Interestingly, the OpenGL benchmarks, which stress workstation graphics/3D performance, have the i820/RDRAM combo well ahead of VIA's PC133 chipset. There's a bit of irony there. The i820 has no niche!
Consider- workstation buyers won't really care about the price premium for RDRAM if it offers the best performance. But with only two RIMMs, you're looking at 512MB RAM as a maximum. For most workstations, 512MB is the MINIMUM. Those folks will look to the i840 instead, and pay the ridiculous premiums required (which is still a lot less than an SGI station! ).
For the home user and casual/hardcore gamer, the VIA chipset is the solution, since it beat the i820/RDRAM combo in EVERY SINGLE TEST of office/productivity/games application performance, including programs like CorelDRAW and Photoshop. That market considers price as highly important, and will not pay the premium required to use RDRAM, nor are they willing to pay more for i820/SDRAM, when it is clearly slower than VIA 133A/SDRAM (or even BX/SDRAM).
Simply put, neither market finds i820 a compelling solution. That may be why Intel is rumored to be holding back the i815 chipset. At this juncture, it is probably going to do more damage to i820 than it will to VIA's chipset.
The "Latency" of RAMBUS memory should be defined by the length of time it takes from the start of a memory transaction, to the time the first byte of information traverses the bus to the target. The article only addresses the amount of time for the clocked control lines to wiggle. Since RAMBUS memory is protocol based, a message must be formulated and sent from source to target to initiate any data transfer. This process does take time and is the "high latency" people talk about with RAMBUS.
The bottom line is that it takes longer for RAMBUS memory to get small amounts of information across the memory bus if they are from several different locations. RAMBUS transfers are more efficient for large continous transfers where the initial hit for the request is made up over thousands or even millions of bytes transfered after the control protocol message.
With greater knowledge comes greater understanding!
With greater knowledge comes greater understanding!
For PC133 the same is true. But you
45 + (2 x 7.5) + (3 x 7.5) = 82.5ns for PC133 SDRAM
A good PC133 (say, Micron grade -7E)
can run 2-2-2 at full 133MHz,
therefore the correct answer again is:
(2+2+2)*7.5 = 45ns.
Only 45 nanoseconds!!! All the above
assumes the worst definition for
latency - from the end of previous
transaction to first Q-word of data.
Now please do your homework about
Rambus, but do not forget the following:
a) Cas-to-data delays on all chips
are artifically increased to be equal
to the longest path, or up to 7.5ns,
to avoid data collisions.
For your information, a 16-chip RIMM
is specified to have exactly 2.06ns
of finger-to-finger propagation time.
So two RIMMs are already 4.1ns, plus controller-to-RIMM, plus between RIMMs.
So, the COL-to-data delay is 20ns+5ns at least.
b) The data from RAMBUS are coming in
16/18-bit words, while all chipset
internals operate at 64/72bits.
Therefore to assemble the whole
internal 64-bit word you need to wait
for the END of the 4-bit times, or
c) The Ras-to-Cas delay is always
8 clocks, or 20ns;
d) I have no idea how long it takes
to open a new page, but assuming that
the physics of DRAM is still the same,
it will take no less than 15-20ns as
for best PC133, but the whole RAMBUS
timing goes in chunks of 10ns,
therefore take 20ns as well.
Now do your math.***
Then, you write:
"The superior electrical characteristics
of a RDRAM eliminate the two-cycle addressing problem, requiring only
10ns to drive the address to the RDRAM."
First, there is no "two-cycle" problem.
The necessity to provide 2-2-2 clocking
for correct internal operation of SDRAM provides perfect opportunity for a
good controller to drive all necessary addresses in advance.
There is no problem what-so-ever.
Everything matches perfectly.
Second, do you have any idea about real "quality" of the signals?
I am surprised that I came to this site by following a link from the Rambus wesite.
Although the article itself shows that Rambus may have some spiffy technology, there are quite a number of informed readers who don't seem to agree. What I find most interesting is that I have only sen one positive reply to the article. But it contained no facts and displayed hostility that anyone would even question that Intel + Rambus was not the fastest solution available.
I realize this is a late reply to this article and there is possibly a follow-up article already posted. But I am glad to see that Sanders is not shrinking from the task of proving proof for his claims.
Although I am interested in who actually has the better solution for real world applications. I find it doubtful that there is only one correct answer. In life there are many different situations, and so the advantages of both RDRAM and the variations of SDRAM could possibly be utilized in different applications.
Personally, I don't see RDRAM ever entering my home in the form of a PC (I do plan on purchasing a PlayStation2 which utilizes RDRAM). It is far too expensive of a solution for my wallet. And I much prefer a good price/performance ratio over "top of the line at any cost" when it comes to PC's.
I think Everyone knows that Hardware Central is a site of Intel Fanatics (you wont find much support/reporting for AMD or Alternate -NON INTEL technologies here). You article really missed the point and your analogies with Edo Ram is off. Apart from the other points raised you cant justify half the price of you new "$5000 TOp of the line model" to the cost of the RAM ESPECIALLY WHEN YOU HAVE ALTERNATIVES. The benefits are marginal at best, and hardly noticable - and I cant believe RAMBUS is telling everyone that a new PC similarly configured is only 160$ more with RIMM.
