alt.certification.a-plus - why was rdram used in p3 computers??

Author

why was rdram used in p3 computers??

james yi

2003-01-25, 2:23 pm

ok now another follow up question. why did the p3's use rdram memory which
ran faster than pc133 sdram?? if the fastest fsb for the p3 was 133mhz,
shouldn't the pc133 have been sufficient since it was in sync with the bus??
the intel 820 chipset used rdram.

VernonOJ

2003-01-25, 8:23 pm

"james yi" <jamesyi@worldnet.att.net> wrote in message
news:9ABY9.13830$zF6.1043097@bgtnsc04-news.ops.worldnet.att.net...
> ok now another follow up question. why did the p3's use rdram memory which
> ran faster than pc133 sdram?? if the fastest fsb for the p3 was 133mhz,
> shouldn't the pc133 have been sufficient since it was in sync with the
bus??
> the intel 820 chipset used rdram.

Hype, pure hype.

Barry Watzman

2003-01-25, 9:23 pm

The P3 with RDRAM was a mistake, not withstanding that RDRAM is, in my
view [others will argue - vocally] the best memory type available.
However, it offered no benefit for the P3, because PC100 SDRAM could
fully meet the bandwidth requirements of the CPU.

Also, note that at the time of the introduction of RDRAM for the P3,
there was not yet any PC133 SDRAM, it was just PC100. PC133 came later,
really at the end of life of the P3, when the P3's with clocks above 1
Ghz began to appear. However, when RDRAM was introduced for the P3, the
fastest P3's were still in the 800 MHz range (of course the faster ones
came out later).

RDRAM for the P3 was a "pump priming" exercise for Intel and, Intel
thought, the memory makers.

Unfortunately, not only was it not a pump priming exercise, it was a
disaster due to Intel's screw-ups. Don't get me wrong, there was
NOTHING technically wrong with the 820 chipset or RDRAM, and, in fact,
the RDRAM P3's were the highest performing processors .... but 128 meg
of RDRAM cost $1,000 (yes, a GRAND), and the performance benefit was
like 1% to 3%.

Meanwhile, however, because of the cost, there was a string of
catastrophic disasters. The worst of these was the 820MTH chip, a chip
that was supposed to allow the 820 chipset (RDRAM) to work with SDRAM.
The 820MTH was simply not reliable, and in the end Intel had to recall
every 820 motherboard that they had made with the MTH on it (using
SDRAM) and replace it with an 820 motherboard without the MTH (using
Rambus RDRAM). Note that the motherboards that were recalled were SDRAM
motherboards, not RDRAM motherboards. But they had an RDRAM chipset on
them (and an RDRAM to SDRAM converter chip that didn't work). There was
NO problem with RDRAM or the 820 chipset, the problem was the MTH and
SDRAM, but it was RDRAM -- Rambus memory -- that took the "PR Hit" and
that got the black eye.

It may well be that Intel was so soured by this experience (it cost them
hundreds of millions of dollars), and some others at the time related to
RDRAM -- but mostly of Intel's making -- that they backed off of RDRAM
for the P4. Which is most unfortunate, because with the P4, RDRAM
provides a real benefit in terms of performance and stability /
reliability, a fact which most people, having used only SDRAM or DDR, do
not fully appreciate (although RDRAM does win virtually all of the
benchmarks). And the price premium for RDRAM is vastly less than most
people perceive, in fact for most of the last month Dell has been
selling RDRAM based 8250 series computers for LESS than [inferior]
DDR-based models.

james yi wrote:
> ok now another follow up question. why did the p3's use rdram memory which
> ran faster than pc133 sdram?? if the fastest fsb for the p3 was 133mhz,
> shouldn't the pc133 have been sufficient since it was in sync with the bus??
> the intel 820 chipset used rdram.
>
>

james yi

2003-01-25, 11:23 pm

=)
nice long descriptive answer.
but wouldn't performance actually decrease if the memory was running faster
than the fsb speed?? rather than a 1%-3% performance benefit? more like a
1%-3% performace penalty.

