JUNE 2005 MA XIMUMPC 47
same front-side bus, and even though
the cores share a common “wall,” the
CPUs can communicate only through
the FSB.
Q
What’s cooking at AMD?A
AMD’s dual-core procs will
feature 1MB of L2 cache per
core with two independent
cores based on the Athlon- But instead of being joined at the
front-side bus, like the Pentiums, the
processors are hooked up to a “system
request interface,” which in turn hooks
into a crossbar switch plumbed into
the on-die memory controller. AMD
says its design is far more elegant and
suited to higher performance than
Intel’s two-chips-welded-together
design. AMD’s first desktop dual core,
the Athlon 64 X2, is expected to launch
at 2.4GHz this summer. But AMD
might actually beat Intel to market by
introducing and shipping dual-core
Opteron CPUs at 1.8-, 2.0-, and 2.2GHz
by the time you read this.
Q
So which chip is more ad-
vanced: Intel’s or AMD’s?A
That’s hard to say without
some field testing, but ac-
cording to Kevin Krewell,
editor in chief of Micropro-
cessor Report, AMD’s design is clearly
more sophisticated. Krewell says some
industry observers have joked that
when Intel makes dual-core procs, it
just cuts apart every other die, instead
of every die, as it does to produce a
single-core CPU. That’s a reference to
the simplistic way the Pentium D and
EE communicate: One core sits next to
the other, but they only communicate
via the front-side bus.
On paper, AMD’s dual core shows
more sophistication through its use of
the on-die crossbar switch and system
request. This design should lower the
overhead inherent to multi-threaded
apps and render AMD’s dual-core
procs more efficient. On the other
hand, Krewell says, the Pentium D and
the next-gen 65nm Intel Presler CPU
exhibit a simple elegance with their
straightforward, if not primitive, con-
nection.
Eventually, Krewell tells us, multi-
core processors might intermix their
guts more and even share such crucial
components as the cache. Shared
cache in a multi-core design would
help when the two cores are working
on the same data—think multi-thread-
ed apps—but it could hinder perfor-
mance when the CPUs are running two
different tasks.Q
Why not just make the dual-
core chips run at the same
speed as the single-core
chips?A
Neither Intel nor AMD want
their customers’ PCs to
spontaneously combust. It’s
entirely possible that running
a dual-core proc at full speed could
overheat the part and bring down the
power grid in your neighborhood. The
single-core 3.2GHz Prescott wasn’t
exactly a cool CPU, after all. While
nearly doubling the size of the diedidn’t double the thermals—thanks to
some power-saving sleight of hand on
Intel’s part—they haven’t figured a way
short of liquid-nitrogen cooling to jack
up the clock speed to 3.8GHz without
melting down your system.Q
What do these CPUs look
like, and will they fit in my
Intel or AMD motherboard?A
Both processors look identi-
cal to their single-core cous-
ins, because the brains of
the chips will continue to be
hidden beneath heat spreaders. Pry off
the heat spreader, however, and you’ll
see a much larger die. Both procs will
pop into older motherboards, but that
doesn’t mean they’ll actually work in
them.
If you own a Socket 939 mother-
board capable of running an Athlon 64
FX-55, you’re good to go. All the board
maker need do is drop in a new BIOS
to enable dual-core support. With
an Intel mobo, unfortunately, you’reThe Scarcity of Multi-threaded Applications
Dual-core processors won’t achieve their full potential until software is recompiled to take
advantage of them. Five years after Intel introduced Hyper-Threading CPUs, precious few
applications have emerged to take advantage of them. Why are software developers so
reluctant to embrace the new tech?
We recently put this question to Christopher Riley, a project manager with Abbyy
Software House (a leading developer of document-recognition software). “I suspect,” Riley
responded, “the biggest problem [facing developers] is that the process requires a large
revamping of code. Not on the actual changing of code, but a lot of performance plan-
ning, research, and testing for each specific application. It’s always hard to move to a new
paradigm.” The upshot is that getting good performance from a multi-threaded app isn’t
as simple as toggling a switch in a compiler. The programmer must plan how the different
threads of the app will interact before he or she writes the first line of code.
Riley knows of what he speaks: Abbyy just completed a rewrite of its FineReaderOCR
app to support multiple threads. “It’s more difficult, not because the programming is
extremely hard,” Riley said, “but because the planning required is greater.”
Several other developers we spoke with described the situation as a classic chicken-
or-egg dilemma: They won’t expend the resources to adapt their software for dual-core
processors until people have dual-core processors.
Dean Lester, Microsoft’s general manager of Windows graphics and gaming technolo-
gies, was slightly more encouraging: “Now that multi-core processors are expected to
hit the street this summer, we should see applications and games that take advantage
of the technology. Most likely next year. The reality is that it’s really helpful for game
developers to have access to the hardware before they can customize their code for
dual core; more importantly, consumers need to have the hardware before they can take
advantage of games written for multi-core processors. Until there’s a good-size installed
base, it doesn’t make sense for game developers to target this technology. That said, the
constant innovation on the Windows platform means that this technology will be widely
adopted in the near future.” Amen to that, brother.