applications. You might recall
that Intel made a big deal about how much SSE2 would help improve the performance
of Internet media. As time has shown, a more accurate claim might have
been that the P4 helps improve the creation of Internet content. Other
Pentium 4 improvements include enhanced branch prediction, 20K of L1 cache,
256K of full-speed L2 cache (later bumped to 512K), enhanced floating point
performance, and arithmetic logic units that run at double the core CPU
frequency.
The original Pentium 4 arrived with a 100/400 MHz bus speed (recall that
Pentium 4 architecture is “quad-pumped,” meaning that data is
passed four times per clock cycle), but, more importantly, it also arrived
accompanied by the 850 chipset. The 850 northbridge utilized Rambus’s
RDRAM memory, which was superior to the conventional SDRAM modules of the
time. The problem was RDRAM’s prohibitively higher price tag. When
AMD, VIA, and others arrived with double data rate (DDR) DIMMs with increasingly
similar performance at far lower prices, Intel finally had to exit its
tight Rambus alliance and bring the Pentium 4 platform over to DDR compatibility
with the 845 chipset.
The Pentium 4 has taken on new innovations, as well, as it has progressed
through several generations. The jump to a 533 MHz FSB (not to mention
0.13-micron fabrication process instead of 0.18-micron) with the 2.4 GHz
chip noted the introduction of the second generation P4, informally known
as the Pentium 4B. Likewise, the Pentium 4C hopped to an 800 MHz FSB at
the 3.0 GHz mark. The Socket 423 PGA platform finally petered out at 2.0
GHz and ceded to the current Socket 478. This was also the speed at which
the Pentium 4’s
256K of L2 cache was universally changed over to 512K, a move that entailed
adding 13 million additional transistors to the chip. Correspondingly, the
Pentium 4’s massive transistor count and higher power requirements
spawned both the necessity for a separate 12V ATX power line to the motherboard
as well as a new heatsink retention mechanism to accommodate the P4’s
hulking heatsink fan units.
After the rise in FSB speeds, Hyper-Threading (HT) may be the single most
important innovation in the Pentium 4 line, although the technology originally
appeared in the Xeon line in March of 2002. In systems where HT is enabled
in the chipset, operating system, and applications, Intel states that the
system’s performance can increase by up to 25% just on HT’s
contributions alone.
A “thread” is essentially a data pipeline across which instructions
are fetched and executed. In the original Pentium 4, as with all preceding
Intel desktop chips, the processors were single-threaded. In multi-processor
environments, the operating system dynamically utilizes the thread potential
in each processor. With an HT processor, resources within the chip are divided
up to create two (or potentially more in the future) “virtual” processors.
Each virtual processor is not as effective as a dedicated physical processor,
but when the virtual processors operate in tandem, the overall performance
is superior. Hyper-Threading is most effective when applications are coded
to utilize HT capabilities, but even non-HT applications working under
an HT-enabled operating system can benefit. All Pentium 4C and later chips
have HT integrated into them.
We should point out that there may be times when you need to watch the
fine print in your processor selection. The “C” generation is based
on the Northwood core while the “E” generation is based on the
new Prescott core. Prescott marks the move from a 0.13-micron fab process
down to 0.09-micron. This feature shrink has enabled Intel to increase the
L2 cache from 512K to 1MB and add support for another 13 SIMD instructions
collectively known as SSE3. The oddball exception in Intel’s current
lineup is the “2.80A GHz” processor. With 1MB of cache and
a 0.09-micron process, the 2.80A looks like a standard Prescott chip save
for the fact that it sports only a 533 MHz FSB and lacks Hyper-Threading.
There is a separate 2.80E chip that is a true Prescott part.
|
|
a standard
512K L2 cache, Gallatin/EE chips add on a whopping 2MB L3 cache. This
extra cache combined with Hyper-Threading can yield a noticeable improvement
in some applications, particularly
those that heavily emphasize 3D graphics performance, and some reviews
have found that the EE beats AMD’s Athlon 64 FX on several benchmarks.
This is why the exorbitantly priced EE chips are marketed at gamers and
power users—the same crowd that doesn’t blink at spending
$400 or more for a video card. Presently, only the 3.20 GHz EE is listed
on pricewatch.com, with the lowest street prices starting at $859.
And not least of all, what of Intel’s news about starting support for 64-bit
extensions in its IA32 processors? Contrary to popular perception, Intel may
not have crafted its x86-64 technology as a response to AMD. In a February interview
at the Intel Developer Forum, Intel chief technology officer Pat Gelsinger indicated
that Intel has been considering 32/64-bit approaches since the early ‘90s.
