Skip to main content
Message Font: Serif | Sans-Serif
 
No. of Recommendations: 50


Asking whether Rambus needs RDRAM might sound like asking whether 800 MHz is
faster than 266 MHz. That is, beyond reasonable dispute. However with the recent
success of Rambus in signing up 5 or 6 (depending on how you count) DRAM manufacturers to pay royalties on SDRAM and DDR, many have concluded that Rambus
needs RDRAM no more than Gloria Steinem needs a bicycle. A lucrative monopoly is at hand, with DDR royalties actually larger than the 1.5% believed typical for RDRAM.
I'm here to tell you that RDRAM does need to succeed for Rambus to reach full value. Yes, Rambus could and will be a cash machine no matter how the DRAM market goes. Rambus is often compared to Qualcomm, but Qualcomm does not own the patents to its rival, GSM. Rambus is in a sweet position, and should easily be worth the $200 price target Morgan Stanley Dean Winter's Mark Edelstone has assigned. If the DRAM market reaches anywhere near the ~$76 billion projected for 2002, and Rambus collects also royalties from the tens of billions of logic controllers that interface to memory (none have signed up as yet, except for RDRAM), even a very modest P/E would do it. Heck, a P/E of 10 would very likely do it.
But if RDRAM flops, Rambus might not be worth more than P/E=10. It would then
risk being seen as the parasite that the Micron-Hyundai-Infineon trio claim. Yes, I'm aware that these same exact trio in 1995 were eagerly pushing for the "evolutionary"burst-EDO instead of the Rambus-IP based SDRAM, making their current claims hollow (1). They fought hard against Intel's decision to introduce SDRAM, and Micron remains the biggest producer of the decrepit EDO technology. But if RDRAM fails, Rambus will have real trouble working with anybody and doing anything except collecting checks for SDRAM and DDR.
To see the difference between associating with success and associating with failure, look no farther than Rambus's experience with the Ninetendo64 and the PlayStation2. Primarily for reasons of marketing, the N64 was never a big hit. By contrast, the PS2 is already enormous in Japan, has been well-reviewed in America and Europe, and seems to be on the way to be a monster hit eclipsing even the original PS.
The consequences? Nintendo is going a different approach for their upcoming GameCube. Specifically, they are going with SRAM, which has certain advantages no
DRAM can equal. (But it cost more --- the GameCube will have less memory than the PS2,
yet the memory will cost far more. The cost of SRAM dwarves RDRAM).
Sony however is delighted with the PS2, and so, therefore, is the PS2 designer, Toshiba. Probably as a consequence Toshiba became the first licensee to sign up for QRSL (Quad
Rambus Signal Logic, a product even faster than RDRAM, probably to be implemented in SRAM). There are reasons to believe Toshiba is interested in QRSL for the next generation
of PlayStations, and possibly for the communications market. The point is, there is nothing like a satisfied customer to bring repeat business.
Similarly, elements of the Rambus packet-logic techniques have been implemented into a serializer/deserializer (Ser/des) device, which PMC-Sierra has licensed. If Rambus is seen as a winner, the efforts to market QRSL, and the Ser/des will both be greatly
enhanced. If RDRAM flops in the PC market, it becomes a much harder sale.
Fortunately there is reason to believe that the Pentium 4 with RDRAM will
be a hit, like the PS2, and not a relative disappointment like the N64. Of course
part of the reason lies with relative marketing power (Sony and Intel obviously having
more than Nintendo). The extremely high cost of the early AMD DDR machines
-- the 266 MHz DDR Atholon machines offered by MicronPC hit nearly $3000 before
monitor, and that with a mere 256 MB of memory -- is also promising. If RDRAM
proves to be more cost effective in terms of total system design (owing, for example,
to the lower pin count and greater ease of motherboard design) as well as faster,
Rambus's star, and stock, will rise sharply. Via and Acer are working toward lower costs DDR systems, but thus far they have simply ignored the difficulties in driving motherboard traces at radio frequencies, and have not produced a product which works.
Although the early plethora of P4 benchmarks will likely confuse many, and provide ammunition for any viewpoint desirable, these benchmarks are hardly created equal. Some test disk speeds, some test video cards, most test whether software has been written to run well with a particular processor. For example, the P4 runs 3 times faster than the 1.2 GHz Atholon on some programs (2), but much slower on others. Because it is newly released, on the whole less software is optimized for the P4. Nonetheless, it is already perfectly possible to tell just how fast the P4 with RDRAM is. It is much faster than the Atholon, even on a per clock basis. For example, on the SiSoft Sandra benchmarks of effective memory bandwidth, the P4/RDRAM system scores TWO TO THREE TIMES the level of DDR/266 MHz. Note that this software is designed to include effects just as latency, and get an effective memory availability number. RDRAM simply annihilates DDR (3). Bulletin board arguments can exist even on whether Apollo ever reached the moon; but no serious hardware designer seeing these numbers will wonder whether 800 MHz is faster than 266 MHz.
Likewise, the SPEC*fp*2000 score of the P4 with RDRAM, 558, is faster than a 1.2 GHz Atholon ON A PER CLOCK BASIS (about 46% faster with a 25% increase in clock speed). This benchmark tests a large variety of floating point computations mostly drawn from the field of scientific computation. However it has, by design, no dependency on software optimization. Thus the P4 (i850 with RDRAM) is so much faster than the Atholon that the later becomes irrelevant. It may take time for optimized software to be introduced, but the end is clear: the Atholon is not remotely competitive. Given that the AMD speed road map looks like a deceleration ramp, that situation is unlikely to change soon. The sole remaining issue in whether RDRAM wins in the market is whether Via can introduce a P4 with DDR that does not carry the same exorbitant costs as the 266 MHz Atholon does. By which, of course, I mean a DDR chipset which works, unlike the current Acer and Via DDR systems. Since PC600 RDRAM is both cheaper and faster than 266 MHz DDR, it does not look promising for Via either.
Ultimately, Rambus is trying to position itself as the means for ultrafast chip-to-chip
communication. If they succeed in doing so, Rambus could indeed be one of the most
profitable companies in the world. Their chances of doing so will rise sharply if the Pentium 4 WITH RDRAM wins in the market place. Then Intel would feel much more comfortable about turning to Rambus in the future. QSRL and the Rambus Ser/des will have a big boost in producing new significant revenue streams.
If your goal in owning Rambus at this point (price around $60) is to double or even triple your money in two years, you will probably succeed so long as Rambus can uphold its patents. If your hope is that Rambus follows the path of a Dell, Cisco, or Microsoft to become a major player in the technology industry, with a market cap to match, hope fervently that RDRAM takes a big chunk of the DRAM market over the next two to three years. There won't be any 10 baggers for Rambus in just collecting royalties on SDRAM.


