A reader responds to comments about RDRAM in the Northwood P4 article
Page 1 of 1
This page contains a reader's response to comments I made about Rambus RAM
in the Intel Northwood Pentium 4 CPU article. John S. is an engineer who does
system-level, rather than IC design. He lives in Florida, and has no connection
with Rambus, other than owning their stock.
His reason for responding was simple frustration. He says, "I hate seeing conniving
organizations roll over the little guy and marketing inferior versions of their ideas
just to avoid paying their due."
Admittedly, I haven't researched Rambus RAM very much. All I know is that it
provides performance comparable to that of DDR RAM, yet costs ~4X more per
MB. This reader makes compelling points, and includes a brief history of RAM
"The problem with Rambus DRAM is
that it's significantly more expen$ive
than DDR RAM, without providing a correspond increase in performance.
It's rumored that Intel cut a deal with Rambus, which awarded Intel mega-
buck$ if they helped Rambus sell a crapload their RAM."
Rambus RAM is more
expensive than DDR RAM, but prices are dropping almost
daily. Rambus is to SDRAM (including DDR) what SDRAM was to DRAM (including
EDO RAM). A little background info:
When asynchronous Dynamic Random Access Memory (DRAM) became too
slow for growing processor speeds, Synchronous DRAM (SDRAM) was created.
DRAM makers used various tricks to squeeze out more performance, but in the
end, Extended Data Output DRAM (EDO) max'ed out asynchronous memory
performance at ~66MHz.
In the beginning, SDRAM wasn't much faster than EDO RAM, because they were
both running at 66MHz. But SDRAM was at the beginning of its design envelope.
Before long, SDRAM went to 100MHz, and then to 133MHz. There is no longer
By using a trick borrowed from Rambus [heck, SDRAM itself is a trick borrowed,
or stole, from Rambus, but that's another subject entirely. See the end of this
page for more insights along these lines] they managed to push the effective
clock-rate to 266MHz. But now SDRAM is against a wall, making further speed
increases difficult, if not impossible.
Enter Rambus. Right out of the block, RDRAM runs equally as fast as DDR. And
it's just getting started. Rambus has already announced plans for the next few
years to crank up clock speeds and quadruple bus width to yield 64-bit with a
throughput of 9.6GB/s.
And that was before they announced their new Yellowstone technology, which,
even with current 16-bit widths and clock speeds, is capable of doubling the
data throughput rate!
As for Intel, Rambus does not pay Intel to use RDRAM. Rather Intel pays
Rambus for the right to use its technology. In fact, Intel recently expanded
their license with Rambus to include more of their technology & extended it
to 5 years.
"I haven't researched these rumors enuf to know if they're true, but I know that
there are a lot of people who don't like Rambus. It's a matter of principle with
them *not* to buy Rambus DRAM."
Most people who don't like Rambus subscribe to the propaganda being
disseminated by memory manufacturers. They don't like selling Rambus
memory because they must pay a royalty on each unit they sell that uses
You see, Rambus doesn't manufacture memory themselves. They merely license
the technology to others (manufacturers).
In the past, new techniques for improved memory performance were designed
and patented by the memory makers themselves. Since most companies were
responsible for one innovation or another, no one paid any royalties, because
[Cross-licensing is like: I invent a new suspension, you invent a new steering
system. We agree not to pay each other, & both sell a better handling car.]
But Rambus doesn't manufacturer memory. The company merely consists of
very smart engineers, who design memory. If they don't charge a price for
others to use their design, they have no way to profit from their efforts.
Perhaps you can see how it might be objectionable for memory manufacturers
to pay royalties when they've never had to do this before. So they badmouth
Rambus, promote DDR, and pray the industry will embrace DDR long enough
for them to come up with a design that gets around Rambus' patents.
"The price of Rambus DRAM has dropped dramatically over the last few months,
but it's still significantly more expensive than DDR RAM, which performs
comparatively, and perhaps even better (lower latency)."
Lest you think that the royalty they charge is responsible for the higher cost,
Rambus receives, on average, less than $3 for every $100 worth of RDRAM
sold. And the fee-per-part goes down with the price.
RDRAM prices are still more expensive than DDR, but continue to fall .. for the
very same reasons that the first P4's sold for close to $800. Current, state-of-
the-art, 2GHz P4s, less than 1 month old, sell for half that much now.
New designs like RDRAM are *very* expensive to implement (think Billions).
Tweaks to existing designs are relatively cheap (say, tens of millions to yield
a faster version of an existing part).
That extra expense has to be recaptured somewhere. If volume continues to
rise for RDRAM, in a few more months, they will be cost-competitive with DDR,
royalties and all.
As for performance, similar data rates yield similar results. But because of the
different interfaces involved, results do vary. DDR does appear to be marginally
better at "traditional" business applications (such as word-processing).
Games and audio are pretty much a dead heat. Compute-intensive and linear,
repetitive operations like video compression favor RDRAM, sometimes by a wide
"The future is a Northwood Pentium 4 with DDR SDRAM and that's where you're
money is invested wisely for the long run. RDRAM systems, if they still exist after
Intel's current contract with Rambus expires, will do better with MPEG content
and video .. "
It should be noted that this ZD article was written *before* Intel announced
the expansion and extension of their contract with Rambus.
One last thing...
