INTEL’S NEW CPU
4K-loving, 7th-generation Kaby Lake explained, overclocked, and rated!
LET’S BEGIN WITH a note of clarity and candor. There is much about Intel’s new seventh-generation Core processor family, codenamed Kaby Lake, that makes us want to do bodily harm to the leading maker of PC processors with a boxed copy of Windows 98. Because Kaby Lake is yet another incremental step that tops out with four cores in desktop trim, and is overwhelmingly optimized for mobile PCs. There are no radical new architectural changes, no promises of revolutionized performance.
It feels as though Intel has been rolling out updates like Kaby Lake for years and years now, safe in the knowledge that its only rival, AMD, simply cannot compete with it. And yet Kaby Lake actually looks kind of cool. It’s not cool in the old-school sense of shaking up the benchmarks, or tearing up the rule book for CPU architectures. But Kaby Lake will allow you to do a few genuinely worthwhile things with your PC that haven’t been possible before.
Much of that involves a major upgrade to the chip’s 2D video encode and decode abilities, the upshot of which will be twofold. First, Kaby Lake will allow 4K video playback on whole new classes of devices. Very likely, it will also mean that 4K streaming services are finally going to be supported on the PC. That’s right, if you’ve been frustrated by a lack of 4K Netflix support on your machines, then give it up for Kaby Lake. It could be the solution.
What’s more, Kaby Lake isn’t a completely busted flush from a traditional enthusiast’s point of view either. Granted, Intel has had some serious problems with its chip production technology of late. But that’s not exclusive to Intel—the whole industry has been struggling as chip feature sizes shrink to the point that they begin to flirt with the very fabric of the universe.
Despite that, it seems as though some useful headroom has been achieved. In other words, Kaby Lake might just be nicely cut out for overclocking. Plus, as ever with a new chip family from Intel, there are new chipsets. So there are also some platform-related upgrades to sweeten the deal. Kaby Lake may not be the radical game changer you’d hope for, but it will probably still be the chip of choice for at least the next six months.
KABY LAKE REPRESENTS no less than the third new processor family to be released on Intel’s 14nm process. That is the critical factoid that puts Intel’s latest CPU into proper context.
For starters, it’s symptomatic of the problems Intel has had shrinking the transistors inside its chips down to ever smaller proportions. Prior to Kaby Lake, Intel had been running to—or at least attempting to run to—a self-imposed cadence of product launches known as “Tick-Tock.” The idea was simple: One year Intel would shrink an existing processor architecture via a new manufacturing process node, the next a new architecture on the existing node. Rinse and repeat, long live Moore’s Law, and so on.
Except Intel couldn’t keep up. With each new node, there was a little Tick slippage, leaving several of the Tocks looking short-lived. Tick-Tock was introduced in 2007, the year Intel’s first 45nm chips went on sale. Follow the logic of Tick-Tock, and the schedule should have been 32nm in 2009 (that slipped into 2010), 22nm in 2011 (didn’t happen until April 2012), 14nm in 2013 (September 2014 in reality), and 10nm in 2016. 10nm obviously hasn’t happened yet, and probably won’t until late 2017, by which point Intel will be a full year behind its Tick-Tock schedule. Whoops.
None of which is a critique of Intel. The entire chip industry has been struggling to keep up with the two-year cadence for doubling transistor densities as famously dictated by Moore’s Law. Instead, Intel’s struggles are the background to a new paradigm being introduced with Kaby Lake, namely “Process, Architecture, Optimization.” So that’s a new production process in year one, a new architecture in year two, and an optimization of that architecture in year three. Call it TickTock-Tock, if you like, but as the third 14nm processor family, Kaby Lake is an optimization under the new regime.
LAKE-SIDE LAUNCH
Of course, given the timing of “Process, Architecture, Optimization,” it’s hard not to conclude that it’s a retrospective wheeze in response to troubles ramping up 14nm production. But the change of strategy helps to explain the slightly disjointed launch of the Kaby Lake family. Since August 2016, mobile PCs with Kaby Lake chips have been available, starting with the super-slim 9.8mm Acer Swift 7 Ultrabook. Today, there are plenty of Kaby Lake laptops, including the Dell XPS 13 refresh, with its stellar 22-hour battery life (which hints that Kaby Lake is pretty interesting), but no desktop chips.
