As a manager of a computer manufacturer, please let me express a few points:
1. From a technical point of view, I regret that there was no technical benchmark of two competing technologies: DDR vs RDRAM. Benchmarks our company did and many other studies show that RDRAM is the looser in most of the particuliar software applications tested against DDR. Certainly if you consider the chipsets for both AMD and INTEL CPU's supporting 266Mhz DDR this year !
2. But I think the economical point of view is much more important: Around 55% of all computers sold are business computers, and the 45% rest consumers users. Both type of users always have a budget range, and all computer buyers will try to maximise performance for the price. So stating that the price difference is not important is total nonsense.
The point here is simple: for the next generation of computers, 128MB to 256MB RAM will be standard. So considering this, a difference for the end-user (consumer or business-user), of around 1.000USD for 256MB (or 500USD for 128MB) IS very important. What can you do with 1.000USD difference per PC ? You could buy the fastest CPU available, the fastest HDD or VGA, take an ADSL Internet, or a combinatoin of those, or even buy a second computer ! Imagine that: ONE RDRAM costs more than a whole new PC ! All those alternatives are money much better spent.
Essentially speaking, this is why (nearly) all the international OEM manufacturers are refusing the RDRAM solution (not in public, Intel is too dangerous as a ennemy). It is not economically justified to pay 600% more to get 5% more !
So I don't blame Sander for being wrong. Pressure from the Intel / Rambus / and some other parties is extremely strong and effective. But it should have been clear to him that all the OEM's (PC and memory manufacturers) in the world (except Dell, but everybody knows why) can't be stupid when they stubbornly refuse RDRAM during more than 12 months now, and sustaining the alternative technology DDR for the near future.
Sorry for the long reply, but many technical reviews do not consider the performance/ cost ratio, and this is an essential mistake. When you see a performance difference of barely 5% in most cases, but a price difference of 50% up to 600%, I am not even interested in reading further anymore, because it make absolutely no sense.
Rambus Direct RDRAM, need I say more? This week I've posted my follow up to the original 'Lies, Damned Lies and a Different Perspective' article, covering in great detail the workings of RDRAM, problems with SDRAM and other memory architecture misconceptions. My objective was to provide a 'Different Perspective', as RDRAM has been taking quite some negative press lately, and there is only one other article I was aware of that actually provided some in-depth information about its workings.
Hence I set off to counter most of the bad press and misconceptions that floated around the net, not to please Rambus or Intel, but in an attempt to educate the reader and to provide a broader perspective, which provided room for further discussion as well as a more solid ground to base one's conclusions on.
I was surprised to find out that most readers took disgust at my approach and slated the article as being 'Rambus/Intel Propaganda', which was never my intention from the onset. More disturbing however is the fact that these people have a take on the Rambus/Intel issue that almost borders on paranoia. Every good word being written about it must have come from the Rambus/Intel propaganda machine, every benchmark conducted has been doctored to make the Rambus/Intel come out on top, every benchmark used was especially written and used to make sure Rambus/Intel would do best, and so on. Needless to say this is all nonsense as I made sure all parties validated the benchmark applications and all configurations were well documented to provide enough information for those that might be willing to reproduce my results.
If those were isolated individuals I would not have worried so much, what does worry me though is the fact that there's herds of them all flocking together to 'kill' any pro-Rambus/Intel information from the start. It is not so much the anti-Rambus/Intel part that I'm worried about, but it is the motives of their actions. Anybody in the right state of mind would welcome an article covering some aspect of a technology not yet presented before; it broadens the view and provides more room for discussion. I'm not biased towards either technology, SDRAM or RDRAM, it is not my job, nor my task to have a bias, I'm just providing information based on research I did, benchmarks I've run and arguments and misconceptions I wanted to address. Neither am I asking anyone to agree with my conclusions, they are presented after an evaluation of the data at hand and provide a write up of my findings and a verdict of what I feel is an accurate representation of facts, not fantasy.
