Intel Announces More Details of Kaby Lake

Intel officially announced Kaby Lake (7th generation Intel Core) at Intel Developer Forum (IDF) in San Francisco in August 2016, with mobile SKUs available in September 2016 and desktop processors launching in January 2017. The Kaby Lake announcement formalized Intel’s departure from the two-decade “tick-tock” strategy — alternating between process node shrinks (“tick”) and new microarchitecture designs (“tock”) every year — replacing it with a three-phase “Process-Architecture-Optimization” (PAO) model. Skylake (6th gen, 2015) was the architecture phase on 14nm; Kaby Lake was the optimization phase, also on 14nm (labeled “14nm+” internally, with improved fin geometry and higher drive current enabling higher clock frequencies at similar voltage). IPC (instructions per clock) was essentially unchanged from Skylake — the same Skylake microarchitecture, core design, L1/L2/L3 cache hierarchy, and ring bus topology — but clock speed improvements were meaningful: the desktop Core i7-7700K reached 4.2 GHz base / 4.5 GHz single-core turbo versus the Core i7-6700K’s 4.0 GHz / 4.2 GHz on the same 14nm process. The 14nm+ improvements that enabled higher clocks also reduced leakage current, contributing to slightly better power efficiency in mobile SKUs.

The media engine improvements were the most significant new capability in Kaby Lake relative to Skylake. Intel’s Quick Sync Video engine in Skylake could hardware-decode H.264 (AVC) but required CPU-assisted software decode for H.265/HEVC 4K content and VP9. Kaby Lake’s media engine added full fixed-function hardware decode for H.265 8-bit (standard HEVC used in H.265 Blu-ray and 4K streaming), H.265 10-bit (HDR content with 10-bit color depth, used in Blu-ray HDR and Netflix HDR streams), VP9 8-bit and 10-bit (Google’s codec used on YouTube for all 4K content and most 1080p60 streams), and 10-bit H.264 decode. The importance was practical: streaming services and YouTube were rapidly expanding 4K and HDR content in 2016, and CPU-based software decode of 4K H.265 or 4K VP9 required 30–50% CPU utilization even on a fast quad-core, making laptop playback on battery impractical. Hardware decode consumed roughly 1–2W of iGPU power versus 15–25W of CPU power for software decode, dramatically extending battery life for 4K video playback on Kaby Lake laptops. Kaby Lake also added hardware H.265 encoding support for 8-bit and 10-bit via Quick Sync, used for screencasting, streaming, and video editing workflows.

Intel’s 10nm process had been originally targeted for 2016 (the next “tick” after Skylake); the delay that made Kaby Lake’s optimization phase necessary stemmed from Intel’s unusually aggressive density targets for 10nm — Intel’s 10nm aimed for ~100 million transistors per mm² (comparable to TSMC’s 7nm at the time), while TSMC 7nm targeted ~90 million/mm² and Samsung 7nm ~90 million/mm². The higher density goals required refinements to EUV lithography and multi-patterning techniques that took longer than Intel’s development schedule had projected. Cannon Lake (originally the next “tick” on 10nm after Kaby Lake) was effectively cancelled after reliability issues; Ice Lake (August 2019) became the first consumer 10nm desktop product, three years behind schedule. Coffee Lake (8th gen, October 2017) added a third optimization pass of 14nm++, pushing mainstream quad-core Core i5 and hexa-core Core i7 desktop processors on the same process lineage that had served since Broadwell in 2014, demonstrating how far Intel could stretch a mature manufacturing node through continued refinement at the cost of TSMC and Samsung pulling ahead in process technology leadership.