ptnewell on TOM'S CONVERSION...
conversion.
Why did Tom 'Suddenly' Convert?
Most long time followers of the industry remember that Tom's Hardware led the charge of the hardware sites against RDRAM.
Shortly after the i820 chipset, coupling the Pentium III to RDRAM, was released, Tom's Hardware overclocked a 440BX
chipset to see how the PIII would work with PC133 and a 133 FSB. He found that it performed just about the same as the P
III with the (at that time) far more expensive RDRAM. 'Rambus is a horrible joke' he told the world.
Then he got nasty. Confessing a pathological hatred for Rambus (he said it made him sick just to write about the company), he
put up stickers on his site showing RDRAM with a large red-X overlaid. The PC community got the message. Anyone who
missed it at Tom's saw other hardware sites large and small echoing the message to avoid RDRAM. The i820 flopped. When
Rambus tried to contact Tom, to explain that Intel's coupling RDRAM with a bus capable of transferring only 1.04 GB/s (the
same as PC133 provides) effectively neutralized the extra bandwidth, he was indignant. “They tried to blame Intel for their
failure.”
Yet now Tom's Hardware declares Rambus the memory of the future. “DDR slows the Pentium 4 to a snail's pace” is a
headline that cannot easily be misinterpreted. Moreover, he provides benchmarks to show that the RDRAM advantage over
DDR increases considerably at higher CPU speeds, so that a 2.4 GHz P4/RDRAM beats a 3 GHz P4 with DDR. So why did
Tom 'suddenly' change?
By using quotation marks I've already signaled my answer: The change was not sudden, but has been happening slowly over
the last 18 months. Events since the fall of 2000 have gradually brought Tom's Hardware out. First, in August through October
of 2000, various DDR chipsets for the Pentium III were released (all of course coupled to Intel's pokey 133 MHz bus
transferring 1.04 GB/s). Via's performed horribly, SiS and ALI did worse yet. Many long time DDR supporters were shocked.
At that time, the Pentium 4 was supposed to take a very long time to ramp up (and was considered a bit of a dog anyway),
hence the conventional thinking was that Via and Sis could clean up with high performance PIII/DDR chipsets. If you look
back I believe you will find that the first bad things Tom Pabst said about DDR stemmed from those days, as did the first
doubts of many long-time DDR proponents. (I declared DDR not dead, but stillborn, which turned out to be true only of the
PIII/DDR chipsets). These PIII/DDR chipsets flopped even more miserably than the i820 did. And the thought entered more
than a few minds that maybe there was something to this idea that coupling a high-speed memory with a slow FSB and an old
CPU said less about the memory than the chipset architect.
In November of 2000, the Pentium 4 was released with the i850 (dual-channel Rambus) chipset. Here again, Tom got a shock.
On the SiS Sandra Soft tests designed to test the memory subsystem (which actually includes the FSB, and the capabilities of
the CPU itself to handle bandwidth), the i850 score was nothing less than spectacular. The combination was capable of
handling roughly 1500 MB/s, versus the few hundred MB/s effective rate common then (the numbers go up a bit with faster P4
speeds released subsequently). Burt McComas of InQuest (a small company with strong ties to Micron) had declared that
dual-channel RDRAM was not capable of even 1/3 that transfer rate. McComas had reached that conclusion by testing the
i840, which does have two Rambus channels, but still has the same inadequate bus as the i820 and i815. In other words, Intel
had connected the bus that was not fast enough for single channel RDRAM to dual-channel RDRAM, which must be the
industry's first encounter with absurdity squared. (There is a small improvement in latency, to be fair). McComas convinced
Tom's Hardware, AnandTech and other sites that 'the memory subystem of the i850 is exactly the same' as the i840.
The release of the i850 was the first time Tom's Hardware said anything positive about RDRAM. (He also fired anti-Rambus
fanatic Van Smith, presumably for unrelated reasons, who then worked briefly for McComas, trying to convince the world that
Pentium 4s have a thermal throttling problem). For example, Tom noted that the P4 seems bandwidth hungry, and stated there
was considerable doubt whether DDR could keep up. Nonetheless, Tom's conclusion at the time remained that Rambus cost
too much, and the company represented evil besides. (See my article on “Tom the Recidivist” for my take at the time [2])."