"...value is not my #1 consideration..." "...I'm far more interested in system
stability than saving a hundred bucks..."
"'Another factor is the stability and product quality of a system: while all Athlon
processors suffered from occasional instability in our tests, the Pentium 4
platform ran without a glitch.'"
I would note that *all* Athlon processors were tested with DDR-SDRAM memory
while *all* Pentium 4 machines were running RDRAM! Coincidence?
If you wanted to drive out into the middle of nowhere, hundreds of miles from
any help, would you rather go in a car with an engine tweaked for every last
ounce of performance? .. one that had to crank the engine to 6000 RPM, just
to do 60 MPH?
Or would you rather go in one that has an engine with power to spare? ..
one that does 60MPH @2500RPM, while asking, no begging you to do 100?
Yeah, me too. =)
Enjoy your new machine.
Note: Stock prices can sometimes tell us something about how well a company's
products are doing in the industry. Here is a 5-year graph of Rambus' stock price.
After climbing like a rocket early last year, it's gone from a high of ~US$120/share
in the middle of last year, to its current price of ~$8 per share, losing over 90% of
its market value in the process. It bottomed out in late August at ~$5 per share.
To be fair, I've included comparisons with two of the world's largest memory makers.
The first is Micron. Here is a 5-year graph comparing Rambus to Micron. The second
is Infineon. Here is a 5-year graph comparing Rambus to Infineon. Both Micron and
Infineon look similar to Rambus.
So the point could be made that the problem isn't with Rambus, but rather the whole
industry. John says, "The fall of RMBS share price is due entirely to SDRAM! Sales of
RDRAM are up this year.'
This section includes John's follow-up comments. In addition to comparing RMBS
stock price with that of two other major RAM manufacturers, he notes:
article contains many of the reasons why I say that SDRAM stole
Rambus. Many of the industry's biggest players conspired to find a device that
would be faster than DRAM, but without paying Rambus.
For several years they got away with it because Rambus had applied for, but not
yet been issued, a patent for their technology. Once the patent was issued, Rambus
approached the manufacturers and asked them to sign royalty agreements on both
SDRAM and DDR-SDRAM.
Rambus had warned the industry they intended to do, back when they first converged
on SDRAM as a standard (before patents were awarded). All but three manufacturers
agreed. Infineon, Micron & Hyundai decided to fight.
It was these royalty agreements for SDRAM and DDR, coupled with the
high prices for memory [64MB of SDRAM used to be US$100+, remember?] and the
expectation of even higher demand that caused the price of RMBS to soar last year.
The value (real or perceived) of RDRAM had nothing to do with the stock's run up.
At that point in time RDRAM was less than 1% of the memory market and a tiny
fraction of Rambus' revenue. Rambus is more than RDRAM.
Now let's jump ahead a few months. The vision of a rapidly-growing memory market
is proven wrong. It's been replaced by the pessimism that no one will ever buy
anything electronic again. Demand for memory is low. Prices have dropped so low
that memory is being sold for less than what it costs to manufacture.
Wall Street throws out the hi-tech baby with the bathwater. It's convinced that
no recovery will come in this lifetime. The NASDAQ drops to ~1600 from over 5000,
a loss of ~70%!
Intel's stock (no one questions their market acceptance) drops 75%. Most memory
makers lose 80% or more. Memory prices themselves have dropped more than 80%.
Recall that lower sales price means lower royalties for Rambus.
In this market environment, add an initial loss in the ongoing legal battle between
Rambus and the three holdout companies. What do you get? That's right: a 90% drop
in share price. All because of SDRAM!
This is why I say that the current RMBS
stock price is a poor
reflection of how
RDRAM is being received by the industry. Revisit this issue a year from now and
I think you'll get a much better picture.
As for the issue of reliability .. with regard to the system you're planning to
you should know that the reliability and interoperability issues with DDR are real.
That means, if your new system comes with Micron DDR and, at some future point,
you want to add more, you may have to make sure you get Micron DDR to expand.
This is not an issue for RDRAM. This is IN ADDITION to the general reliability issues
experienced by the folks at Tom's.
As for the price difference, I agree - it's
tough to swallow. But the price of RDRAM
has been dropping steadily. Who knows how far it will fall by the time Northwood
Furthermore, it's possible that, when the 845 DDR chipset comes out, the added
demand caused by all the major manufacturers trying to put DDR systems in their
lineup may create more of a demand than current DDR manufacturing can meet,
causing DDR prices to go up.
I suspect that, by the time your ready to build your new system, the premium for
RDRAM will be much less.
I am an engineer with a fair amount of experience in the field. I do not now, nor
have I in the past even worked for Rambus. I live in Florida, like hurricanes better
than earthquakes and do system level rather than IC design, so I'm more likely to
work for IBM (and have) designing systems that use RDRAM and P4's than to design
the chips themselves. I do own stock in Rambus, but I also have shares of Micron,
their biggest competitor.
My reason for responding to your initial article was simply frustration.
Farmwald had a great idea and Mark Horrowitz did a fabulous job working with
him to come up with a revolutionary new concept in memory design, offering
not only higher speeds, but also greater reliability.
I hate seeing conniving organizations rolling over little guys and marketing inferior
versions of their ideas just to avoid paying that little guy his due.
And now I'll step down off my soap box.
[Intel Northwood Pentium 4 CPU article]