>Do the Rambus bashers really think that they know more about
>semiconductor design and integration than Sony, Intel, and Sun?
>These are the people who have defined their respective industries.
>Each of them could have chosen DDR if they wanted to. They did not.
>You might want to spend some time understanding why that was.
Intel may be wondering that themselves, what with the myriad problems they've had with RDRAM. Enough problems that their next server product will support DDR instead of RDRAM. I think it's fair to say that even Intel has found the going rough with RDRAM. Of course, they stand to make lots of money if RDRAM becomes the next standard, and they'll have more say in how the DRAM market develops. Also, note that their agreement with Rambus requires that they promote it vigorously, which they've done. But a few months back, they also convened several industry giants to help define the next DRAM standard, and it wasn't based on Rambus designs.
As for Sony, I think someone already pointed out that there's a difference between hardwiring DRAM chips to a circuit board, and mounting them on a RIMM. RDRAM lends itself well to consoles, with their proprietary and non-upgradeable designs that allow them to mount single RDRAM chips on the circuit board.
So yes, many people have wondered why Intel and Sony and Sun would turn to RDRAM. And the answers aren't as cut and dried as they may seem at the start.
And please refrain from the "anti-Rambus" bit. I am not "anti-Rambus", and I'll note that I haven't been one of the handful of people who have labeled anyone else. I'd appreciate the same courtesy in return.
Sir, I am a graduate assistant in my schools legal writing program. I grade the first year law student's papers, and as I explain to the students, if you do not have solid counter arguments, the best you can hope for is a C. I am afraid that your article received a C. In your defense however, all of the other reviewers on this topic received the same grade. It seems that people either zealously support one side of a computer issue or another, without much logic or reason behind their decisions. I am glad to see a positive review of Rambus, because the other articles that I have read, certainly did not explore its good sides. However, it is not acceptable to leave out the fact the DDR ram is just around the corner. This is a very important fact, since DDR Ram will likely be superior to Sdram and for Rambus to succeed it will have to be better than it. Personally, I don't understand your logic. You say that Rdram may catch on in a year or two, well my experience with the computer industry is that a technology that takes several years to catch, never does. In most cases something far superior is developed in the interem and the old promising technology becomes completely obsolete. In any case, you should at least point out the holes in your logic, and try to overcome them with sound reason and logic. This is especially true whenever you write an article comparing and contrasting competing technologies. Afterall, its neither fair nor realistic just to tell one side of the story, even if everyone is telling the other side.
and has as expected already taken some flak for it... mainly I think for being accused of not addressing the latency issue as well as openly saying a lot of his info comes from directly talking to a RAMBUS official. He does however address the DDR SDRAM comparison, which a lot of folks here were wondering about in Sander's article.
So Sander.. dont feel too bad.
On an aside note: To the fellow who was quoting Toms article about how he hated the changes at AMD - a) that quote has NOTHING to do with the RDRAM/SDRAM debate, and b) I read the whole article you took that quote from, and as I recall, the REST of that article was critical of Intel and its roadmap and also discussed its CPU shortages etc.. so you are taking a snippet quote and taking it out of context if you were/are attempting to use it as support for Intel/Rambus and against AMD.
The last info Compaq/Alpha/Rambus I saw was about 1 and half years ago. The heading read:
MOUNTAIN VIEW, Calif. — Compaq Computer Corp. (Houston) will announce on Monday that it plans to use Direct Rambus DRAMs in future AlphaServer designs, defying the conventional wisdom that the Rambus architecture is not well-suited to multi-gigabyte memory systems."
Originally posted by Sander Sassen: But if you feel Rambus is bad from the roots up and isn't worth the penny in your pocket, why read the article in the first place?
While I don't speak for everyone, I occasinally read RDRAM articles in order to try to find one reason I should allow Intel to force this issue upon the market and charge me for the priviledge. I haven't found one yet. That is why I read the article.
Also, as to why I then sometimes reply: when the article has errors and a forum is available to point them out, I will do so. Therefore, the novice reader is not led astray in their purchase decision based on errors on websites, whether intentional or not. The fact that you offer the forum makes me believe it isn't an Intel/Rambus marketing ploy, else you would have quelled any ability for the viewer to talk back. I agree some have gotten too paranoid about the issue. However, while you have been attacked by some, not all posts were hostile even though they pointed out faults or other considerations to your article. For example, one author states the reason for observation A is reason A, but forgot or overlooked the possibility of reason B, is he being paranoid for pointing it out to you? As a webmaster and journalist, you need to find the ability to separate the wheat from the chaff...