THE HIGHER PERFORMANCE IS ALWAYS THE TECHNOLOGIES THAT GIVES US GREATEST LEVEL CAPABILITES AT A GIVEN PRICE LEVEL, IN A PARTICULAR CATEGORY OF MACHINE. Somehow I dont think that RAMBUS will ever be there because of the "Inherent manufacturing complications" and Greedy licensing. It will probably go the way that DAT recorders and Minidisk went, with an honorable mention when the history books are written.
The Above Line summs up your article with the same name.
It is obvious that you were biased in writing the article. First, you opened the article siding with Intel and Rambus. Second, you were continually portraying how Intel was being opressed by AMD and VIA and the problems they had getting the i820 chipset to market. Third, you are damned Hypocrites: You critisize using theoretical data for marketting when you yourself only had theoretical data and no benchmarks to support your claims.
Sure Rambus looks nice on paper, but it runs hot as hell (as almost any review will tell you), and therefore hogs a lot of power. It's price is also insanely high. The performance of the RDRAM vs. SDRAM can not be compared completely objectively, since for the Different RAM types you need a different chipset.
I will admit, the benchmarks of systems using RDRAM are impressive, but they are also using Coppermine CPUs which have On-Die cache which makes up for a considerable performance boost.
SDRAM is proven and stable, and available in much larger amounts than RDRAM, I have seen boards with 16 DIMM slots, and I don't know if there is a limit to the amount of SDRAM you can get, but it is certainly way above the measly 2 slots that RDRAM has, and if you rembember as far back as before RDRAM was released, intel had to redo their motherboards using only 2 slots because with any more, RAMBUS doesn't work. It's an unstable architecture to start out with, and the fact that it has shown to be slower than than some SDRAM configurations is not surprising.
And on a closing note: If you want to make a biased article, at least put the effort into running some (sabotaged) benchmarks to prove your point instead of hypocritically using theorhetical data.
Conceptual is a very nice thing, but just like it happens on cars the bottom line is not how much horsepower a car has, but how much of it actually makes it to the tires, that's why we all take test drives before we buy one. In theory Rambus delivers, but i have yet to see it doing it on real life. Besides at that price the least I will expect is a two fold (if not a fourfold just like the price) performance increase.
Rambus Direct RDRAM, need I say more? On Monday May 8, 2000 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 had both Betamax and VHS recorders for a long time and you could visually see the difference in quality between the two. If you can sit me down at two comparable systems and I can easily see the visual difference in speed between the two, then I'll stand behind you 100%. Otherwise don't make such an outrageous comparison. Betamax was (and is) better, but it lost because of price, not performance or quality. That is why Betamax is still the standard in the film industry even after all these years.
You say that it is easy to debunk something. It is not easy to debunk it with benchmarks, as I have seen done. You say everyone is being negative about RDRAM, well I don't remember this kind of bad press for SDRAM. Sure everyone was skeptical of it, but it tested better virtually every time. It did well, the benchmarks proved it's performance levels, and it was never half as expensive as RDRAM. That's not counting the fact that SDRAM came about when smaller sizes of memory were the norm.
I have been going to Tom's Hardware for three years for good information. He does not just pooh on everything as you seem to imply. He and his staff thoroughly test every opinion they offer with extensive benchmarking, and let the benchmarks do the talking. If the product performs, it is praised. If it does not, then it is chastised. I realize that Tom does not have your computer science degree, but his staff on the other hand is quite educated. Anyone who can give complete electrical specs for building your own Athlon overclocking card, and EXPLAIN those specs, obviously has a LOT of technical know-how. I hope you are not saying that Tom and his staff have no technical expertise. Tom's site publishes a book on computer hardware for crying out loud. Show me your book.
You seem to know a lot about what you're talking about, but time after time other sites and posters here debunk you with methodical explanations. How do you justify that?
Wow Sanderson I am impressed. With your lack of comprehension. It isn't that anyone is anti-Intel it is we are anti-lies. RDRAM has been proven to be an overpriced underperforming memory. PERIOD. I don't care one bit about theory as theory won't make Quake run faster or Exchange run smoother. And in case you haven't noticed Intel has pretty much decided they *@*&ed up since the 815 board is now being rushed to production. Also no one will buy an RDRAM server, no way you are going to justify to an accountant a 10k server when for 4k you could have the same thing and one using proven technology (and faster). Not too mention the heat issue which is a big deal in a server enviroment. Our server room here at work is about 85F, add a ton of RDRAM and I bet it would go up another few degrees shortening the lifespan of each system. Also to claim that benchmarks prove nothing is pure idiocy! What theories do? Benchmarks are the only way to see anything, if SDRAM runs Q3 faster than it is better, if it runs Word 2k it is better, etc... How can this not be so?! Like the car example above, so you tell me that you have RDRAM with 300hp and I have SDRAM with only 200hp. Then you say "my car is faster than yours since it has 100hp more" my reply "wanna race?" yours "no that wouldn't prove anything, trust me my 100hp is better". Then I wrangle you into a race and you lose, everyone says "see SDRAM is better and it was cheaper too!" and you say "ya but mine has 100 more hp". SO WHAT YOU LOST. Just like RDRAM in real life...