Links:

(1) http://www.techweb.com/se/directlink.cgi?EET19950116S0018
(2) http://www.sharkyextreme.com/hardware/guides/pentium4/20.shtml

(3) http://www.sharkyextreme.com/hardware/guides/pentium4/8.shtml





Print the post Back To Top
No. of Recommendations: 2
Hey Patrick,

I can not take you seriously until you teach your word processor to spell Athlon (sic Atholon). Makes me think of a friend of ours who had a problem with Inquest (Inquist). LOL Thanks for the post.
Print the post Back To Top
No. of Recommendations: 1
I see now that you are right, RDRAM is the future of computing. The obvious performance increases in unoptimized benchmarks like SPECfp just show that RDRAM is the obvious choice for workstation environments. With the future going for RDRAM, there is no need to have DDR or SDRAM, as no one will use it 2 to 3 years from now.

I will go into more detail in my next post on this thread.

--Sboy
Print the post Back To Top
No. of Recommendations: 2
Basically, that last post was to prove how easy it is to get recs on this board. I bet I will get at least one, despite the entire post being found on falsehoods.

I posted about 3 months ago that RMBS needs the DDR and SDRAM claims to be upheld, and that is more obvious today than it has ever been in the past. P4 was built completely around RDRAM and performs well in optimized benchmarks. Just for the record SPECfp is P4 optimized if compiled with P4 optimized compilers, which Intel did. On workstation apps with non-optimized software, the Athlon generally performs better than P4.

There is no doubt that RDRAM is the bandwidth champion. Sandra and Linpack (large matrices) show this. But, as everyone has been saying for the past 6 months, bandwidth =/= (does not equal) performance. 800 Mhz is more than 266 Mhz and that's very nice. 1500 Mhz is also more than 1200 Mhz, but that doesn't mean it is a better product or better performer.

All publications have stated that the market is refusing to embrace RDRAM as a technology. There are currently only 5 motherboards using i850 under developement and around 15 using AMD760, which are of similar age. When Nvidia's and Ali's and Micron's chipsets come out (all supporting DDR), there will be a lot more options for DDR than RDRAM.

