Intel’s mantra for
Alder Lake was
“Wider, Deeper,
Smarter.” That meant
more execution ports
and deeper buffers,
among other changes.SIX YEARS OF
INCREASING
POWER
To put it bluntly, Alder Lake and many of
its underlying technologies are arriving
incredibly late to the party. Intel’s first 14nm
chips, the Broadwell architecture and 5th Gen
Core launched in 2015. Broadwell was mostly
a die shrink of the previous generation
Haswell chips, combined with a large 128MB
eDRAM cache inside the CPU package.
Broadwell was also late, which is why the
desktop version was almost entirely ignored,
but Skylake came out just a few months later
as the second iteration of 14nm.
That was the last time Intel officially
moved desktops to a smaller manufacturing
node. Since then, we have had refinements
to the 14nm process to try and keep it
competitive. However, you only need to look
at power requirements to see that these
weren’t keeping pace with the competition.
Broadwell’s i7-5775C had a 65W TDP, whereas
Skylake’s i7-6700K bumped that up to 91W,
and Kaby Lake’s i7-7700K stayed with that
value. The Coffee Lake i7-8700K added a
couple of cores and increased TDP to 95W, but
from then on, things started to go downhill
pretty quickly.
The i9-9900K was the Coffee Lake refresh
and still had a nominal 95W TDP, but peak
power (PL2) was set to 210W. Generally
speaking, the 9900K never managed to use
quite that much power, but it also rarely
stayed at or below 95W under load. Comet
Lake and the i9-10900K added two more cores
and came with a 125W TDP and PL2 increased
to 250W. While the CPU should theoreticallyonly run at heightened power draws for up to
56 seconds (Tau), in practice, many enthusiast
motherboards simply ignore that value.
Rocket Lake was the final voyage for 14nm
and the first time Intel shifted desktops to a
fundamentally new CPU core architecture.
Everything from Skylake through Comet
Lake used variants of the same Skylake core.
But Rocket Lake switched to the Cypress
Cove architecture, which was derived from
the Sunny Cove cores used in the mobile-only
10th Gen Ice Lake CPUs, except that it needed
to be backported from 10nm to 14nm. The
result was a rather large CPU core, again with
high power requirements. The Core i9-11900K
has a 125W TDP and 250W PL2 and, under
load, tends to stick closer to that 250W limit
on most Z590/Z490 motherboards.
Alder Lake finally ditches the 14nm
baggage and moves to a truly new
manufacturing node. Formerly called 10nm
Enhanced SuperFIN, Intel decided to rename
its manufacturing nodes to align with
the rest of the industry—meaning TSMC
specifically, but also Samsung. That’s fair
enough, as there’s nothing in TSMC’s N7
process that actually correlates to 7nm, nor
is there anything directly linked to 5nm in
the newer N5. Intel makes the claim that its
new Intel 7 process basically matches many
of the characteristics of TSMC N7, and since
the process naming schemes are effectively
broken, we’ll take that at face value.
AMD has up to 16-core CPUs with the Ryzen
9 5950X, using TSMC N7 running with a
105W TDP. Like Intel’s PL2, AMD CPUs can
exceed the TDP, but the maximum power
consumption is capped at just 142W. While
it’s certainly possible to make the argument
that Intel’s Rocket Lake cores can outperform
AMD’s Zen 3 cores in some single-threaded
workloads, the fact that AMD can run 16 cores
at around 4.7GHz while using only 142W,
Intel’s eight cores at 5.1GHz drawing 250W
is a serious problem, one that Intel aimed to
correct with Alder Lake.
Except that it has only partially addressed
that issue. The Alder Lake i9-12900K still
comes with a 125W TDP, but Intel has now
changed nomenclature and lists MTP, or
Maximum Turbo Power. This is still PL2, but
there’s no longer a time limit (Tau) for how
long a chip can stay at MTP, and Intel doesn’t
list an official TDP. MTP on the 12900K did
drop a bit, to 241W, but that’s still 70 percent
higher than AMD’s maximum 142W. The
difference is that per-core performance has
improved—and by a lot.
Across a suite of single-threaded tests,
the 12900K comes in 19 percent faster than
the Core i9-11900K and the Ryzen 9 5950X. It
also leads the 5950X in many multi-threaded
workloads. Power use is still higher, but
efficiency is far more competitive than the
previous generation 14nm chips.the empire strikes back
30 MAXIMU MPC FEB 2022
© INTEL