"In the Spring of 2001, Via released a P4/DDR chipset (P4X266). Tom's Hardware noted that the i850 tied or beat it on
EVERY test. Other hardware sites were taking notice also. Anand, initially declared that the i850 was not worth reviewing,
since the i840 results ('which have the same exact dual-Rambus memory subsystem') proved that RDRAM would 'cripple' the
P4, also compared the two. He also found that the i850 won or tied every test. Of course Anand concluded that the high cost
of RDRAM made the Via chipset a better buy, but at least RDRAM was starting to win nods from the hardware sites as the
choice for pure performance. (Incidentally, Via has since fallen back on the old rule that 'If you can't beat 'em, trick 'em'. Via
now overclocks the test systems that it sends to reviewers, so that its '2.0 GHz P4', say, is running about 4% faster than the 2.0
GHz P4 on an i850 board. Tom's Hardware was the first to catch, or at least the first to point out, this bit of trickery).
Thus Tom's Hardware has not made a 180 degree turn. Over the last 18 months, the anti-Rambus rhetoric has cooled, as has
the pro-DDR reports. For example, Tom has mentioned that i850 motherboards are the most stable available. The same
pattern holds true of other hardware sites also, as far as I can glean (without actually reading much of them). The recent 'Clash
of the Titans' article by Tom's Hardware concluded that “DDR hinders the Pentium 4 quite a lot”. They had used overclocked
P4s, running up to 3 GHz to discover that DDR just could not keep up. The faster the P4 ran, the further DDR fell behind, so
that a 3 GHz P4/DDR could not equal a 2.4GHz P4/PC800.
It has been suggested that Tom's Hardware must already have Intel's forthcoming i850E chipset (P4/PC1066 w 533 MHZ
FSB) in hand [3]. If not, Tom could easily make his own by overclocking, since most PC800 made by Samsung works fine as
PC1066. Tom's Hardware loves to overclock, and they clearly know that the performance advantage of RDRAM over DDR
is about to dramatically improve, and will continue to improve with each faster P4 speed grade released. It probably seems
wise to come clean with RDRAM BEFORE the i850E is released, especially for a hardware site that aspires to be the industry
opinion leader.
Finally, please note that although Tom Pabst must have personally approved the pro-Rambus article, he did not personally
write it. It might be easier to eat crow if you have your assistants do the job."
"Does it Matter What Tom's Hardware Says?
The problem Rambus has always faced is the enormous success Team DDR has had in convincing the world that DDR is
simple, easy, and inevitable, while RDRAM is complicated and probably will be gone in six months. Quarter after quarter
RDRAM shipments have grown, and yet I do not think you can find any time between mid-1999 and the present when more
than a few months have passed without some media outlet declaring that RDRAM would soon be gone [4]. The perception
issue has been self-fulfilling. If everyone thinks DDR will be easy and RDRAM hard, obviously one should build DDR chipsets.
What IT department will recommend computers that will be gone in six months, no matter how good the performance? Why
stock RDRAM computers at Best Buy if “Intel is replacing them with DDR”?
Many Rambus longs have wondered what sort of an advertising campaign was possible to change perceptions. I cannot
imagine any campaign as advantageous as the industry's leading hardware site declaring that “There is no way around Rambus.”
Everyone I know who has the least interest in PC hardware reads Tom's site. All who were formerly reading about how DDR
would replace RDRAM are now reading that RDRAM is better today, and will have an even bigger advantage tomorrow.
A conversion is rare in any debate, and carries greater influence. Which hypothetical would impress you more -- Tate stating
that Rambus was the memory of the future, or Sherry Garber stating the same? [5])
The general public will never know enough about memory to chose RDRAM on their own, excepting only a few percent
gamers and a few other small categories. This is true even though RDRAM beats DDR in Pentium 4 benchmarks. The
magazines and PC hardware web sites have to TELL the public, and IT buyers, what to buy. An excellent start is a headline
warning that DDR reduces a Pentium 4 to a “Snail's pace.” Again, I believe that other hardware sites such as AnandTech have
been softening toward RDRAM for some time. But an explicit, clear, flat out declaration that DDR just does not equal
RDRAM's performance today and will fall even further short in the future is what is needed to break through the cobwebs built
up about RDRAM. On February 15, 2002, Tom's Hardware again acted as the industry media leader.
[1] tomshardware.com
[2] “Tom, the Recidivist”
boards.fool.com
[3] stkhwk's view:
boards.fool.com
[4] For example (one for which I can find the reference if necessary), Andrew Young, of the Register, in the fall of 2000,
predicted the disappearance of RDRAM by spring 2001.
[5] Or Bilow, or elixe. Although I rather think if one ever converts, all three will."
Cheers |