As with all things everyone's entitled to their own opinion, that's one of the most precious of all human rights. 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? We all know that the general public feels Rambus Direct RDRAM might not be the next best thing, I'm not stating it is. I am however providing valid arguments, supported by real world benchmarks that it isn't all bad, and that the current RDRAMs might not perform as well as we'd expected them to. As the performance we've come to expect was based on all of the marketing mambo-jambo surrounding its release, but it does offer a good concept, high bandwidth and good scalability, although price still remains an issue.
What I see happening here is what happened to Betamax, VCR-2000, DCC and all of those other concepts that did have potential, but never made it big time. All those concepts have quietly gone away due to the fact that negative press was blown out of proportion and completely did away with any positive press. The people that actually made those headlines usually weren't the ones who had the technical know-how to fully back their arguments, but they did have somewhat of an established authority in the industry, combined with a receptive audience, they could have made people believe that your 8086 would run Windows2000 when you patched it with product X.
Point being it is always easy to criticize or debunk a product, an article or even an individual, but providing arguments based on facts and logic, adding to the discussion instead of tearing it down is something many aren't capable of. If the general consensus is that something is bad then it is easy to come up with all the disadvantages, but you'll be hard pressed to find any advantages, as the negative press will have gradually filtered those out. What I tried to do is 'reboot' the system and provide some insight into what initially motivated Intel to adopt Rambus as a memory architecture and what those advantages were. What you do with that information is up to you, but don't shoot the messenger.
I'll be going over the arguments and other information provided in this thread in the next few hours and will try to read through the BS that some felt they had to post. Expect a new post with some more feedback in the next couple of hours.
Iíve read through the thread, and apart from the sometimes really personal and rather childish comments (if you canít write in a neutral manner and have to shout, *****, call me names and make acquisitions to get your point across youíve got a serious problem) I think that thereís a couple of very valid arguments mentioned.
I donít presume to have all the answers and Iím certainly not going to be so ignorant to dismiss your arguments just because I havenít thought of them, I would like to propose the following two options:
Iíd like to do a follow-up to the article, discussing in detail the arguments youíve mentioned and any misconceptions or simply Ďerrorsí that may be found in the initial article. I will however depend on you to provide me that information, complete with URLs, references, quotes or any other means for me to verify the argumentís validity. I will not be doing that research myself for the plain and simple reason that Iím not going to have my findings questioned a second time. If you have a valid argument and are able to back it with facts make sure you can Ďput your money where your mouth isí. Iíll work those arguments into a follow-up article which will also include my comments.
I have freelancer positions open, if anyone has the guts to stand up to his claims and is willing and able to write a follow-up to address any errors or misconceptions in the article as well as provide new details and arguments (pro or con). Iím more than happy to give him/her/them the opportunity to do so. Even better, the person(s) in question will be paid for the work he/sheís done by standard freelancer rates. If anyone is willing to take me up on this offer it might be worthwhile to have the support of the others, backing him/her with the necessary information. I will not edit nor change anything in this article, other than clear attacks against me, HWC or other people.
I for one hope that this will finally do away with the acquisitions that HWC is biased, or that weíre bought. This is your chance to make yourself heard and express all of those concerns, arguments and other things you feel you have to mention.
Is it possible that every debate, article and post about Rambus vs. other memory technologies is ignoring a very critical, obvious and yet easily overlooked element?
To date, no one has mentioned the fact that memory and processor must still communicate across the 133MHz processor bus, and further, most posters seem to dwell on the aggregate bandwidth per second of the memory bus and not the speed at which a byte of memory crosses the processor bus. The Easy to confuse and overlook, but different for analysis purposes.
Thus, I assert that present processor bus speed (133 MHz) is the performance bottleneck, and that every faster memory interface (DDR, Rambus, etc.) cannot be fully exploited (or even properly measured) until the FSB is increased accordingly.
Before I start, I am ignoring the i820 chipset. It seems safe to say that it is a dog and should be put out of our collective misery.
Tom's Hardware demonstrated that a BX chipset at 133MHz on average performs within 2% (I've done a simple averaging of performance deltas using the raw numbers that he provides) on the benchmarks he uses.
