10 IS THE MAGIC NUMBER
Intel’s 10th-gen Core CPUs are here, and with them its 10nm production process. But, as Jeremy Laird found, the power of 10 is proving a little problematical.
It’s been five long years since Intel’s very first 14nm CPUs came to market. That’s worth repeating. It’s taken fully five years for Intel to roll out the 10nm successor to its 14nm production node. This from the company that has led the industry in manufacturing integrated circuits for decades. More than anything, it’s Intel’s advantage in chip manufacturing that has defined the company. But only now is it ramping up production of 10nm silicon as part of its new 10th generation of Core processors.
Of course, this is Intel, and nothing is ever simple. Strictly speaking, Intel trickled out a few dual-core 10nm chips in 2018. But if you can point us at an available retail product that uses these ultra-rare Cannon Lake chips, you’d be smarter than us. 10th-generation Core CPUs thus represent the first true availability of 10nm processors from Intel. Likewise, while Intel’s new 10th-gen line-up does include both 10nm silicon and an exciting new microarchitecture, it also encompasses not just one but two 14nm designs based on existing microarchitectures. Intel 10th-gen, in other words, means a lot of different things.
That’s just 10th-gen as we know it today. Don’t be surprised if it becomes even more complicated. And that’s just the products themselves. When it comes to branding, something Intel seemingly sees primarily as an opportunity to confuse its customers, the company is arguably achieving new lows with 10th-gen. The new nomenclature is so complicated and contradictory, it feels like you need a couple of MBAs, a law degree, and a PhD just to grasp the basics. But don’t despair. There is some genuinely new technology buried beneath the brain-dead marketing. What’s more, we’re here to do the donkey work for you, and make sense of Intel’s marketing madness.
Let’s begin...
Before we deep-dive into all those new 10th-gen chips, there’s an elephant that needs escorting from the path ahead. We speak of the extraordinarily long delay to Intel’s 10nm production process, the successor to 14nm in all its many forms. According to Moore’s Law, 10nm tech ought to have been in full flow by late 2016. After all, Moore’s Law famously predicts the doubling of transistor density every couple of years, and thus essentially dictates a new process technology from Intel every two years.
2016 sailed by, as did 2017. In 2018, Intel released a very small number of dual-core processors based on its 10nm node. But they were vanishingly rare and not true retail products. Even Intel described those initial 10nm products as “low volume.” If we were cynics, we’d say they existed only so that Intel could claim that 10nm was in production and on sale.
Whatever, it’s only now that 10nm CPUs from Intel are truly available, albeit with restrictions. At launch, 10nm is purely for mobile platforms. So, what happened? Inevitably, Intel hasn’t been expansive in its explanations, but two fundamental issues have very likely been at play.
One: Intel was particularly aggressive with 10nm with regard to transistor density. For some time, the nomenclature attached to production nodes has become disconnected from the size of features in chips. In other words, the features inside a “14nm” Intel processor – the transistor gates, the logic cells, and so on – are not actually 14nm. By Intel’s own figures, 14nm feature size actually ranges from 42nm through 399nm. That’s the same for all major makers of integrated circuits. The node names have become labels, indicators of a given generation, rather than directly reflecting feature size.
Anyway, with feature size decoupled from the node nomenclature, the label “10nm” doesn’t dictate transistor density. That’s down to decisions made by Intel. For 10nm, Intel aimed for “hyper scaling” of density, in this case a 2.7x improvement in the number of transistors by area of integrated circuit. That compares with a scaling of 2.1x with the transition from 32nm to 22nm, and 2.5x improvement when 14nm came on stream.
Making matters worse, Intel went after this hyper scaling just as it was also running up against the limitations of its existing approach to producing chips using lithography based on ultraviolet (UV) light. It’s the wavelength of the light used that informs the minimum size of features in a chip. UV light wavelength is in the order of 200nm. Using multiple masks and other tools, it’s possible to produce feature sizes smaller than the actual wavelength. But with 10nm, Intel was pushing up hard against the limitations of UV light. 10nm is the end of the line for UV lithography. At the same time, the technology for extreme UV or EUV lithography, with wavelengths below 15nm, wasn’t ready.