Meanwhile, both Intel’s multi-core server chips, with up to 22 cores, and its closely related enthusiast-class Core i7 processors for the LGA2011 platform remain based on the older Broadwell architecture, a full two generations behind the latest laptop chips.
If that’s the background, what are the tech specs of the new architecture, and what CPU models can we expect when the desktop-orientated Kaby Lake-S variants arrive in early 2017? The second half of that question is straightforward. At the time of writing, Intel hasn’t officially announced any desktop models, but leaked documents have revealed the launch lineup in detail.
As seventh-generation Core processors, the new chips take the 7000 Series moniker. The top chip at launch will be the quad-core, eight-thread Core i7-7700K, clocked at 4.2GHz base clock. That’s 200MHz more than the existing Skylakebased Core i7-6700K. The quad-core, quadthread Core i5-7600K, meanwhile, jumps to 3.8GHz from the 3.5GHz of its 6600K predecessor. However, it isn’t known for sure what the new 7700K and 7600K will deliver in terms of Turbo clock speeds. As we’ll see, there are some reasons to think these new Kaby Lake puppies will be quite the thing for overclocking.
Across the 11 new Kaby Lake processors that will be available at launch, base clock speeds are up by 100–300MHz. Once we know what the official Turbo clocks are, of course, the next question will be overclocking. As we suggested, laptops such as the Dell XPS 13 and its borderline ridiculous battery life hint at significant progress when it comes to the process technology, despite Kaby Lake remaining
14nm. Could that play into overclocking headroom with the desktop S series?
It may make sense. That’s because Intel hasn’t carried over its 14nm production tech untouched for Kaby Lake. Instead, it’s been revised, and is now known as 14PLUS. Two major changes are involved. Taller fins for the FinFET gates (so-called for their fin-like 3D structure) mean less driving current is needed, and thus less leakage, and, in turn, power consumption and heat dissipation are less of a problem. The other feature is a wider gate pitch, which, at first glance, is an odd one.
Gate pitch is a critical metric when determining transistor density. So, wider gate pitch would, on the face of it, mean lower transistor density, which, in turn, could be regarded as a step backward. For now, the impact of this is unclear in terms of Intel’s die size and chip production costs. Intel has reportedly said that it won’t have an impact on die size. What we can say is that the intended benefit is that the wider pitch allows the heat generated by each transistor to dissipate more efficiently, rather than multiplying with that of nearby transistors.
Put the wider pitch together with the taller FinFETs and, in theory, you have access to greater voltage range and higher frequencies. Certainly, base clock speeds are up over Skylake chips. Will overclocking headroom benefit, too? We hope so. There’s at least a chance that Kaby Lake could be the best overclocking chip from Intel for several generations.
CLOCK-WATCHING
That’s the process tech dealt with. What about architectural tweaks? Intel has said little about Kaby Lake’s execution cores, other than to indicate that little has changed. You should not, therefore, expect much, if any, improvement in throughput when it comes to performance-per-clock. In other words, a 3.5GHz Kaby Lake processor will perform the same in most tasks as a 3.5GHz Skylake chip. Any improvements would need to come from higher clock speeds.
Intel hasn’t changed the architecture of its 3D cores in its integrated graphics engine, either. But there is one exception to this tale of mostly static per-clock performance: 2D video. There are three ways a processor can handle 2D video. It can brute-force it on the general-purpose CPU cores. It can use the integrated 3D graphics cores to accelerate some parts of the job. Or it can hand the job over to the purpose-built or fixed-function 2D video engine for full hardware acceleration.
In practice, it’s a little more complicated. Depending on the spec of a given CPU
and the video codec, there can be some overlap, with some parts of the workload done in software, others accelerated by the 3D cores, and yet others hardware accelerated. The big news for the Kaby Lake fixed-function media engine is that Intel has added some serious fullhardware support for the latest codecs, right up to 4K resolutions.