Bert McComas has performed an excellent analysis of DDR vs. the OR840 with dual Rambus channels. One observes two things in his article that are worthy of note. While DDR does sometimes beat the OR840, it's hardly clearcut, and it is indeed highly benchmark dependent. Also, under NT, DDR and dual Rambus perform quite similarly.
So it all begs the question: If Rambus and DDR are so great, why don't we see one winning consistently (i.e, clearly and almost always beating rival technologies)?
No matter what, it takes 7.5ns for a quad word to travel across the processor bus, and in aggregate, you can get *at most* 1066 MB/s across this bus. Even if DDR, PC133 and Rambus are all going at their maximum bandwidth, the stream of bytes must be slowed down to processor bus speed. The maximum theoretical communication speed between memory and processor is 1066 MB/s. Period. No amount of memory magic is going to get around this.
This ends the "proof" of the assertion.
So how do we find out which is better?
If one factors in price, Rambus loses. End of story. We can all get on with our lives.
So, let's forget price for a moment and consider performance only.
We need a benchmark created which simultaneously stresses all subsystems to truly see. Just for the sake of discussion, let's say this benchmark creates N threads for each of the subsystems it's going to stress. One for some processor intensive task (to apply stress from the processor bus), one for a graphics intensive task (to apply stress from the AGP4X side -- say using a GeForce 2), one for a DMA oriented task (to stress the memory bus without ever involving the processor -- probably a memory-to-disk task or some memory-to-memory DMA task if that is even possible).
If Rambus is truly worth it's salt and can live up to expectations, then it should be able to service all of these requests. The reason is that if you add up the numbers for just two of these (AGP4X and processor bus), you find total maximum demand on memory from these two subsystems is exactly 3.2GB -- the theoretical maximum for dual channel Rambus. Indeed, it would also in theory show whether DDR could handle such a load.
Even if you disagree with my analysis, I think we can all agree that FSB is most likely the real bottleneck which prevents any meaningful benchmarking. Further, even if you think I'm full of it, the benchmark proposal is a good one for stressing out such fast memory subsystems.
So if you gotten this far , what do you think?
[This message has been edited by RasputinNOLA (edited 05-11-2000).]
Just read your reply - very reasonable response. The part that read -
"We all know that the general public feels Rambus Direct RDRAM might not be the next best thing, I'm not stating it is. I am however providing valid arguments, supported by real world benchmarks that it isn't all bad, and that the current RDRAMs might not perform as well as we'd expected them to. As the performance we've come to expect was based on all of the marketing mambo-jambo surrounding its release, but it does offer a
good concept, high bandwidth and good scalability, although price still remains an issue."
- would have been great if it was in the original article.
Unfortunately, the original did seem to me as too close to RMBS hype, and did not communicate the above.
I personally think there is a place for Rambus memory (don't think it will ever cut it in a computer IMHO) in many other devices.
Not fantasy at all. In the real world, people often spend their OWN money on computers. Value counts, it really does. The sdram solution blows the doors off a similarly-priced rdram solution. Even I wouldn't fork out the extra money for a 1ghz, but would be happy with a much cheaper 950 athlon!
I choose two 128mb sticks because of an article I saw on cluboverclocker.com today benchmarking the 'interleave RAM' feature of the via 133 chipsets. Puts rdram even further in the performance hole.
With regards to your response to all the feedback, I'm not as technically savvy as some on here appear to be, but I do know how to write essays/columns, as I had to do a fair bit of it when getting my degree in Political Science, so I'm volunteering to write a condensed "RAMBUS/RDRAM Rebuttal" for
people - provided that they send me their objections/proof/URL Links to me.. there certainly seems to be more then enough objections to it, but maybe ppl arent sure they can write a column.. so I'd be happy to do it for ppl. Uh.. I dont know how that would work for getting paid a freelancer fee, since obviously I'd just be putting together ppls idea and compiling them into 1 essay, and I could understand if ppl wanted to get some compensation out of it, but.. I'll worry about that once I see if there are any takers to my offers.
