Tuesday 8 March 2016

Skylake Goes Mobile

Skylake Goes Mobile

Intel’s range of CPUs has penetrated every nook and cranny of portable PCs. Which do you want?

Intel’s new sixth-gen Core processors represent a departure for the big blue chip-making machine in more ways than one. First of all, it decided to deliver both a tick and a tock in the space of a couple of months. It also figured that it would rather get the desktop chips out the factory door first, then follow them up later with a whole slew of mobile processors, rather than the other way around.


Last summer saw Intel’s Broadwell desktop parts finally arrive—with more of a whimper than a bang, it has to be said—only to be followed shortly after by the debut silicon of its brand-new Skylake architecture, again on the desktop. Broadwell was the die-shrink tick, with Skylake being the processor designed to really make use of all the goodness a 14nm production process can offer.

Launching the new Skylake architecture desktop-first may seem a bit odd, given Intel has focused on efficiency since the launch of Sandy Bridge. That push for efficiency means it has taken a mobile-first design approach, and then essentially iterated more powerful versions of the laptop chips to make the standard desktop CPUs.

Intel also has its enthusiast desktop platform, of course, the beefy X99 boards and Haswell-E processors that are packing six or eight cores. But for those, Intel went to the workstation guys making the Xeon server CPUs, stripped out a few features, unlocked the multipliers, and dropped them onto the consumer market.

Intel hasn’t specifically designed consumer desktop CPUs for years, and Skylake is no different. Whatever the reason behind holding back the mobile launch, we are now seeing a staggering range of laptop and tablet processors arriving with the sixth-generation Core in swathes of mobile devices across the world.

From configurable TDPs of just 3.5W all the way up to 45W—and even overclockable K-series laptop CPUs—Intel has created the broadest range of mobile chips it has ever made. Which presents a bit of a problem for end users, with all the new naming conventions, and vastly different feature and component sets. Which mobile Skylake processor do we actually want to see as the beating silicon heart of our next device? It's time for some explanations.

It isn't an exaggeration to say that Intel has created its broadest range of mobile processors ever with the new Skylake silicon. There are nearly 50 different CPUs designed for tablet, 2-in-1, AIO, and laptop devices. It can be overwhelming at first glance, but it's easier to talk about them in three classes, grouped by ascending TDP.

At the bottom of the power-draw spectrum are the Skylake-Y chips at 4.5W, then the Skylake-U series at 15W and 28W, and finally the 45W high-end Skylake-H series. They are designed to be dropped into different devices but all share the same basic Skylake architecture. So they all have the same 14nm sixth-gen Core design at their heart, and the same graphics technology, but are differentiated by feature set, CPU cores/threads, and the number of graphics execution units they contain.

That’s an incredibly impressive place to start, if you think about it. Intel has created a Core architecture that is so scalable that it can go from a miniscule 4.5W dual-core part, which will be used in tablets and be passively cooled, all the way up to a powerful 45W quad-core gaming processor that can take on desktop CPUs.

The main aim of an architectural tock, as opposed to the 14nm die-shrink tick of the Broadwell generation, is to drive home the advantages that change in production process can offer, and design the overriding CPU architecture to match it. So Skylake is all about improving the overall performance—seemingly with a focus on really beefing up the graphical performance—while at the same time making it far more efficient.

When it comes to the mobile side of the computing world, that efficiency is king. All-day battery life, generally around 10 hours, is the name of the game, as is advanced stand-by modes that don’t drain precious battery power unduly when the machine is not being used. And with the increasingly visual nature of the computing world, doing all that while also improving the graphical performance of Intel silicon is absolutely vital.

Speed Shift


Skylake brings with it improvements to the clock management of its processors, which effectively replace the old Speed Step feature, and were introduced via a Windows 10 update back in November. By an OS update? Speed Shift both is and isn’t a hardware-based feature—in order for it to work, it needed operating system support, which meant it wasn’t available at the launch of Skylake’s mobile chips. What it does is allow the operating system to completely hand over step changes in processor frequency to the CPU itself. Basically, it allows the processor to govern its own clock speeds, with the goal of making them happen much faster than if it had to jump through OS hoops first.

The whole idea is to improve efficiency by allowing the CPU to quickly ramp up to maximum operating speed when a task is delivered to the processor, which means it can be completed faster, and then return to its idle frequency faster. Intel estimates that it takes around 100ms for its processor to reach maximum frequency without Speed Shift; but with it, an example Skylake chip can jump from 1GHz to 2.5GHz almost instantly, and hit its 3.5GHz maximum speed in just 35ms.

All told, Intel believes it can complete some tasks in around 50 percent of the time it would normally take. By hitting top speed faster, it doesn’t need to run for as long. That improved efficiency means less drain on the battery, and longer up times all round.

Skylake-Y


That’s the codename, but the consumer name is now Core m, and Intel is trying to bring the previously super-confusing nomenclature more in line with its standard just-plain-confusing naming conventions.

