MaximumPC.com Gets a Makeover
Our revamped website offers everything you love about the magazine
and then some!X Marks the
Warmth
Betcha didn’t know this: When
you spread thermal paste
across a CPU—even a fancy
name-brand paste—you
lose about 40 percent of the
CPU’s cooling effi ciency to the
ceramic or metal particles that
make up the thermal compound.
And according to IBM, the
CPU’s thermal budget is further
depleted by the peculiarity of
the paste’s fl ow behavior.
Researchers at the comput-
ing behemoth noticed that as
a CPU heats up, paste particles
bunch together along the diago-
nals of the chip, forming an X
pattern of sorts. Obviously, this
runs counter to the “spread
evenly” recommendation we give
for thermal-paste application.
To address the issue, IBM
has designed a trenchlike,
microscopic irrigation system
for CPUs. This lowers paste
thickness, reduces the pressure
between the CPU and the cool-
ing unit, and ultimately doubles
a chip’s cooling effi ciency. The
only downside? There’s no ETA
for this technology to hit con-
sumer chips.JUNE 2007 MAXIMUMPC 09
EMI is no podunk publisher—its open-format
offerings are sure to please many music lovers.
We’ve redesigned
MaximumPC.com
to deliver the latest
tech news, early
looks at hardware,
exclusive how-to
content, our favorite
content from back
issues, and the No
BS podcast—featur-
ing Gordon’s infa-
mous weekly rant.EMI to Sell DRM-Free Music
Will the industry’s fears of music-pirating
mayhem come to pass?A
pparently, one of the major music labels took
Steve Jobs’s open letter against DRM to heart.
EMI is fi nally stepping up and offering its music
catalog (except for the Beatles’s songs) sans copy
protection, so the tunes can be played on any type
of digital music player. Coming from the third-larg-
est music label, this is good news for consumers,
but the freedom comes at a cost.
The music tracks, which will be available
through iTunes beginning in May, will be priced
at $1.29 apiece—a 30-cent premium over iTunes’s
DRM’d tracks. While EMI and iTunes attribute the
added cost to the tracks’ 256kbps bit rate (versus
the standard 128kbps), we’ve gotta believe that
consumers are paying in part for the absence of
DRM. Still, at least it’s a start.I
n the last few years, PC processors have exploded
from single-core designs to dual-core and now
quad-core chips. The first eight-core PC processor
will probably appear by the end of this decade.
Beyond that, 12- and 16-core chips are possible.
What then?
Some CPU architects believe the current road is
a dead end, much like the pursuit of superfast clock
speeds with single-core processors. Although the
x86 cores that AMD and Intel use in their multicore
chips are more efficient than previous cores were,
today’s cores are still too large and power hungry to
sustain the momentum of multicore progress. Some
engineers argue that multicore chips must give way
to “manycore” chips.
What’s a manycore chip? There’s no hard defi-
nition. Basically, it’s a microprocessor with tens,
hundreds, or even thousands of processor cores.
It’s like multicore on fertility pills.
Intel recently hinted at this future by showing
an 80-core chip, code-named Polaris. It caused a
stampede of breathless coverage in the mainstream
press and on technology websites. However, Polaris
is strictly a prototype, not a product. Its 80 cores
are relatively simple FPUs, not x86-compatible
cores. Polaris is based on an experimental VLIW
architecture optimized for floating-point math.
In reality, Intel is playing catch-up in this race.
Other companies have already produced manycore
processors with hundreds or thousands of cores.
I’m talking about finished designs in actual produc-
tion, not lab experiments. Manycore processors are
running today in real-world applications, such as
base stations for wireless networks.
Nevertheless, Polaris is an important star to
steer by. The reason: Intel has another project to
develop a low-power x86 core for ultramobile PCs
and embedded systems. This core will probably
consume less than one watt. Intel could use it to
build manycore processors for desktops, notebooks,
and servers, too. Such processors could have 100
or more cores without a meltdown.
Of course, manycore processors don’t solve the
problem of writing parallel-processing software that
usefully exploits so many cores. Manycore chips
might also reveal software bugs that bite only when
running numerous threads (see last month’s col-
umn). But for CPU architects, there’s little choice. If
multicore processors reach a dead end, as high-fre-
quency single-core processors did, then manycore
processors might be the only alternative.Tom Halfhill was formerly a senior editor for Byte magazine
and is now an analyst for Microprocessor Report.Multicore
vs. Manycore
FAST FORWARD
TOM
HALFHILL