A written list of what you think is wrong with RDRAM,(or what is better with SDRAM - any variety) as well as benchmarks/sites/URL's/proof of what you say, plus any other suggestions can be sent to my email address: firstname.lastname@example.org -- I also will give proper credit to people for what they send in and what I deem to be good enough to include in any subsequent posting.
If I get enough of a response.. I'll submit a column to you Sander. Thats my counter-offer to anyone who wants to say something but might be afraid to post a column.
Since on this article his purpose was not to "glorify" RDRAM, but to compare VIA, 440BX and 820 chipsets.
On the page where he compares 640x480 resolution Quake III the test result is:
440BX 1GHz 115.5
440BX 800M 114.1
694x 1GHz 107.7
694x 800M 104.3
440BX 600M 102.8
i820 1GHz 97.5
694x 600M 94.0
On the other tests the results were similar, the i820 with 800Mhz RDRAM lost every single test Sander put it to. (Note that on the Sysmark 2000 the RDRAM had problem on 800MHz so he went down to 700MHz...)
Eventhough i820 and RDRAM lost in every test his conclusion was:
"If youíre thinking of doing an upgrade, or even the purchase of a whole new system today, weíd opt for an Intel 440BX or VIA 694X platform, as these two platforms currently offer the best price/performance ratio and will undoubtedly last for another year or two. By that time, RDRAM may have matured along with supporting chipsets, creating opportunities for the next upgrade or purchase."
So, less than two months ago when he made a detailed test his conclusion was that RDRAM wouldn't be good until 1-2 years from now (when the price went down). Well, the DDR will have replaced SDRAM by then and I have not seen anyone speculating in even higher RDRAM than 800MHz at this moment, probably since the yield for 800MHz is too low.
So the first article he wrote was with test results etc and it was too benefitial to SDRAM, then he went back and made a new article without testing to back it up. This last article he made the SDRAM platform with wrong drivers etc in order to show some benefit for RDRAM... wasn't your first testing good enough?
Sander, did you buy stock in RMBS when the price was high? Are you trying to recoup your losses by writing these articles?
Since your original testing, Van Smith's, Bert Incomas, DELL testing all show that even SDRAM PC100 is challenging the 800MHz RDRAM why create an unfair testing setup for the sole purpose of giving RDRAM higher speed (which you actually failed to do in all tests...)?
I really hope that Micron can provide you with a DDR system, I think that all of us reading your articles would love to see how you end up comparing them.
Thanks and I look forward to your next article, they are much more fun to read than all those pro-SDRAM, your article have a tendency to challenge people, to make people think, that's really good! Keep up the good work!
(Ps. I would sell my Rambus stock now before DDR comes out, don't wait any longer you will end up loosing more than you expect)
I must say that I am not surprised at the amount of Intel/Rambus bashing that has been going on in this post. It almost seems that it has become "fashionable" to bash anything that has to deal with Intel. What I am about to do may seem like social suicide but to each his own right. I fully support Intel. And on top of that I do not support AMD. I am not saying that the Athlon is a bad processor, but that the current slot-A version is not living up to being a 7th generation processor. At lower clock speeds the Athlon keeps up and sometimes beats the PIII, but at higher clock speeds the cache kills it. This is not saying that the Thuderbird will not beat the PIII, but that is what was excpected in the first place. But this article is not on Intel vs AMD, but rather Intel/Rambus vs. the rest of the observable universe. I thought the article was very well written, had very godd technical data to explain itself, and finally, was unbiased in its representation. I think that based on the data put in front of us, it showed a clear advantage to Rambus. He even stated that current software was meant to run on older systems. I good paralleism that I can think of is the current situation with Video Cards. The new Geforce 2 is an incredible card. (I think that Nvidia will beat the pants off 3dfx in the next product cycle. V5-6000, give me a break.) But it is hurt by bandwith issues. Primarily the lack there of. I'm not saying that Rambus is better than DDR Ram, and I don't see anywhere inthe article that he ever states that, but rather that Rambus is better than SDRam. Remember that i820 is a very new chipset and that the bios needs to be tweaked. The BX chipset is old and probably close to perfection. Even though the VIA chipset is new, it is based on an old idea and that alone gives it an advantage. Just look at situation with the Athlon boards. The original boards by AMD had issues, but allready with new chipsets from VIA, they are getting better. I believe that givin time, the i280e, which is replacing the i280, will eclipse the older BX and VIA chipsets. Finally, I believe that attacking Sander based upon his article is childish, and only shows the lack of wit that the majority of population has. I'm all for Intel/Nvidia/Microsoft just because history shows that the industry giants are the ones that plow the way and are the least affected by times of trouble. I am finished now and will brace for impact. lol
Your analysis of AMD vs Intel is questionable, but hardly the topic of discussion here.