"Barry Watzman" <Watzman@neo.rr.com> wrote in message
news:3E334BA3.10806@neo.rr.com...
> The P3 with RDRAM was a mistake, not withstanding that RDRAM is, in my
> view [others will argue - vocally] the best memory type available.
> However, it offered no benefit for the P3, because PC100 SDRAM could
> fully meet the bandwidth requirements of the CPU.
>
> Also, note that at the time of the introduction of RDRAM for the P3,
> there was not yet any PC133 SDRAM, it was just PC100. PC133 came later,
> really at the end of life of the P3, when the P3's with clocks above 1
> Ghz began to appear. However, when RDRAM was introduced for the P3, the
> fastest P3's were still in the 800 MHz range (of course the faster ones
> came out later).
>
> RDRAM for the P3 was a "pump priming" exercise for Intel and, Intel
> thought, the memory makers.
>
> Unfortunately, not only was it not a pump priming exercise, it was a
> disaster due to Intel's screw-ups. Don't get me wrong, there was
> NOTHING technically wrong with the 820 chipset or RDRAM, and, in fact,
> the RDRAM P3's were the highest performing processors .... but 128 meg
> of RDRAM cost $1,000 (yes, a GRAND), and the performance benefit was
> like 1% to 3%.
>
> Meanwhile, however, because of the cost, there was a string of
> catastrophic disasters. The worst of these was the 820MTH chip, a chip
> that was supposed to allow the 820 chipset (RDRAM) to work with SDRAM.
> The 820MTH was simply not reliable, and in the end Intel had to recall
> every 820 motherboard that they had made with the MTH on it (using
> SDRAM) and replace it with an 820 motherboard without the MTH (using
> Rambus RDRAM). Note that the motherboards that were recalled were SDRAM
> motherboards, not RDRAM motherboards. But they had an RDRAM chipset on
> them (and an RDRAM to SDRAM converter chip that didn't work). There was
> NO problem with RDRAM or the 820 chipset, the problem was the MTH and
> SDRAM, but it was RDRAM -- Rambus memory -- that took the "PR Hit" and
> that got the black eye.
>
> It may well be that Intel was so soured by this experience (it cost them
> hundreds of millions of dollars), and some others at the time related to
> RDRAM -- but mostly of Intel's making -- that they backed off of RDRAM
> for the P4. Which is most unfortunate, because with the P4, RDRAM
> provides a real benefit in terms of performance and stability /
> reliability, a fact which most people, having used only SDRAM or DDR, do
> not fully appreciate (although RDRAM does win virtually all of the
> benchmarks). And the price premium for RDRAM is vastly less than most
> people perceive, in fact for most of the last month Dell has been
> selling RDRAM based 8250 series computers for LESS than [inferior]
> DDR-based models.
>
>
> james yi wrote:
> > ok now another follow up question. why did the p3's use rdram memory
which

> > ran faster than pc133 sdram?? if the fastest fsb for the p3 was 133mhz,
> > shouldn't the pc133 have been sufficient since it was in sync with the
bus??
> > the intel 820 chipset used rdram.
> >
> >
>

klassick

2003-01-26, 12:23 pm

"Barry Watzman" <Watzman@neo.rr.com> wrote in message
news:3E334BA3.10806@neo.rr.com...
>snip<

Wasn't there a legal battle over patent infringement also?? I seem to
remember some shaky business practices from Rambus' end.

Barry Watzman

2003-01-26, 7:23 pm

No, not necessarily -- and the RDRAM may well have operated
synchronously, they were not using PC800 RDRAM, they were using PC600
and PC711. In any case, if you go back to benchmarks from the time (I
think that this came out in November of 1999), they WERE the fastest
sysems availalble and they slightly beat SDRAM systems with the same CPU
clock, but the difference was insignificant, and the cost penalty was
astronomical.