I am not disputing that RMBS may be a good investment. All companies may be a good investment if they find a golden goose. My only statement is that there are a lot of companies searching for that goose, and so far, RDRAM has promised gold and delivered pyrite. It probably has delivered gold (increased stock price) for a lot of you, but for the future, I don't know if it is foolish or Foolish. I just don't see the company maintaining the level of speculation and outrageous stock price without delivering a product that they legitimately own the rights to. Which comes back to the beginning, if RMBS owns DDR and SDRAM, they will be a profitable company, but I'm not betting my money on that because **I don't know**. Do you???

--Sboy
Print the post Back To Top
No. of Recommendations: 0
I see now that you are right, RDRAM is the future of computing. The obvious performance increases in unoptimized benchmarks like SPECfp just show that
RDRAM is the obvious choice for workstation environments. With the future going for RDRAM, there is no need to have DDR or SDRAM, as no one will use it 2
to 3 years from now.

I will go into more detail in my next post on this thread.

--Sboy


What changed your mind? Or is this some rare form of sanity breaking out in unexpected places?

Good luck on your mu and amd "investments".

BTW just how much RMBS do you own now?

Eric
Print the post Back To Top
No. of Recommendations: 13
Basically, that last post was to prove how easy it is to get recs on this board. I bet I will get at least one, despite the entire post being found on falsehoods.

Sboy,

I guess it's not as easy as you thought. So far no recs, but I'm sure some Rambus haters will go back and give you some. Speaking of falsehoods...

All publications have stated that the market is refusing to embrace RDRAM as a technology. There are currently only 5 motherboards using i850 under developement and around 15 using AMD760, which are of similar age. When Nvidia's and Ali's and Micron's chipsets come out (all supporting DDR), there will be a lot more options for DDR than RDRAM.

There is a big difference between talking the talk and walking the walk. While it's true that Team DDR and the hacks in the media love to talk about how great DDR the fact is that so far no one wants to sell it. Despite the fact that dozens of companies are singing the praises of their DDR chipsets and motherboards the only box maker selling them is Micron. Micron! The MM that is pushing DDR hardest of all. Beyond that they are already in a lawsuit with Rambus and have nothing to lose and everything to gain by selling DDR systems.

Now this is what I call walking the walk.

http://www.intel.com/pentium4/launch/indsupport.htm?iid=P4PLaunchFlash+021&

Look at how many people are selling P4s with RDRAM on the first day it's out. Caliber, Compac, Dell, Hewlett-Packard, IBM, Powerspec PC, NEC, and Systemax. I just went to Gateways site and they're for sale there too. I thought Gateway was a big AMD supporter? Why no DDR systems? With all the support DDR is supposedly getting you would think Gateway would be on board. Before you reply I already know the answer. "Just wait another two or three months and then you'll see what DDR can do."

Either way Rambus wins. Show me the money.

Jackson





Print the post Back To Top
No. of Recommendations: 0
gwj1

"there will be a lot more options for DDR than RDRAM."

This is just another way of saying that there will be a lot of little fish trying to swim in a very small pond.

More fish don't make the pond bigger. Neither are a lot of little fish likely to eat the shark's dinner.

I'm betting the Shark has the fish on its dinner plans. What do you think?

;?)

DougK99

Print the post Back To Top
No. of Recommendations: 0
qq{ Asking whether Rambus needs RDRAM might sound like asking whether 800 MHz is faster than 266 MHz. }

Just wanted to point out that 800MHz RDRAM has the same bandwidth as 200MHz DDR. This is due to the fact that RDRAM is only 16-bits wide, whereas DDR is 64 bits wide. This is also why RDRAM's latency is 4x higher than SDRAM's.

MHz is only half of the speed equation.

Dave
Print the post Back To Top
No. of Recommendations: 1
Just wanted to point out that 800MHz RDRAM has the same bandwidth as 200MHz DDR. This is due to the fact that RDRAM is only 16-bits wide, whereas DDR is 64 bits wide. This is also why RDRAM's latency is 4x higher than SDRAM's.

MHz is only half of the speed equation.