More specifically, there’s full hardware support for encode and decode of 4K HEVC Main10 profile video streams. Skylake chips can decode HEVC Main10 up to 4K at 30fps, but they use a hybrid process that shares the workload between the generalpurpose CPU cores and the GPU’s shader cores. The upshot of which is that Kaby Lake can decode 4K HEVC video streams at a fraction of the powerload of Skylake. In fact, the engine is so powerful, it can decode up to eight standard 4K HEVC streams simultaneously, and can support ultra-high bitrate streams, too, up to 4K at 60fps and 120Mbps, which is pretty preposterous when you think about it.
Intel has also given Kaby Lake proper fixed-function 8-bit encode and 8/10-bit decode support for Google’s VP9 codec. Again, Skylake only supported that in a hybrid mode. VP9 is Google’s preferred codec for high-quality video on YouTube, so full hardware support is a real boon.
THE 4K FUTURE
The really intriguing development when it comes to streaming video arguably lies elsewhere. As you probably know, Netflix doesn’t allow 4K streaming to PCs. That’s largely, if not exclusively, down to piracy concerns. However, Kaby Lake’s video engine is thought to have additional DRM features that may allow it to be certified as a platform for 4K Netflix streaming. On a related note, Intel has confirmed that Kaby Lake will be certified for Sony’s 4K movie and television streaming service sometime in 2017. It’s known as ULTRA, and has hitherto been restricted to Sony’s own TVs, which tends to confirm that Kaby Lake does indeed bring new DRM features to the mix.
A final related feature enabled by the new 2D video engine are some tweaks to enable HDR and wide color gamut in Rec.2020 format. Intel hasn’t gone into much detail, but increased per-channel color bit widths are surely part of the package. Intel has also indicated that Kaby Lake’s HDR support leverages both graphics execution units and the 2D video engine, so there is scope for hardware acceleration and assist to enable HDR visuals within a low-power profile.
That pretty much wraps up the video part of the Kaby Lake package,
so what about the minor matter of the CPU cores and general performance. Is there any good news? Apart from the aforementioned increase in clock speeds, the answer is no. Intel has introduced Speed Shift v2. That’s a feature that, via a software driver, allows the CPU itself to control its own Turbo mode. This latest version allows for quicker adjustments in CPU frequency. Kaby Lake can react in 10ms to Skylake’s 30ms. But this is a feature that’s most relevant to mobile PCs.
All of which leaves one final frontier: the new platform that comes with Kaby Lake. We speak, of course, of the new 200 Series motherboard chipsets. As we go to press, full specifications are still being sorted out, but a number of intriguing new features are likely. For starters, the native PCI Express support goes up from 20 PCI Express 3.0 lanes on 100 Series chipsets, such as the Z170, to 24 lanes. Access to PCI Express bandwidth is becoming ever more critical as storage technology increasingly switches from SATA interfaces to PCI Express.
Specific support for Intel’s upcoming Optane storage and memory technology is also in the mix. Exactly what form this will take remains to be seen. You could say the same thing for Optane, to be fair. A general affinity for high-bandwidth interfaces will feature, too. Expect support for up to 10 USB 3.0 sockets built into the PCH chip natively. Thunderbolt 3 support built natively into the platform is also expected, though this isn’t quite a total, nailed-down certainty.
Of course, these features will vary depending on the CPU and spec of the motherboard chipset. The 200 Series chipsets and Kaby Lake maintain the existing LGA1151 socket. Indeed, the 200 Series chipsets are backward compatible with sixth-generation CPUs—Skylake, in other words. Needless to say, if you use a Skylake chip in a 200 Series mobo, you’ll default back to the 20 PCI Express lanes built into the CPU, and native Thunderbolt 3 support will be a goner, too. Anyhow, the usual stratification of consumer chipsets will feature, with mainstream H250, H270, and Q270 chipsets capped by the more enthusiast-orientated Z270 chipset.
Kaby Lake certainly isn’t the most compelling CPU Intel has ever cooked up. It won’t blow away any records for pure performance. But when you factor in some of the new platform features, and the extra bandwidth available for storage and peripherals, along with the new 2D video engine, and the promise of DRM support for premium 4K streaming services, the likely upshot is that Kaby Lake will be pretty compelling for anyone building a new PC. That’s not something you’ve always been able to say about Intel’s CPU refreshes of late.