GENERATIONAL SHIFT
If that’s the 10nm story, what about the 10th-gen narrative? While Intel’s 10th-gen Core processors include new 10nm chips, there’s much more besides. But we’ll start with the 10nm hardware, because that’s where the newest tech is found.
10th-gen chips that are 10nm also debut a new CPU architecture, known as Ice Lake. Depending on how you measure these things, Ice Lake represents the first extensively revised Intel x86 architecture since at least Skylake from 2015, and just possibly the first Intel Core chips in 2006.
For Ice lake, Intel has cooked up both a new microarchitecture for its CPU cores, Sunny Cove, and new graphics tech, denoted Gen11 Graphics. Together they form the basis of Intel’s
“EVEN INTEL DESCRIBED THOSE INITIAL 10NM PRODUCTS AS “LOW VOLUME.” IF WE WERE CYNICS, WE’D SAY THEY EXISTED ONLY SO THAT INTEL COULD CLAIM THAT 10NM WAS IN PRODUCTION AND ON SALE.”
key CPU and graphics technologies going forward. The latter half of the equation includes Intel’s upcoming discrete graphics cards, which are likely to be based on the same architecture as the Gen11 integrated GPU in Ice Lake.
Ice Lake features start with a claimed 18 percent improvement in per core, per clock CPU performance compared to Skylake. Next up is a 2.5x increase in what Intel calls AI performance, the addition of AVX512 instructions, two times faster HEVC video encode, integrated Thunderbolt 3, DisplayPort 1.4a, and a whole lot more. In other words, Ice Lake is no simple die shrink with a few tweaks. It’s the first major overhaul from Intel in a long time, worthy of the 10th generation tagline.
For now, Ice Lake is limited to mobile applications and is exclusively available in dual and quad-core variants for low and ultra-low power applications. Thus, Ice Lake is targeted at thin-and-light laptops and below, the latter encompassing even slimmer form factors, such as tablets.
Arguably less worthy of the 10th-gen designation are the other members of the line-up. For now, the most straightforward 10th-gen CPUs are the new HEDT (high-end desktop chips), the beasts based on Intel’s workstation and server technology. Known as Cascade Lake-X, they’re 14nm chips, ranging from 10 to 18 cores.
Feature-wise, they’re dead ringers for their Skylake-X predecessors, which offered essentially the same microarchitecture and the same core counts. However, the big news with Cascade Lake-X is pricing. In short, Intel has pretty much cut pricing in half. The reason? AMD and the massively increased competition it is now presenting, especially in the form of its latest Ryzen 3000 series processors and the new third-gen Threadripper CPUs.
The final part of the 10th-gen puzzle, for now at least, is another set of mobile CPUs, this time based on existing 14nm production technology, an architecture derived from Skylake, and designated Comet Lake. Just like the Ice Lake chips we mentioned a moment ago, these Comet Lake processors are low and ultra-low power models, but along with dual and quad-core variants, Comet Lake also marks the first time that Intel has offered low-power six-core configurations.
Yes, you read that right. You can now buy a laptop based on a brand new 10th-gen Intel Core processor. And if it’s dual or quad-core, it may be low or ultra-low power, it might be 14nm or 10nm, it might based on Skylake from 2015, or it might be a brand new architecture. Exactly how it makes sense that both of these families of CPUs, plus the HEDT multicore chips, are all designated 10th-gen is beyond our comprehension. But what we can say for sure is that it makes for a massive headache when it comes to actually recognising these CPUs in shipping products, particularly with regard to the mobile CPUs.
Allow us to elaborate. Traditionally, Intel has designated its low and ultra-low power chips with “U” and “Y” signifiers, respectively. By way of example, the eighth-generation Core i5-8265U was a quad-core low power CPU, with a 15W TDP, and a maximum boost clock of 3.9GHz, while the Core i7-8500Y was an ultra-low power dual-core model, with a 4.2GHz boost clock, and a 5W TDP. Is it a little confusing to have a dual-core Core i7 and a quad-core Core i5? Perhaps. But the U and Y designations at least made it clear what class of CPU you were dealing with, so there was logic of a kind.
