ALDER LAKE:
A BRAVE NEW
WORLD
Intel’s latest architecture is undoubtedly
one of the most complex undertakings in the
history of the company. Creating one new
architecture requires plenty of effort. Moving
to a new process node involves another huge
chunk of work. Couple that with the second
new CPU architecture, plus the hybrid core
design with its need for updated software,
then add in support for DDR5 and PCIe 5. It’s a
massive undertaking, and it mostly pays off.
Intel has matched AMD’s Ryzen 5000 series
in terms of raw core counts, though it trails in
the total number of threads. The new Golden
Cove architecture more than makes up for
that in most tests, and Alder Lake can now
lay claim to the overall desktop performance
throne. It’s not a perfect victory, however.
Power and efficiency tend to favor AMD’s
Zen 3 architecture. Intel’s talk about how
efficient the E-cores can be sounds good, but
in the real world, at the clock speeds Intel
is pushing, the 12900K still sucks down a
substantial amount of juice. That’s mainly
because the E-cores aren’t about competing
with AMD; they’re meant to fend off Arm.
You only need to look to Apple to see Intel’s
Arm fears made manifest. Several years back,
Arm and smartphones were one of the biggest
threats to Windows PCs. We’re still happily
running Windows 10, but Apple has ditched
Intel for its own homegrown M1, and now the
M1 Pro and M1 Max are kicking things up a
notch. Apple’s control of both the hardware
and software running on its Macs gives it
a lot of opportunities to tune performance,
and we’re witnessing the results of those
efforts. The M1 Max might not be able to beat
a Windows desktop in every task, but it wins
enough to be a serious threat.
Intel’s answer is a rather large 19 percent
boost to IPC on the Golden Cove cores, but
just as important are the efficient Gracemont
cores. In fact, the next generation Raptor Lake
processors will apparently double down on
the E-cores, so we’ll get a chip with the same
eight P-cores as the 12900K, but with sixteen
E-cores. The new design philosophy appears
to be providing enough high-performance
cores to take care of tasks that need higher
IPC and clock speeds, but for workloads that
are able to scale with lots of cores and threads,
spamming a bunch of lower performance but
more efficient cores can ultimately lead to a
better overall result.
For its part, AMD’s Zen 4 architecture
appears to be sticking with a more traditional
approach—it won’t feature a hybrid
architecture. But given the proliferation of
big.Little in the Arm arena and with Intel now
WELCOME BACK,
OVERCLOCKING
Overclocking is far
more satisfying with
Alder Lake than we’ve
seen on many other
recent CPU launches.
Make no mistake, Alder
Lake can use a lot of
power and thus needs
good cooling if you want
to push beyond stock
speeds, and there’s
not a massive amount
of headroom at the
top. However, unlike
AMD’s Ryzen CPUs,
it’s possible to tweak
clocks based on how
many cores are active,
and there are a lot of
other knobs to tweak.
The E-cores, as you’d
expect, tend to benefit
less from pushinghigher clocks. The
P-cores meanwhile can
generally hit 5.3GHz on
a good motherboard
without batting an eye.
Compare that to AMD’s
overclocking, which
often requires you
to reduce maximum
single-core clocks
to hit higher all-core
overclocks. What’s
more, there are a lot
of possibilities for
improving performance
by tuning the memory
subsystem.
It’s too early for us to
provide a detailed Alder
Lake overclocking
guide, but so far
sticking with Gear 1
and pushing memoryclocks as high as you
can reach with stability
seems to be the way to
go. DDR5 modules are
also in their infancy,
so we could see much
higher memory clocks
down the road—and
maybe Gear 2 will
become more useful
when we’re talking
about DDR5-8000. And
if you’re into extreme
overclocking using
LN2? Professional
overclockers have
already eclipsed 7GHz.Overclocking
headroom on Alder
Lake appears to be very
impressive—liquid
nitrogen not required.moving to a hybrid design, don’t be surprised
to see that Zen 5 or Zen 6 end up mixing
different types of cores.
With 14nm finally fading in the rearview
mirrors, Intel has plans to quickly move to
7nm-class, 5nm-class, and then 3nm-class
processes. It fell behind the competition, TSMC
and Samsung, during the past seven years,
but plans a decisive return to the lead. While
Alder Lake is the first step in that direction,
TSMC is already shipping 5nm-parts and
talking about 3nm production starting in- Intel still has plenty of work to do.
FEB 2022 MAXIMU MPC 35