Abstract ▼ AI-Helper

Recently, mobile CPUs are trying to use many high-performance features (e.g., out-of-order execution, large cache size, etc.). Those features help improve performance of mobile devices though it increases power consumption of the mobile devices. Considering the mobile workloads which have different ...

Recently, mobile CPUs are trying to use many high-performance features (e.g., out-of-order execution, large cache size, etc.). Those features help improve performance of mobile devices though it increases power consumption of the mobile devices. Considering the mobile workloads which have different characteristics compared to the general SPEC benchmark workloads, it would merely increase power consumption of the mobile devices without any performance benefit.
This thesis analyzes mobile workloads in detail and reveal the contemporary mobile CPUs have quite more resources than those really required by the workloads. This thesis also proposes a simple, yet efficient design methodology by using weighted sum. By utilizing the simulation results with various CPU configurations, the proposed weighted sum-based methodology finds the optimal configurations for mobile workloads. In evaluation results, mobile workloads favor narrower issue widths and commit widths considering both power consumption and performance. In terms of cache sizes, mobile CPUs also favor smaller L1 data, L1 instruction, and L2 cache sizes. It implies mobile CPUs have many idle and unused resources when executing mobile workloads. The proposed methodology can be used in design space explorations when designing mobile CPUs.

Keyword