Event: Industrial Affiliate Program
Intel Performance Hybrid Architecture
Lecture Description:
General purpose compute PnP optimizations is becoming harder and complex as the process shrinks. Architectural techniques drive transistor count and process technology no longer mitigate all power problems. The world of SW has changed significantly, the market demands diverse computing. While single and lightly threaded application still dominant, there is a steady growth in heavily threaded applications for parallel data computation, content creation and AI. In addition, users today are multi-processing by running multiple applications at once, for interactive usages mixing foreground and background applications. The Multipurpose one size fits all approach is no longer applicable. This is a fundamental industry wide challenge that demands a paradigm shift.
Intel performance Hybrid is the future for Intel to mitigate these challenges and scale performance per watt per mm^2 at the same time on a single product. It required resolutions in the areas of SoC development such as power management, power delivery, integration, cores compatibility, software and performance and power tunning. The performance Hybrid architecture is delivering three types of efficiency, (1) Energy (Performance/Watt) (2) Architectural – seamlessly scaling core count and core mix (3) Cost effective – performance/mm^2. It introduces two types of cores, a performance Core (P-Core) and an Efficient Core (E-Core) on a single SoC, sharing resources such as Last Level Cache and power delivery. It introduces Intel Thread Director (ITD) technology which guides the Operating Systems to make best thread placement decisions based on the performance monitoring unit and Realtime ML workload predictions. With ITD, Intel ensured that thread placement decisions are driven by HW hints building Intel’s differentiation from competition. There was significant software enabling effort in Operating system (Windows11) to enable support for ITD materializing hybrid goodness