It wasn’t until “1999 or 2000,” however, that the company undertook
to cement a path that will culminate in the approach debuting in its Xeon chips.
If four years from planning to implementation sounds suspiciously long, recall
that the Opteron arrived nearly two years late and the Athlon 64 family’s
schedule was repeatedly smacked back like a final Microsoft antitrust ruling.
At the recent Intel Developer Forum in February, Intel announced that it would
add 64-bit support to two forthcoming Xeon chips: a dual processing version code
named Nocona and an MP version named Potomac. Moreover, Intel revealed that it
will add 64-bit functionality to the Prescott core. (Many conjecture that the
hardware is already there but is presently disabled.) Intel’s position
is that x86-64 support will only be made available to business machines, meaning
servers and workstations in the Xeon line, until further notice. Buzz on the
street speculates that “further notice” may arrive toward the end
of the year. No one would be surprised if the timing also proved to be fairly
close with Microsoft’s release of 64-bit Windows.
Additionally, such a long-range hinting of future 64-bit Prescott chips plays
the FUD (fear, uncertainty, and doubt) game at which Intel is so masterful. As
enthusiast consumers and businesses refresh their systems around the holiday
buying season and know they want 64-bit functionality, AMD will have the first-mover
presence and reputation in the market. “But Intel will have 64-bit very
soon,” the prospective buyer may think. “Sure, it’s the buying
time, but I don’t want to buy the wrong thing. Maybe I should wait just
a little longer.”
Your advice will be critical for customers as they waver on this point. Which
64-bit solution will be better? We have no idea, and neither does anybody else
outside of the processor R&D labs. Perhaps the best way to answer the question
is to side-step it and return to the perennial maxim of PC sales: The best time
to buy a product is when you need it. If the Athlon 64 or 64 FX has valuable
features that the user can employ and Intel still has yet to provide a 64-bit
equivalent, then sell the AMD. If both rivals stand evenly pitched in the market
with equivalent products that offer equivalent value to the customer, then perhaps
you need to revert to a selfish approach and examine which product poses the
best overall margin opportunity for your operation.
Customer first, margin second.
|
|
The Channel View
on Chips
We won’t try to snow you. The desktop processor space is a tough
market that has faced wave after wave of commoditization over the last
20 years. The price point for an average mainstream system has fallen drastically
from its $1,500 level during the mid-‘90s, but the recent desktop
PC market has stayed nearly flat since the pre-Y2K sales surge, denying
an increase in shipment volume to compensate for the profit decline in
falling ASPs.
“The growth opportunities in desktops can be fairly limited,” says
Dean McCarron, president of Mercury Research. “The desktop market
has been under a lot of pressure as the portable market is building. Within
desktops, outside the 64-bit class, there isn’t much change or
growth. Now, there are some niche areas, especially in the small form
factor and quiet PC areas, that are growing, but they use the same chips
as standard desktops. The difference is that product selection may be
done more carefully to optimize for thermal issues and things like that.”
The news seems even more grim when you consider that industry projections
show notebook sales surpassing desktops in 2007. This begs the obvious
question: Where will you find opportunities in the desktop processor
field?
Not everyone thinks the desktop processor market is grim. While Louis
Burns, vice president and general manager of Intel’s Desktop Platform Group,
wouldn’t give us specific numbers, he did tell us that the desktop
outlook for 2004 was “very, very good” and “far from
flat.” Allan Schroeder, director, marketing and product management
for MA LABS Inc., feels that the replacement cycle for all those Y2K
machines is primed for coming around at last. While most PCs get turned
over on an average of every three years, Schroeder contends that the
replacement cycle has been disrupted because of the poor economy combined
with an influx of surplus used systems left over from the dot-com fallout.
However, even these surplus machines from 2001 and 2002 are now growing
long in the tooth, and the tech economy is at last showing some indicators
of turning around.
Most sources we speak with agree that the low end of the desktop market
is an increasingly lost cause for small system builders. The momentum
for resellers and the channel as a whole is shifting into the high end.
“Everything is moving upscale,” says Schroeder. “For
a while, it was the war of the Celeron against the Duron, but now Intel
and AMD are both focused on high-end products.”