Dave


Wow, latency is now 4x as bad as Sdram. Its seems to keep increasing. Stuff gets hot too. Damn expensive.
I think most peoples ideas on latency are off quite a bit, heat also, price too

Bandwidth is a mile wide. The door is closing, can you feel it...Pete/Goaltender95
Print the post Back To Top
No. of Recommendations: 2
That was an interesting opinion ptnewell. The problem I have is that for the next 6 to 8 weeks it is just an opinion. Until DDR motherboards are out and for sale to the general public and the early adopters have found all of the problems we will not have a good look at the two technologies. Right now looking at many sites shows the Athalon 1.2 Gig (running DDR) is at about par to the P4 at 1.5 Gig. This means that the P4 wins some benchmarks and the Athalon wins some benchmarks. I don't want to make this into an AMD vs. Intel post but there is no P4 DDR motherboard out right now so the only comparison available is AMD vs. P4. This means that a true DDR vs. RDRAM comparison is not going to happen until ??

So where does this leave Rambus? If in court Rambus' patents are thrown out for SDRam and DDRram Rambus will depend on DRDRam for all income. I will leave it up to everyone to go read up on the lawsuits and counter suits but my opinion is that Rambus will lose the claimed rights to SDRam and DDRRam. Then DDRRam will become the standard due to lower prices and performance within 1% of DRDRam. In which case Rambus will be in a world of hurt.
Print the post Back To Top
No. of Recommendations: 0
Just wanted to point out that 800MHz RDRAM has the same bandwidth as 200MHz DDR. This is due to the fact that RDRAM is only 16-bits wide, whereas DDR is 64 bits wide. This is also why RDRAM's latency is 4x higher than SDRAM's.

Now, I'm NOT a Rambus support (personally I think they will fail, at least short run < 1-2 yrs) but this is absurdly wrong. By your own argument you prove that RDRAM can deliver 64-bits in four clocks (or one clock of DDR at 200 Mhz). In actuality, for BOTH memory systems, NO data is delivered until 64-bits have been collected. This would be accomplished in the same amount of time if you look at ONLY these very few pieces.

In reality, things are MUCH more complicated than this and your post merely flaunts your ignorance in a manner akin to standing in the middle of Times Square, butt-naked, yodeling, and waving a giant hunter's orange flag that says "Look at me!".

Michael Mathers
Print the post Back To Top
No. of Recommendations: 0
In reality, things are MUCH more complicated than this and your post merely flaunts your ignorance in a manner akin to standing in the middle of Times Square, butt-naked, yodeling, and waving a giant hunter's orange flag that says "Look at me!".

BTW will that be allowed after Gore gets in :)

Cheers
Cor
Print the post Back To Top
No. of Recommendations: 1
qq{NO data is delivered until 64-bits have been collected. This would be accomplished in the same amount of time if you look at ONLY these very few pieces. }

Please show me where I stated that memory would be delivered piecemeal? RDRAM memory transfers are only 16bits wide, so obviously 4 transfers would have to take place before the memory controller would have the requisite 64bits to deliver to the CPU. Nothing I stated denies this requirement.

The aggregate bandwith for PC800 RDRAM and DDR1600 is *identical*. In terms of memory clock cycles, the latency of RDRAM is 4x that of SDRAM. Timewise it's not 4x, but is still higher than that of DDR due to the overhead involved in de-multiplexing.

Dave
Print the post Back To Top
No. of Recommendations: 1
The aggregate bandwith for PC800 RDRAM and DDR1600 is *identical*. In terms of memory clock cycles, the latency of RDRAM is 4x that of SDRAM. Timewise it's not 4x, but is still higher than that of DDR due to the overhead involved in de-multiplexing.

This is so completely wrong. Let's use some numbers to illustrate. Let's say one SDRAM clock = 10 ns. RDRAM would spit out data to the bus ever 2.5 ns (this included the demultiplex operation, of which there are four). 2.5 * 4 = 10


-Mike
Print the post Back To Top
No. of Recommendations: 2
"The aggregate bandwith for PC800 RDRAM and DDR1600 is *identical*. In terms of memory clock cycles, the latency of RDRAM is 4x that of SDRAM. Timewise it's not 4x, but is still higher than that of DDR due to the overhead involved in de-multiplexing.

Let's dispell the myth right now that RDRAM has a higher latency (component or system) that SDRAM.

Please reference the white paper from Samsung:
http://www.usa.samsungsemi.com/new/rdram-whitepaper.htm

Note that the RDRAM's timing granularty advantages give a far superior component latency that SDRAM (BTW, poor coarse timing granularity is a significant detractor from DDR) and overall system latency is lower due to the inadequate system bus generally partnered with SDRAM.