Core m is still Intel’s premium low-power processor, but it's now rocking a lower-case "m" as opposed to Broadwell’s big-boy Core M. The reasons will become clear soon—where the first Core M processors sported names such as Core M 5Y70, Skylake is now aping the Core i3, i5, i7 terminology it uses with its more power-hungry silicon.

As such, we now have Core m3, m5, and m7 to denote the varying levels of performance to expect from the respective powersipping CPUs. Beneath the lot, though, just to keep things really confusing, is a single Pentium version of the Core m, with basic HD graphics cores and no Turbo mode. There’s a "Y" in the codename to help denote the Pentium 4405Y as a Core m part, though.

All the Core m parts carry the same dual-core, quad-thread design, and come with the same Gen9 GT2 graphics core with 28 execution units (EUs) inside, and a 300MHz base frequency. Unlike the rest of the Skylake line, they also only operate with DDR3L memory, as opposed to the DDR4 standard the rest of the Skylake chips interface with.

The Core m3-6Y30 has the lowest operating frequency, at just 900MHz, but the key to Core m is the power of its Turbo Boost mode combined with the new Speed Shift feature. The Core m3’s single-core Turbo Boost is a healthy 2.2GHz. There is a pair of Core m5 parts, but for all intents and purposes, their only difference is a 100MHz advantage on one of their singlecore Turbo Boosts. They both have a base of 1.1GHz, with a Turbo of either 2.7GHz or 2.8GHz, and the GT2 graphics have a Boost clock of 900MHz.

The top Core m7-6Y75 has a base of 1.2GHz and a single-core Turbo clock of 3.1GHz, with a GT2 Turbo clock of 1GHz.

These chips are designed to go into 2-in-1 designs, tablets, and teeny PCs such as the Compute Stick. The difficulty Intel found with the Broadwell Y chips, though, is that while they could potentially go as low as 3.5W or as high as 7W, thirdparty manufacturers weren’t designing the chassis to cope, which ended up hobbling the burst performance of the chips. Intel says it is being more strict on the manufacturer guidelines this time around.

Skylake-U


We’re back in more familiar straight Intel Core territory now, with the Skylake-U series of chips. These come in i3, i5, and i7 forms, in two broad categories: 15W and 28W. The raison d'être for these parts is to form the basis of the broadest range of Intel-based devices. We’re talking thin and light notebooks, portable AIOs, and mini PCs, such as the NUC, as well as power tablets, such as the Surface Pro 4.

Once more, the core configuration is pretty straightforward with both the 15W and 28W parts restricted to two cores and four threads, no matter their i3, i5, or i7 leanings. On the 15W end of the TDP scale, the i5 and i7 variants have four discrete chips in each category, joined by a single i3. Each is rocking either HD Graphics 520 or Iris Graphics 540.

It’s interesting that you’ll be able to pick up a mobile Skylake CPU in the cheaper Core i5 bracket, which will still come with the Iris Graphics component. The GT3e part of the Iris Graphics 540 silicon comes with the same 48 EUs as the last generation Iris Pro parts, though only has 64MB of eDRAM, as opposed to the top Broadwell SKU of 128MB eDRAM.

The easy way to tell if a U-series chip has the Iris Graphics is that its name won’t contain a double zero. A Core i5-6300U has the base GT2 24 EUs graphics part, while the Core i5-6360U comes with Iris Graphics.

That works with the 28W U-series parts, too, because all four come with Iris Graphics as standard, though they get the slightly higher-tier 550 version. It still has 48 EUs and 64MB of eDRAM, but has a higher TDP and max potential clock speed—though only the top i7 and top i5 chips in this bracket are capable of it. The good news, however, is that there is an i3 version in the 28W lineup, which comes with Iris Graphics, but without the i5 or i7’s Turbo Boost tech or the higher GPU clock speed.

You can tell whether it’s a higher TDP version of the U-series Skylake CPUs by checking whether there is a "7U" at the end of its model number.

Skylake-H


These are the big-daddy 45W CPUs of the mobile Skylake game. They come with the highest clocks, and represent the only option if you want the full quad-core, eightthread mobile processing enchilada.

If you were hoping that this would be where you got the complete Iris Pro graphics package, you will be disappointed. All of the Skylake-H parts come with the standard HD Graphics 530 core, with 24 EUs and no eDRAM. Given they’re designed for top mobile performance laptops, these are the processors that are going to get matched with beefy discrete graphics cards.

They are also the most familiar for any of us used to dealing with the vagaries of processor names and die configuration. The sole i3 chip, the Core i3-6100H, is a dualcore, four-thread CPU; the two i5 chips are straight quad-core, four-thread parts; and the four i7 processors are the peak quadcore, eight-thread CPUs.

They scale in a fairly linear way in terms of base processor speeds and Turbo Boost clocks, with the i3 running at 2.7GHz, no matter what, right up to the Core i7-6920HQ, with a 2.9GHz base clock and a peak Boost clock of an impressive 3.8GHz. That’s going to need some serious cooling.