The reasoning is clear: The feature sets found in the latest desktop
processors—64-bit
extensions and NX based-security for AMD or the Hyper-Threading, SSE3,
and large cache sizes in Prescott-based Pentium 4 chips, for example—lend
themselves to complete solutions. Companies that want to run 64-bit servers
and terminals today have an increasing number of options in the Linux
world with distributions from the likes of Mandrake, Red Hat, and SuSE,
and a growing number of high-end consumers are taking an interest in
64-bit Linux, as well. (For a better sense of this space, check out an
article from last December at OSNews: http://www.osnews.com/story.php?news_id=5428).
Of course, the gamer market is doing nothing but growing, and the latest
chips from both vendors make bold appeals to this group.
“There’s a lot of buzz around 64-bit,” says Mercury Research’s
Dean McCarron, “but volumes will be limited for the next six months.
The feature set does command a good price premium, though, and the public
interest level is high.”
Availability is a problem that has sporadically plagued both Intel and
AMD over the years, and AMD has taken its turn most recently with the
Athlon 64 line. However, Geno Marcoux, general manager and vice president
of the Ingram Micro components business unit in the U.S., says that he
sees processor supply in the channel becoming much more stable than in
years past.
“The CPU business has been good,” says Marcoux. “Manufacturers
have really taken a look at trying to get more linear and managed, trying
to control their production cycles better. There are still peaks and valleys,
but I think you’ll see the delta between over- and under-supplies
smoothing out.”
Over-supply, of course, is part of what continually feeds the microprocessor
gray market. Companies from throughout the PC industry are working to
suppress gray market opportunities. Not only does gray market supply
erode Intel’s
profit and drop ASPs throughout the channel, it thwarts the business
development opportunities provided through vendor channel programs and
opens the door to counterfeit/remarked parts being sold to end-users,
a phenomenon that still happens periodically.
“The reason the gray market exists is because of arbitrage,” says
Marcoux, “either through pricing in different regions or different
programs that allow that arbitrage to create a delta and allow people
whose business models are based on taking advantage of that arbitrage
to exist. Ultimately, the vendor is the one who has to step up and put
some effort into cleaning it up. I think Intel did that, and then AMD
quietly followed suit.”
Of course, not everyone feels the gray market is the scourge of the processor
industry. MA LABS is one of the largest IPDs around, yet the company
continues to do a huge business in gray market chip sales. In fact, rather
than seeing his company as facilitating an industry problem, Allan Schroeder
perceives that MA LABS is doing the channel a needed service.
“The manufacturers are the ones who control the gray market,” says
Schroeder. “They’re the ones who have relationships with OEMs
who either need the quantities of products they buy or buy those quantities
and then get rid of some of them somewhere. We’re trying to help
our resellers compete with Dell. Our resellers offer local service and
local availability, but Dell has huge economies of scale and purchasing
power with their OEM relationships. We view ourselves as a service organization
for system builders and VARs. It’s our job to find them the lowest
possible price. We actually make the same amount of money whether we’re
selling an authorized product or gray market. If we can find the same product,
we’ll let resellers make the choice.”
Ultimately, the real money to be made in desktop processors lies in creating
a complete package with software and peripherals that relies on the CPU’s
most cutting edge advantages. You can see this most clearly in the media
center PC movement, which may prove to be the biggest and most profitable
desktop opportunity of 2004. Not only are Hyper-Threading and multimedia
extensions a huge boost here, but the 64-bit element will be a tremendous
help as 64-bit Windows and third-party PVR applications evolve.
When we spoke with Intel’s Louis Burns, he was passionate about home
entertainment and convergence being at the center of tomorrow’s
mainstream PCs. His only fear was whether Intel will be able to provide
enough performance for next-generation usage models.
“Say I’m encoding two HD shows off a satellite and putting
those on the PVR,” says Burns. “I’m also streaming one
HD stream and I’m streaming a HD movie, but it’s not going
to be able to be played in HD, so it’s transrating and transcoding
that. In another part of the house it’s playing my music, and someone’s
still doing the books or doing their checkbook. You’re gonna need
a serious amount of processing for all that. But people are also going
to want high-definition audio and high-definition video. If you look at
the penetration already of HDTVs or HD-capable TVs, there’s been
like a hockey stick move on that because people aspire to that level of
entertainment, and they see that the content’s starting to show
up in that space.”
So while you may not see a large trend upward in desktop processor sales,
there is still plenty of opportunity to make the space more profitable,
both on the front side of the sale and the back side with the vendor’s
channel programs. If you can turn the latest technology innovations into
meaningful solutions rather than faceless commodities, you’ll find
there’s far more of a goldmine in processors than just their wiring. 
|
|