I have argued the technical merits of the article were lacking based off of my own experience as a practicing EE and from my personal geeky reading of various web articles. I have no personal bias against either Sander or HWC. I simply believe the article to be incomplete and in some cases inaccurate (much as I believe your assessment of AMD and Intel to be).
Excelent point; however, how do you explain the increased frame rate of highly AGP bus intensive programs such as Q3 under PC800? This is not an aquisition, but an honest question.
could you e-mail me? I have some questions about your post that don't warrent being posted here.
[This message has been edited by OneEng (edited 05-12-2000).]
[This message has been edited by OneEng (edited 05-12-2000).]
With greater knowledge comes greater understanding!
Nice job on a tough, controversial issue. Notice how the anti-Rambus faction starts by attacking your integrity, because they can't attack your data? I just wish one of the idiots would consider why Sony, Intel, and Sun have gone with Rambus for high-bandwidth high-speed bus solutions.
is just the first in a series of positive reviews about to appear on popular gaming and hardware sites. It looks like Rambus management and PR realize that mindshare is something they must build, give the number that has been done on them by shorts and Intel bashers.
is going to be publishing a review of Rambus in a couple of days that is also based on discussions with people in the company. Here is Anand's preview:
"In any case, let's talk about what's been going on over here. I had a meeting with Rambus last week (that Mike was horribly late to ..) and got quite a bit of interesting information. I've stayed pretty quiet on the issue of Rambus (or should I say the "how should we attack Rambus" debate) but from the minute Rambus sat down in the lab I asked them just about every question that you've been asking me and I've been wondering about as well (including thirteen variants of the question "$*#(@$*)#@ $800 for 128MB?!?!") and I'll be putting all of that into an article which will hopefully make it up sometime in the next two weeks, it should be quite interesting."
Even Tom Pabst--who original Rambus articles were written by Bert McComas, a well-known DDR marketing flack--has recently been putting Rambus in a more positive light, starting with some unusual criticism of AMD:
"At the same time there have been a lot of changes at AMD that I personally don't really appreciate. Many important and good people have left this company...disgruntled by AMD's changing internal policy. I wonder if AMD will do well without them. One thing that I have noticed bitterly is the demise of AMD's dealings with the press. While in the past AMD was a lot more pleasant to deal with than Intel, it now seems as if the new success went to the heads of the few remaining old executives of this chipmaker. Today it's a lot more fruitful and enjoyable to deal with Intel than it is to deal with AMD."
Dr. Pabst then followed up with a brutal condemnation of VIA and problems with it's KX133 Athlon chipset.
You seem a bit thin-skinned. Despite a number of posts that deal with the issue at hand and do not slip to the level of personal attacks, you still focus only on those people who have criticized you personally. You ignore a number of posts that call into question the technical accuracy of your article. You ignore some direct contradictions of the information you've presented. And you dismiss it by asking that someone else write an article addressing those points that have been raised. Yet again, you sidestep valid criticisms by taking aim at the very messages you're better off ignoring.
You ask "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?". You yourself have stated, several times, that the aim of your article was to present a different side of the issue. Thus, I would expect that you wanted those who didn't like Rambus to read your article. And anyway, crying that "if you didn't like it, why did you read it?" is a poor defense.