The thing i find interesting about that rambus article is that the memory scores i get off my athlon 600 on a K7V Motherboard (With Cheapo pc 133 ram running i think at CAS 3 it might be cas 2) are 421/526 If this is really a review of Ram then this means that even PC133 isn't running up to it's full potential on those P-!!! Systems. That's just my opinion but please tell me how that in the one test that showed the "superiority" of Rambus, with a Athlon 600 and Generic PC133 i can best Rambus by almost the margin it beat it's contenders.
Onegambit, that actually buttresses my point about the FSB being the bottleneck. If FSB maximum is about 1GB, then DDR vs. Rambus can't be answered on performance merits. Price merits however overwhelmingly favor DDR, should it make an appearance in the relatively near future.
Regarding SDRAM vs. RDRAM..., you have simply not asked the proper questions and you have deliberately ignored that which is inconvenient to your purposes. You clearly wrote it to be a biased puff piece (you as much as indicated thus in one of your replies). Well, it was just as you intended and the result is that you have gotten properly flamed for a variety of good reasons.
Understand that I have been following Rambus for a decade (or longer, I can't remember when I first read about the fledgling technology concept in EE Times...). I have had very high hopes for it and have been awaiting its debut in the general computing world. It is just plain disappointing to see such a good idea so poorly implemented. But hey, that's reality _in_practice_. If anything, I've been inclined to give Rambus the benefit of the doubt and some extra time to get things right so don't accuse me of being one of the "negative" types. It's just that the truth is the truth no matter how much it is spun. The term "Truth" includes the WHOLE truth and that is what we have to live with and make decisions upon. Nothing less will do and your article, although on the surface appearing to be comprehensive, lacks so much of import and inflates so much of relative insignificance that it falls into the classic debating category of "sandbagging".
Here are a few (certainly not all) of the relevant issues as to why virtually your entire piece is essentially useless other than to show RDRAM as a _marginal_ technology (Except for zealots on boths sides - Yes, BOTH sides since you have given an entire ammo dump to the anti-RDRAM crowd).
1) You can state all the _facts_ you want (the _logic_ is marginal too) but the comparisons made are hardly valid since you are comparing what is supposed to be THE BLOCKBUSTER breakthrough technology to what is now rather _pedestrian_ standard memory. The (more or less readily available) state of the art in SDRAM is CAS2 PC133 running at its top speed. And using VIA's chipset for the PC133 comparison is ludicrous (the BX is still superior for Intel CPU's). If RDRAM can only eke out a few percentage points of performance benefit in a few circumstances against OLD technology, what does that say about it? Pretty weak stuff IMO. And what about using virtual channel or other so called "advanced" memory technologies. (It may have made for a _very_ interesting article with RDRAM hard pressed to show any advantage at all.)
2) You succeeded in showing that RDRAM can demonstrate some performance and cost benefits under certain conditions but they are not apples-to-apples. The reference to the Sony Playstation2 is a non-PC application (yet a valid technology demonstrator) but is irrelevant to the purposes of the article. Why is it even mentioned?
3) Where is the comparative commentary on DDR SDRAM and published benchmark comparisons? Conspicuous by its absence and very, very suspicious. Go ahead and pick DDR apart(if you can) and then publish the standard benchmarks to prove your point. (There's the gauntlet thrown down, now pick it up and post the results, good, bad or ugly.)
4) It is very nearly ridiculous to hype a performance difference of five percent or less, especially when using synthetic benchmarks. Of course, techno-weenies just love to wax rhapsodic about how one configuration or another "beats" something else so they can either claim personal superiority or superior knowledge. It used to be fast cars or stereo systems (Hmmm, still is for some). How juvenile. Just give us the facts without the breathless hype, please! (Sander, I am not really addressing you in this regard.) In the REAL world the mix of applications and operating circumstances is so varied as to completely bury such trivial differences. Talk about lies and damned lies!
IGNORING THE TRULY SIGNIFICANT
5) A really NON-MARGINAL difference is in the all important price/performance ratio. RDRAM is a monster loser here, in some cases by nearly _an_order_of_magnitude! Now THAT is worth talking about and making decisions on. You don't have to be "paranoid" to see that RDRAM is a lousy investment.
IGNORING LONG TERM ISSUES
6) It has been my long experience that any investment in a "new" technology should either be made with either a very short term perspective or a rather long term view (it is difficult to qualify a middle ground).
6a) The short term is great for "gotta have that feature/performance/function NOW" kind of year or so timeline and can often conveniently ignore long term issues such as maintenance costs, expansion costs, repair costs, rate of depreciation, technological obsolescence, etc.