Remember, any given chipset is either RDRAM or SDRAM (or DDR), so you
are comparing systems with different chipsets. It's hard to draw exact
conclusions as to what aspect of the system(s) is responsible for
whatever performance differences are seen.

james yi wrote:
> =)
> nice long descriptive answer.
> but wouldn't performance actually decrease if the memory was running faster
> than the fsb speed?? rather than a 1%-3% performance benefit? more like a
> 1%-3% performace penalty.
>
> "Barry Watzman" <Watzman@neo.rr.com> wrote in message
> news:3E334BA3.10806@neo.rr.com...
>
>>The P3 with RDRAM was a mistake, not withstanding that RDRAM is, in my
>>view [others will argue - vocally] the best memory type available.
>>However, it offered no benefit for the P3, because PC100 SDRAM could
>>fully meet the bandwidth requirements of the CPU.
>>
>>Also, note that at the time of the introduction of RDRAM for the P3,
>>there was not yet any PC133 SDRAM, it was just PC100. PC133 came later,
>>really at the end of life of the P3, when the P3's with clocks above 1
>>Ghz began to appear. However, when RDRAM was introduced for the P3, the
>>fastest P3's were still in the 800 MHz range (of course the faster ones
>>came out later).
>>
>>RDRAM for the P3 was a "pump priming" exercise for Intel and, Intel
>>thought, the memory makers.
>>
>>Unfortunately, not only was it not a pump priming exercise, it was a
>>disaster due to Intel's screw-ups. Don't get me wrong, there was
>>NOTHING technically wrong with the 820 chipset or RDRAM, and, in fact,
>>the RDRAM P3's were the highest performing processors .... but 128 meg
>>of RDRAM cost $1,000 (yes, a GRAND), and the performance benefit was
>>like 1% to 3%.
>>
>>Meanwhile, however, because of the cost, there was a string of
>>catastrophic disasters. The worst of these was the 820MTH chip, a chip
>>that was supposed to allow the 820 chipset (RDRAM) to work with SDRAM.
>>The 820MTH was simply not reliable, and in the end Intel had to recall
>>every 820 motherboard that they had made with the MTH on it (using
>>SDRAM) and replace it with an 820 motherboard without the MTH (using
>>Rambus RDRAM). Note that the motherboards that were recalled were SDRAM
>>motherboards, not RDRAM motherboards. But they had an RDRAM chipset on
>>them (and an RDRAM to SDRAM converter chip that didn't work). There was
>>NO problem with RDRAM or the 820 chipset, the problem was the MTH and
>>SDRAM, but it was RDRAM -- Rambus memory -- that took the "PR Hit" and
>>that got the black eye.
>>
>>It may well be that Intel was so soured by this experience (it cost them
>>hundreds of millions of dollars), and some others at the time related to
>>RDRAM -- but mostly of Intel's making -- that they backed off of RDRAM
>>for the P4. Which is most unfortunate, because with the P4, RDRAM
>>provides a real benefit in terms of performance and stability /
>>reliability, a fact which most people, having used only SDRAM or DDR, do
>>not fully appreciate (although RDRAM does win virtually all of the
>>benchmarks). And the price premium for RDRAM is vastly less than most
>>people perceive, in fact for most of the last month Dell has been
>>selling RDRAM based 8250 series computers for LESS than [inferior]
>>DDR-based models.
>>
>>
>>james yi wrote:
>>
>>>ok now another follow up question. why did the p3's use rdram memory
>>
> which
>
>>>ran faster than pc133 sdram?? if the fastest fsb for the p3 was 133mhz,
>>>shouldn't the pc133 have been sufficient since it was in sync with the
>>
> bus??
>
>>>the intel 820 chipset used rdram.
>>>
>>>
>>
>
>

Barry Watzman

2003-01-26, 8:23 pm

klassick wrote:
> "Barry Watzman" <Watzman@neo.rr.com> wrote in message
> news:3E334BA3.10806@neo.rr.com...
> >snip<
>
> Wasn't there a legal battle over patent infringement also?? I seem to
> remember some shaky business practices from Rambus' end.
>
>

There is a huge legal battle over Rambus' patents on memory technology
and it's still going on, it has never been resolved.

Basically, Rambus filed a monster patent in 1990 that took a decade to
get fully processed. The patent office said it was too big and that it
had to get broken up into many smaller patents, which was done. This
same patent was also filed in Europe and was published by the EPO
(European Patent Office) BEFORE Rambus joined Jedec, so it was not a
"secret". Also, separately, Rambus signed "NDAs" (non-disclosure
agreements) with over 3 dozen companies and fully disclosed their
technology to these firms. Then Rambus joined Jedec, the body that sets
industry wide standards for memory.