Interestingly, the latency advantages are even more apparent in SERDES applications. See below white paper for advantages of RDRAM in SERDES applications. Note discussion on effective bandwidth advantages due to increased number of memory cores:
http://www.rambus.com/developer/downloads/OC192_App_Note.pdf

Another 11% Nice day of performance.

- Matt
Print the post Back To Top
No. of Recommendations: 1
qq{ This is so completely wrong. Let's use some numbers to illustrate. Let's say one SDRAM clock = 10 ns. RDRAM would spit out data to the bus ever 2.5 ns (this included the demultiplex operation, of which there are four). 2.5 * 4 = 10 }

RDRAM would spit out data to the memory controller every 2.5 ns. The data would not go out over the bus to the CPU until a full 64bits was available. So it would take 4 clock cycles before enough data was available to be sent to the CPU. In contrast, the data from SDRAM would be delivered in a single clock cycle because the SDRAM path to the memory controller is 64bits wide. Thus *in terms of clock cycles* RDRAM's latency is 4x higher. You are correct that RDRAM's latency in per unit of time is not 4x higher than SDRAM, however that does not invalidate my argument, because you are using a different unit of measurement.

Dave
Print the post Back To Top
No. of Recommendations: 2
"Thus *in terms of clock cycles* RDRAM's latency is 4x higher. You are correct that RDRAM's latency in per unit of time is not 4x higher than SDRAM, however that does not invalidate my argument, because you are using a different unit of measurement."

And you are using a different unit of measure in your comparison between SDRAM and RDRAM. A clock cycle isn't a "standard unit of measure" and is only meaningful when either frequency or period (duration of cycle) are known.

Your contrived statement is misleading because what you call "your unit of measure" is a variable rather than A STANDARD UNIT.

RB

PS - If you really need further illustration:
You're looking to move 1000 cars across the ocean. Two companies can lease boats capable of making the journey in 10 days.
Company A can lease you 25 boats = 25 boat loads
Company B can lease 13 boats = 13 boat loads

Which is the better choice for your needs?

I guess it depends on the boat load!
Print the post Back To Top
No. of Recommendations: 1
You are correct that RDRAM's latency in per unit of time is not 4x higher than SDRAM, however that does not invalidate my argument, because you are using a different unit of measurement.

This is the silliest thing I have ever heard. Sure, you're right. The number of clocks is four times higher. Big deal. It's the same amount of TIME! Latency is a timing measurement. It is often measured in clock cycles, yes, but the important part about latency is TIME.

Sheesh!

Michael Mathers
Print the post Back To Top
No. of Recommendations: 0
qq{ This is the silliest thing I have ever heard. Sure, you're right. The number of clocks is four times higher. Big deal. It's the same amount of TIME! Latency is a timing measurement. It is often measured in clock cycles, yes, but the important part about latency is TIME. }

I realize that TIME is the important factor, however in my original statement, when I stated that the latency of RDRAM was 4x that of SDRAM I was *NOT* talking about time. Other people made the assumption that I was saying that RDRAM took 4x as much TIME which is not what I said. One of RDRAM's advantages over SDRAM is that it can run at a higher frequency. However many people overlook the fact that this advantage is also a requirement because RDRAM isn't as "efficient" as SDRAM on a per clock cycle basis.

Most people simply see 800MHz RDRAM, 200MHz DDR, and think RDRAM is 4x faster than DDR, which it is not. I don't know why so many people were so up in arms over what I said. I was just trying to point out that although RDRAM may be running at 4x the clock speed, it's overall performance is *NOT* 4x that of DDR SDRAM.

Dave
Print the post Back To Top
No. of Recommendations: 3
I realize that TIME is the important factor, however in my original statement, when I stated that the latency of RDRAM was 4x that of SDRAM I was *NOT* talking about time.

.. and hence you weren't talking about latency either. Latency is defined as the time required for the memory to respond with data. You can measure a 1 liter container of water in ounces too. That won't mean you have 33.8x the amount of water (33.8 ounces/liter)! Nor will it mean you have STOPPED talking about the AMOUNT of water you are measuring, regardless as to what unit you use.

Other people made the assumption that I was saying that RDRAM took 4x as much TIME which is not what I said.

Well, that may be what you MEANT to say but it is NOT what you said. You talked of latency. Translated: You talked of time. That is what latency IS. TIME.

This is also why RDRAM's latency is 4x higher than SDRAM's.

See.. This is what you said.

Michael Mathers
Print the post Back To Top