Speaking of cooling, there is a bit of an anomaly in the ranks of these Core i7 CPUs—the Core i7-6820HK. Yep, a K-series mobile part (see “Mobile K-Series"), the first of its kind, which means it’s a deliberately overclockable laptop processor.

Whither Iris Pro?


That’s going to be the question on the lips of the graphics-obsessed amongst you reading this far. Even though Intel has officially launched almost 50 shades of mobile Skylake processors, not one of them has come with the peak of its Gen9 graphical prowess.

The GT4e, or Intel Iris Pro Graphics 580, iGPU comes with a mammoth 72 execution units, combined with 128MB of eDRAM, and represents the pinnacle of Intel graphics. It’s the Skylake graphics core with the potential for over 1.1TFLOPS of GPU power. That’s an Intel CPU with the graphical performance of an Nvidia GTX 950M. That means processor graphics that can actually manage 1080p gaming.

And yet they’re still nowhere to be found. Intel has teased that they’re to be used on the upcoming Skull Canyon platform, which is due to be a new NUC machine appearing early this year, but other than that, we’re clutching at rumors.

Toward the end of January, a reported Intel OEM price list appeared showing an upcoming set of Skylake-H processors all sporting the new Iris Pro cores. There’s supposedly a sole quad-core Core i5-6350HQ, running at 2.3GHz, with three further Core i7 models—6770HQ, 6870HQ, and 6970HQ—with base clocks of 2.6GHz, 2.7GHz, and 2.8GHz, respectively.

When these will actually arrive in machines is anyone’s guess, though the Skull Canyon timing of Q1 of this year should mean we won’t have to wait too long. The prospect of an i5 Iris Pro could make for a relatively powerful 1080p gaming laptop, where the battery doesn’t suddenly vanish the instant you fire up a game.

And, while the situation is similar with the desktop chips, where there isn’t really a huge number of tangible performance improvements over the last few CPU generations, upgrading on the mobile side makes more sense. The graphics improvements, boosts in efficiency, and Speed Shift power states mean that the mobile parts are still making gains over their progenitors, even if not in straight CPU performance terms.


Mobile K-series


The launch of a new slew of mobile CPUs is rarely that exciting, especially when Intel keeps back the highest end of its silky new Gen9 graphics cores until Krzanich knows when. But there really is one standout processor from the latest list of laptop processors, and it’s the first dedicated overclockable mobile CPU we’ve ever seen.

The Intel Core i7-6820HK is, on the surface, a relatively traditional quad-core, eightthread 45W processor. Its starting base clock is a modest 2.7GHz, with a maximum Turbo Boost of 3.6GHz, making it not that much different from the rest of the Core i7s in the Skylake-H category. It’s got the same 530 series graphics and the same iGPU clock speeds as the others, and the same 8MB L3 cache.

But it just doesn’t care what the hell you do with its CPU multiplier, with a healthy disregard for its own safety, just like all the best processors. That means, with the right cooling array, you can push well past that 3.6GHz limit—a quick look at HWBot and you’ll see it’s rated on there at 4.067GHz right now, though some demos have shown it going as high as 4.2GHz. That ought to give you proper desktop levels of processor performance from a simple 45W mobile part.

Proper cooling is key, or your expensive laptop is going to find a soggy, molten mess where its internal components used to be. We've seen the Core i7-6820HK in MSI's GT72ST, but the ASUS GX700 takes cooling to another level. It's a desktop replacement with a detachable water-cooling loop for when you want to go nuts with the clocks.

The Core i7-6820HK, then, is the enthusiast-level CPU to be seen with in your gaming laptop—whether you bother tweaking its multiplier or not.

Tick-Tock-Tock?


It wouldn’t be an article about a new processor launch without a little future-gazing thrown in for good measure, and the next generation, post Skylake, represents an interesting shift in Intel’s traditional design cadence.

The tick-tock model has been in use for a decade now, where the tick represents a shrink in production, using existing processor designs, and the tock sees a whole new CPU architecture introduced at that lithography.

That’s going to change with the upcoming Kaby Lake line of Intel processors. The seventh generation will essentially be built on the same architecture as Skylake and the same production process, but should include generational performance enhancements, as well as new features to differentiate itself from the previous CPUs.

This lengthening of a production process’s lifetime is the result of issues Intel has run into with the physical limits of shrinking its transistors. Initial 14nm yields were low and costs were high, which delayed Broadwell’s appearance (hence the Haswell Refresh chips), and the introduction of Kaby Lake will give it time to get the 10nm process right for Cannonlake in 2017.

Cannonlake will keep whatever architectural improvements are introduced with Kaby Lake, but on a new 10nm process. A whole new architecture will follow that, but we haven’t heard about its codename yet. The likelihood of that being followed by a 7nm part the following year is looking incredibly slim, then. Chances are that the two-year die-shrink cadence will be a thing of distant memory as it becomes harder and harder to get transistors down further into the single-digit nanometer world. But we’re not going to start predicting the death of Moore’s Law just yet—we’ve all been bitten by that in the past.