But not with 10th-gen. Oh, no. In total, there are currently 19 different 10th-gen mobile CPUs to choose from. If we begin with the 10nm Ice Lake chips, the new nomenclature goes like this. Take the new Core i7-1065G7 as an example. It’s a quad-core 10nm model with the most powerful “Iris” graphics option. Initially, the branding is fairly straightforward. The “i7” is Intel’s established brand modifier indicating a high-performance model within a given range of CPUs. Next up, the “10”
“WHICH OF THE TWO 10TH-GEN MOBILE FAMILIES SHOULD YOU GO FOR. THAT IS A VERY TOUGH QUESTION TO ANSWER, UNLESS YOU EITHER DEMAND SIX CORES OR WANT THE BEST POSSIBLE INTEGRATED GRAPHICS PERFORMANCE.”
indicates a 10th-gen Core processor. So far, so good.
The next two digits, the “65,” identify both the specific SKU in question and its status as a low or ultra-low power processor, “5” in this case indicating low power – alternatively, “0” would signify an ultra-low power chip. Keeping up? Finally, the “G7” reveals the level of graphics power, with “7” designating the top tier of Gen11 graphics, with fully 64 execution units. If you’ve got all that, you’ll no doubt immediately grasp that the Core i5-1030G4 is a quad-core 10nm ultra
low power 9W Ice Lake processor with a top Turbo speed of 3.5GHz, and Gen11 graphics, with 48 execution units, right?
Now, while this is a break from the past, it follows a certain logic. So, all Intel needs to do is stick with that logic and we all have half a chance of keeping up. Except that isn’t what happens. For starters, even within the Ice Lake series, there’s a Core i71068G7, which breaks the “5” and “0” indicators for low and ultra-low power. But that’s nothing compared to the confusion that occurs when you add the alternate Comet Lake 10th-gen mobile processors into the mix.
That’s because Comet Lake retains the old “U” and “Y” designators. Thus the Core i7-10710U is a low power 15W 10th-gen 14nm processor with six cores, Gen9.5 graphics, and a top Turbo speed of 4.7GHz, while a Core i5-10310Y is an ultra-low power 7W quad-core chip, again with Gen9.5 graphics, and a maximum Turbo speed of 4.1GHz.
Quite how Intel imagines mainstream consumers can follow any of this, we have no idea. It’s not entirely clear, for that matter, why Intel felt the need to offer two parallel low and ultra-low power CPU series within the 10th-generation line-up. However, we suspect that concerns over 10nm production capacity are at the heart of the matter.
The next obvious question is which of the two 10th-gen mobile families should you go for. That is a very tough question to answer, unless you either demand six cores or want the best possible integrated graphics performance. In the first scenario, only 14nm Comet Lake can deliver. In the second, you want an Ice Lake Processor with its fancy new Gen11 graphics. But for everything else? It’s unclear at best.
On the one hand, Ice Lake offers a more efficient architecture that does more work per clock cycle. On the other, Comet Lake SKUs tend to clock higher. As for battery life, that’s confusing, too. You might expect 10nm tech to be clearly superior for battery life, but the 10nm Ice Lake ultra-low power models, for instance, are rated at 9W, while the equivalent Comet Lake options are pegged at 7W.
In practice, your choice will often be dictated by other features, such as screen technology, chassis quality, and so on. But if you’ve narrowed down your options and are left with both flavors of 10th-gen CPU and little else to base a decision on, we’d lean toward the 10nm option for its stronger feature set. Extras such as integrated Thunderbolt 3 and support for DisplayPort 1.4a maximise connectivity options, particularly when it comes to adding future high-res displays.
Back on the desktop, of course, it’s a lot easier. On the HEDT side, the pricing of 10th-gen Cascade Lake-X makes it a massive step forward in terms of value. You get double the cores at any price point compared to Skylake-X. As for mainstream desktop, there are no 10th-gen chips at all, for now.
Overall, then, this is the most interesting new generation of chips from Intel in a long time. But also the messiest. It includes Intel’s first new production node in five years, and its most significant architectural overhaul in at least four years, not to mention debuting what will probably be the basis of Intel’s graphical assault on Nvidia and AMD. But amalgamating the genuinely new 10nm products with a bunch of rebaked 14nm chips under a single 10th-gen banner, then cooking up some of the most baffling branding yet devised, is also a source of frustration. At this point, the term “generation” as Intel uses it means pretty much zilch. Still, you now know exactly what to expect from the CPUs Intel has deemed to describe as 10th-generation Core processors. For a while, at least.