6b) The long view requires a much more demanding analysis and incorporates all of these and more. Many of the replies were dead spot on regarding how the higher cost of RDRAM (and very likely _future_ higher cost) makes steering clear of RDRAM a no-brainer. My clients insist on getting more than three years of life out of their systems (desktops, servers, networks, software) and we very carefully help them select technology which is usually value priced just behind the bleeding edge. The recent i820 recall puts RDRAM into a special class that we call "the hemorrhaging edge".
6c) It is absolutely clear, you should only buy systems using RDRAM if you have apps that can be proven to benefit from it or you don't care about long term issues. (I wonder how all those Dell customers are feeling about their purchases these days and how they will feel 2 years from now.) If Rambus doesn't make it as an economical, mainstream product or is superannuated by DDR or QDR SDRAM it will merely have proven to be a waste of time and money. Currently, I cannot think of ANY application that I'd be willing to spend extra for on the up front cost to buy the marginal performance possibilites of RDRAM. When I factor in the dead-end nature of all current products (both the RIMMs and motherboards) it makes me run the other direction.
APPARENT CONFLICT OF INTEREST
7) The criticisms of the entire site are certainly valid in that there is the appearance of a "vested interest" in the outcome of the article. It may only be an _appearance_ but it does cause one to pause and consider the objectivity of the source. Then, when the article is so, so biased, almost to the point of hyperbole, it makes one wonder. I agree that questioning your parentage, recklessly speculating about your owning stock, being bought by Intel or Rambus, etc. are out of order. But I wonder, did you get paid in any way for this article and if so, who paid you and who paid _them_?
In summation, all your analysis and writing are very nearly for naught since much of it is essentially irrelevant. I know that kind of comment stings but I hope that you can learn and grow from the experience. You are a talented writer and have great potential but you need to learn to _ruthlessly_ steer clear of bias if you want to be taken seriously.
You should however redeem yourself first by acknowledging and addressing the very valid criticisms that have been posted. At this point your credibility is pretty well shot to pieces. Maturity demands taking responsibility when we screw up and you just plain screwed up. It will be interesting to see if you can do this without hedging.
For the record, I'll be glad to take you up on rebuttal option #2 if the rate is high enough to make it worthwhile (I doubt it but let me know and we can negotiate). Better yet, let's see another mano-a-mano test with state-of-the-art products later this summer when new hot rocket CPU's, supporting chipsets and DDR SDRAM are all available. Pitting DDR against RDRAM is where the _real_ debate will be a bit later this year. By Q4 of 2000 we should have a much better understanding of the fate of Rambus and the applicability of their technology to mainstream computing.
Just saw the news on www.theregister.co.uk regarding Intel making a U-turn on the Willamette chipset. It seems that Intel will make two chipsets just in case that Rambus will not come down in price, going Rambus only would yield a Willamette system seriously more expensive than any other comparable system.
However it seems that the other chipset they make will be a SDRAM and not DDR... The Foster (basically the same CPU but in differnt package) will support DDR, so why not Willamette?
Without Intel's 100% support there is no way that Rambus will be the memory of the future! They might get some design wins for special applications like PS2 etc, but not in the mainstream computing.
Anyone know why Intel killed of the "Greendale" chipset for mobiles? It was supposed to come out in May'00, then postphoned to Sept'00, then later killed. It was supposed to be a high-end chipset with mobile-Rambus.
I'm really suprised that most of you are suggesting that I or even HWC is 'owned' by either Rambus or Intel. Allow me to fill you in on some of the work that has gone into writing this article. Most of the info you'll see posted came from Rambus, Intel, Samsung, Micron, VIA and AMD a whole slew of company datasheets and a compilation of other websites' data.
I can see why you might not clearly understood the timing issues with RAMbus then, I have been pouring over those documents trying to glean usefull information, but it seems that they either failed to do the complete theoretical performance analysis, or refused to disclose it to the public: either of which is an unforgivable sin.
I've asked each and every one of these manufacturers to send me a package with some indepth technical background info, which I them compiled into the info you see posted. I think I've done a fair comparison by all means, all the drivers and benchmarks used were recommended to me by the manufacturers of the applicable chipset/motherboard/manufacturer, you can be sure that they make sure I use a driver/benchmark that has their product performing at best.
That is not nesescarily true. in many cases, the people you deal with are marketing, or at best Tech. Support. neither of which are really qulaified to make statements regaurding the performance of their companies products. They ussually just grab the latest version from the repository on the assumption that it is the latest and greatest. In many cases, latest and greatest is defined as being the most stable, not the fastest.
One of the things that really offends me with these kind of flames is the fact that you people tend to absorb all anti-Rambus/Intel articles as the word of God, but flame one that does shed a different light and dares to take a different point of view. Who's being objective here? I disclose all info, drivers used, benchmarks etc. I don't see any of that in those other 'articles', they just summarize their results without posting a single screenshot and you tell me you're buying that?