And bear in mind that the market will determine the success of technological progress, not a handful of naysayers in the press. If Intel hadn't been stumbling so badly with its i820 chipset, and if RDRAM wasn't so expensive, the market would've adopted it without a single qualm. Sometimes, a technology is panned because it is a bad technology, not due to some hidden agenda that is as difficult to rationalize as it is to explain.
In any case, you seem to be angry that anyone would dare to criticize your work. You should dismiss anyone who simply accuses you of being paid-off, and focus on the actual points being raised by several people here, since a number of them seem to point out gaping flaws in your logic. Are you willing to address those, instead of daring everyone else to "[have] the guts to stand up to [their] claims"?
At the very least, explain why you feel that Rambus is so promising when it barely competes with the memory technologies it is supposed to supplant, and why early benchmarks show it losing badly to competing memory technologies that are soon to debut in the market!
*hodb says:*"Notice how the anti-Rambus faction starts by attacking your integrity, because they can't attack your data? "
If you've read through the messages posted, you'll see numerous people questioning, and outright contradicting, specific technical information and data presented. Thus your accusation (not acquisition, as I've seen spelled several times now) doesn't hold water.
And the links you offer hardly provide a ringing endorsement of RDRAM. Firing Squad simply offers a Q & A with a Rambus PR official, and basically tossed him a bunch of soft questions. AnandTech offered nothing more than the promise of an article dealing with RDRAM. The links you provide from Toms Hardware have nothing to do with RDRAM, they are simply critical of AMD and VIA. Perhaps you should check out his latest report, which bashes Intel for further problems with their latest chipsets (Problems, BTW, which arise from Intel's attempt to get the i820 to work with SDRAM and not RDRAM. Funny how even Intel has to turn to a "lesser technology" to help it overcome the problems that 'tomorrow's memory standard' is vexing them with).
And Anand's dream machine is based on the premise that money is no object, so it's no surprise to see that the 512MB RDRAM for it costs $2000. You can get 1GB of high-quality PC133 SDRAM for half of that.
Actually, OneEng, I can't answer that, though I wish I could.
I simply wanted to point out that the 133 MHz FSB, which heretofore had matched the memory clock-for-clock in SDRAM systems and was never a concern, may very well be the limiting factor in answering the question of which memory technology is really superior.
On the one-hand, the Sandra scores seem to favor the Rambus, while on the other, some scores favor the BX. This tells me two things: 1) until the FSB and its bandwidth go up (or everyone buys EV6 Athlon motherboards), we can't really give a definitive answer to the "which is better" question and 2) don't waste your money on Rambus at today's prices if you're building a computer right now.
BTW, Sander Sassen *may* be right about the pricing situation. I've noticed a mini price skirmish has broken out on Pricewatch regarding 128MB Rambus PC800 RIMMS. It went from $560 or so to $520 in a few days (from 5/8 to 5/12). 256MB PC700 seems to have just broken the $1000 barrier. If these price drops are for real, it may be a good sign (well, maybe for Rambus et al).
Also, by the way, since these disclosures seem important , I hold no stock or bond invenstments of any kind in Intel, Rambus, AMD, Micron, etc., etc., because I'm too poor. The only thing that I want to see in all this is for Rambus prices to achieve some parity with PC133. Why? I was given an OR840 motherboard (don't ask) and would love to make use of it, but can't due to the exorbitant cost of RIMMs.
This features the same Intel OR840 SMP motherboard, but has dual P-3/866MHz, a larger Ultra160 10,000-rpm HDD, and the ELSA Gloria (Nvidia Quadro) video card--kind of the GeForce's big brother. $5,500. This is the fastest production workstation on the planet based on the x86 architecture. I have a $100,000 SGI Octane at work, and I think this would be faster.
On RDRAM: the marketplace will decide. RDRAM has to drop to no more than a 20% premium over PC133 SDRAM, and it must be shown to make a significant (20-30%) performance difference on 1GHz+ processors, particularly the Willamette. We will know for sure over the next 12-18 months.
I always come back to this: 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.