What happened next is kind of a mess, Rambus' IP (patented technology)
ended up in the Jedec standard for SDRAM (and later DDR), and Rambus
added claims to their patents to cover things that were in the original
1990 patents, but not in the original claims. [This last point is a
legally complex point that is far too complex to be either explained or
understood by anyone but a patent attorney].

In 2000, Rambus claimed that the Jedec SDRAM standard violated their
patents. Seven of the top ten memory makers accepted this, signed
licenses and ARE paying rambus a royalty on both SDRAM and DDR (plus, of
course, RDRAM), but Micron, Infineon and Hynix charged Rambus with Fraud
in the modification of their patent claims and went to trial.

The first trial was the one with Infineon, and in a trial that I
personally believe was completely out of control, the Rambus
infringement case was thrown out (not even allowed to be presented), and
Rambus, which was the Plaintiff (brought the suit) ended up being
convicted by a jury of 4 counts of fraud, and was ordered never again to
even bring a patent infringement lawsuit against Infineon for Patent
Infringement on these patents.

Rambus, of course, appealed the entire case, and two of the four fraud
counts (on DDR) have already been overturned. The case has reached the
2nd highest court in the land, just below the US Supreme Court, the CAFC
(court of appeals for the Federal Circuit), which heard the appeal on
June 3rd. The 3-judge panel has not yet issued their verdict, which is
being watched by the entire legal community because there are issues
that effect virtually every single industry in the US with a "standards
setting body", which is almost every industry. By the way, Ken Starr
was hired to present the case against Rambus (yes, THAT Ken Starr). The
verdict is way past due and could come at any time, but the court was
still asking both sides for additional information as late as December 11th.

Even if Rambus wins, there will be another trial, because their
infringement case was thrown out (never even heard) in the original
trial. But if Rambus ultimately prevails, the bottom line is that
Rambus' 1990 patents may cover both SDRAM and DDR (or they might be
found to cover DDR only, or neither). If the Rambus patents are found
to cover SDRAM and/or DDR, Rambus would be entitled to a royalty for all
of the applicable memory types produced. [Remember that 7 of the top
10 makers of memory are already paying this royalty anyway; it's
reportely about 0.75% (.0075) for SDRAM and 3.5% (.0375) for DDR, so
it's not a critically huge issue in terms of the dollar impact on the
cost of memory or the price of a computer].

The "shaky business practices" that you refer to are Micron's claims
that Rambus used information from Jedec to make the modifications to the
claims of their patents. They were able to "sell" this view to a
patent-virgin Judge and a lay jury in the Virginia Infineon trial (it
may be relevant that the trial was held in a city with an Ifineon plant
that employs 5,000 people). However, a patent has two parts, the
invention and the claims, and you can't add claims that are not present
in the invention part, which cannot be amended after initial filing.
Conversely, if something IS in the invention part, under US law, the
claims can be added at ANY time, e.g. ten years later. So the issue may
hinge not so much on the addition of the claims but on whether the
claimed inventions were in the original 1990 invention (which has never
been modified or changed). That's a matter for the courts to determine,
which they are doing (this will be a key part of the CAFC ruling when it
comes out), but in any case it's not, as far as I can see, a moral or
ethical issue and based on what I've heard and seen, I will be very
surprised indeed if all of the remaining fraud charges against Rambus
are not thrown out.

The CAFC has "special" jurisdiction over patent cases, it has exclusive
appeals authority on patent cases (except for the Supreme Court) and has
technical expertise (engineers) on-staff to address technical aspects of
patent cases. The CAFC is a "patent expert" court, and the 3-judge
panel in the CAFC that is hearing this case is chaired by Judge Randall
Rader, who is quite possibly the most knowledgeable judge in the US, if
not in the world, on intellectual and patent laws. It should be an
interesting decision, if it ever comes out.

VernonOJ

2003-01-27, 5:23 pm

"Barry Watzman" <Watzman@neo.rr.com> wrote in message
news:3E334BA3.10806@neo.rr.com...
> The P3 with RDRAM was a mistake, not withstanding that RDRAM is, in my
> view [others will argue - vocally] the best memory type available.
> However, it offered no benefit for the P3, because PC100 SDRAM could
> fully meet the bandwidth requirements of the CPU.
>

One way or one reason or another, RDRAM is in the history books with steam
engines, BetaMax, CP/M.

Sponsored Links