I have compared your benchmarks with Tom Pabst's of Tom's Hardware guide, and both your benchmarks appear to be saying basically the same thing, with one key difference. You ommitted the inclusion of the 133 Mhz intel BX set. VIA is notriously bad in memory performance, and compared to a simply overclocked BX they cant compete.
I hold a masters degree in EE and a bachelors in ME, so you can be sure I'm able to tackle any technical issues, thus I might not do the benchmarks you'd like to see, I do however make sure they're objective and as well documented as possible, the same cannot be said about many others.
Well, A Bachelors in ME is entirely Irrelevant to these discussions, and the EE degree could very well be equally useless in terms of digital processor design. I would consider a degree in Computer Engineering to be the ideal qualification (behind of course industry experience), followed loosely by CS, but a degree in Electrical Engineering tends to be Analog oriented, and only a specialized breed of EE's are actually qualified to discuss digital systems design.
Last but not least, the theoretical discussions regaurding maximum memory data rates are all technically incorrect because they do not include the channel overhead. The maximum data rate of SDR SDRAM (with overhead) is 337 MB/s not 800 MB/s as stated by your theretical analysis. This is assuming a 3-2-2 timing, which you might be able to improve to a 2-2-2 timing for 100 Mhz systems for a throughput of 356 MB/s. for a
133 Mhz system, we can achieve 447 MB/s effective max throughput. These numbers are achieved assuming 8 consecutive RAM reads or writes (max allowed by current SDRAM), which gives us 3 clocks for the precharge, and 2 clocks for each aditional xaction. for a total of 19 clocks for out 64 byte xaction. if we take 64 / ( 19 * clock period ) we get the max throughput given above.
RDRAM can be evaluated in the same way. ultimately we can measure the RAM in terms of prep time, and then individual access cost of the form Setup Cost + ( n * access cost ). Using this formula, RAMBUS should be able to fill in the blanks, but they have been surprisingly hesitant to provide these simple numbers for us, even under the ciircumstances which are supposed to show their product in its glorious form. I have gleaned the Access cost (or I think I have, I still can't be sure) from their technical documentation, but in reference to the setup cost, all they do is refer cryptically to "once the bank has been precharged". No timings for this phenomenon are provided. The access times they provided in the documentation I found claimed that their Pclk runs at 2.0 x Sclk, and the Sclk was 1/4 the Master Clk, so this leaves the Pclk at 200 MHz, but again that was unclear because no indication of whther or not this was counting the Master clock as a single edge, or double edge source, was given. However, the one thing it was clear about was that the access cost was 2 Pclk periods for a single access. This would leave them (assuming no access cost at all) with 2 clks * 32 access' (remember it costs us 4 times as many access's to get the same data from RDRAM) for a total of 64 Pclks to get at the data. so our total throughput comes to be 160 ns for 64 B ( assuming that our Pclk is running 2 * 1/4 * 800 = 400 ). This suggests that our Rambus should be faster, for large number of consecutive writes ( which it is), but leaves the question of just how many we have to do to make the difference, since they dont want to tell us the initial setup cost ( which is most definitely not 0). This gives us a maximum Rambus throughput of 400 MB/s. So the real question is what is that setup time. IT would seem that it is the critical piece of the puzzle, and the one rambus doesn't seem to want to document. as long as its small, RAMBUS wins, but if it gets larger than 30 ns, then RAMBUS looses. (typical Precharge time for an SDRAM is 24 - 30 ns.)
I'd like to say I'm not a EE, So I won't debate the technical aspects of the article. But my position is this, I like Intel products, I'm glad AMD exists and I'm completely for any new technology that comes out that increases the speed of my computer. But....As a consumer, both business and personal, I don't like Rambus. It is being forced, repeat forced, on consumers. The company forcing Rambus (Intel) has a financial stake in that company, FACT. If Intel really believed that Rambus tech. was a marked improvement over other Dram tech., why not allow it to prove itself on its own? Why force it? I should throw it out and buy Rambus because Intel says so? Not even factoring in price to performance ratio? Thank heaven for AMD. If rambus truly was a better technology wouldn't computer users embrace it? These are the same "computer geeks", and I say that with pride since I myself am one, that will drop $350 on a new Geforce 2 card just because they can get 10 fps more in quake 3. It's not the media that's dogging Rambus, it's the consumers! Most of the sites and reveiws I've read are pro-Rambus. It's only the newsgroups and replys to advertisements like these that you see so much hostility. We don't want it, not at least at the present time. Let the marketplace decide if Rambus is wanted, not Intel, not Rambus and certainly not wallstreet.