Today, Intel announced its upcoming Kaby Lake hardware refresh. This launch is the first iteration of Intel’s new Process – Architecture – Optimization strategy (dubbed PAO) that replaced Tick-Tock earlier this year. It’s a change driven by the realities of lithography. As die shrinks have become more difficult, it now takes longer to move from one node to the next. This difficulty is somewhat exacerbated for Intel because it continues to perform full node shrinks rather than relying on hybrid proccess nodes like TSMC and Samsung.
Kaby Lake is built on what Intel is calling “14nm+” rather than its original 14nm. While the company isn’t revealing specifics on its design changes at this point in time, it has stated that its 14nm+ offers an improved fin profile, improved transistor strain, and a larger pitch. In semiconductors, pitch is defined as the center-to-center distance between two features. Historically, manufacturers wanted a smaller pitch, not a larger one — but packing chips into tighter and tighter spaces increases power density, which ultimately harms performance at this stage. Intel is claiming that these improvements can yield up to a 12% performance improvement and allows for higher clock speeds in the same power envelopes.
Anandtech has details on the specific SKUs, and the top-rated frequencies at the various power windows have all increased substantially. Intel’s Core i7-7Y75 has a 3.6GHz burst frequency compared to 3.1GHz for the Skylake-derived m7-6Y75. 15W chips are also seeing a large frequency jump, from a similar 3.1GHz top frequency for the i7-6500U to 3.5GHz for the Core i7-7500U. The overall TDP is higher because the CPU’s base frequency jumps — the 4.5W parts all have base frequencies of 1GHz – 1.3GHz, while the 15W chips are in the 2.5GHz – 2.7GHz range. This should ensure that users see at least some small benefits in all cases, but the difference between a 15W chip at 2.5GHz and a 15W chip at 2.7GHz is pretty tiny. Because manufacturers have the option to build chips that target a variety of clock speeds and cooling arrangements, how much time any system spends at its boost frequency will depend on the particulars of the cooler. We saw this create somewhat erratic performance with systems based on the first Intel Core M, and that problem doesn’t seem to have been eradicated.
Kaby Lake’s larger gains are in the realm of video and multimedia processing. Unlike Skylake, which couldn’t support H.265 / HEVC Main10 decoding fully in GPU hardware, Kaby Lake can — and Intel is claiming a huge performance increase and subsequent power consumption boost as well. VP9 8-bit and 10-bit decode and encode are also supported in hardware.
QuickSync and AVC encode also get a boost in this update, which should provide overall improved functionality — but how many people actually use these codecs or settings on a daily basis? It’s not a trivial question — the majority of online video is still encoded in H.264, and while it’s great to see Intel being proactive and adopting these technologies, it’s not going to make much of a difference for owners asking whether or not they should buy hardware to see a meaningful update now. Then again, Anandtech is reporting that 4K Netflix support, when it eventually comes to PCs, may only be available to Kaby Lake owners. Skylake customers may be out of luck due to a lack of support for the DRM that Netflix apparently requires.
As CPUs have evolved, the conditions for “winning” the race have changed as well. The ability to move between turbo and non-turbo clocks and in and out of sleep states has turned 0W into the new 1GHz — by which I mean that the ability to enter and exit low power states has a critical impact on battery life and system performance. With Skylake, Intel introduced Speed Shift — an option that allowed the CPU to handle its own power state shifting rather than relying on the operating system.
With Kaby Lake, Intel is introducing Speed Shift 2 and improving its ability to transition between power states.
Seventh generation Kaby Lake processors can boost to maximum frequency much more quickly and cut their frequencies when they hit turbo limits. In theory this should make systems faster and more responsive. Overall, Intel is predicting that Kaby Lake will boost performance by 10-19% over Skylake. Presumably most of this boost will be in mobile, where power limits and TDP are critically important. Desktop hardware isn’t expected until later this year.