chose to stick with CISC scheme in their chips, as opposed to going with the newer and faster RISC scheme. Then, Intel chose to stick to the old tried and true platform, because the CISC scheme had more headroom. The difference today is Intel knows they cannot achieve the desired performance, nor are they going to be able to keep costs low with existing memory, or enhanced iterations, and therefore RDRAM is the logical long-term solution. And the support is there from many of their partners.
Fourth, in addition to DDR's performance shortcomings, it is not a thoroughly stable memory currently. Yes, memory manufacturers are producing the memory chips, and yes they are selling them at prices comparable to SDRAM (and losing money in the process), but the ease in which the memory manufacturers are transitioning from SDRAM to DDR has been wildly exaggerated, and is not as simple as the press reports. The motherboard manufacturers are continuing to have problems implementing the higher speed DDR memory, and Intel has gone as far as making changes to the JEDEC spec to ensure their low speed version of DDR will work. Intel is doing exactly what they did in 1996 with SDRAM; they are using DDR, but on their terms. For those who want to make it for Intel, must do so according to their specs. The addendums they have made are to ensure stability and compatibility. Currently, we don't think Intel has a lot of confidence in delivering higher speeds of DDR that will work properly, (at least not without major costs to the motherboards). On the other hand, Samsung reportedly has a 1066 MHz RDRAM 4i chip running and ready for a chipset to put it into. For those not familiar with 4i, it is a forthcoming lower cost RDRAM chip that will bring the difference in cost of manufacturing between RDRAM and SDRAM to within 5aenbsp; Therefore, once this chip is in production, the cost will be inconsequential, and the performance will be much faster than any DDR chip. And, more importantly, it is stable and reliable.
Fifth and finally, the reason DDR doesn't work as well as RDRAM is because DDR is a second-rate version of RDRAM that we believe contains concepts stolen from Rambus designs. It's kind-of funny, because we've mentioned this a number of times in the past, and not once has anyone argued with us on this point.
We believe that if Intel's DDR is being introduced to replace anything it is SDRAM for lower-end systems. RDRAM can't be touched when it comes to performance and headroom, and that will migrate up (and down) in time. DDR has nowhere to go but down (and out), really.
The latest round of negativity regarding Rambus comes from the articles and stories popping up all over the place again about the demise of Rambus, due to Intel's launch of DDR, which apparently is coming sooner than expected.
Intel has stated repeatedly, even to the point of publicly countering news reports that were published erroneously early in the week, they will not ship their DDR Pentium 4s until the first quarter of 2002. The trade press, on the other hand, who prefers to publish their side of the story (because it's such a better read), along with a few misinformed analysts both seem to suddenly confuse a ramp-up with a launch.
We spoke with Intel recently, and they confirmed that the P4 chipsets with DDR would begin to be distributed to motherboard manufacturers in and around the November timeframe. But the "world-saving" DDR chipset will not ship to consumers until the first quarter 2002.
Analysts and the press seem to have lost their manual on Intel protocol, and have forgotten that the company, like they've done with so many other chipsets, sends it out to the motherboard manufacturers prior to the launch date.
What the press fails to acknowledge at this time is that when the "world-saving" DDR 200 MHz memory Intel has finally made stable enough to ship with their P4 chipsets, the performance may be a huge disappointment. But that news will have to wait.
While Rambus' RDRAM seems to be the technology in jeopardy in this scenario, the truth is far from it. As we step back and look at Intel's strategy, it is apparent they are going after AMD more so than anyone. By introducing DDR in the Pentium 4 as a performance dud, and the SDRAM successor, the memory is positioned not as high-end, but ho-hum. Which, in turn, makes AMD's highest performing solutions seem mediocre compared to the performance of RDRAM P4 systems. Ultimately, as RDRAM migrates down, and SDRAM goes out, DDR becomes the low-end and AMD has a perception problem that looks like they only support low-end memory. That's our speculation, but as we all know, perception can be a problem.
Intel continued to reiterate their support for Rambus at their Developers Forum, stating that they had no plans to drop Rambus, and in fact, they would be using more of it in the future.
One of those places is in the mobile PC space. As highlighted in the following link from Intel, the new 830M chipset will support a Rambus local memory chip to enhance graphics.
Link to Intel's latest 830M mobile chipset with support for Rambus memory:
intel.com
Click here for more in-depth analysis from Hager Technology Research.
By Fred Hager
All Rights Reserved Copyright JAGfn.com |
