Measuring Hyper-Threading and Turbo Boost

The text discusses an experiment conducted by Sung Park to measure the performance impact of Hyper-Threading (simultaneous multithreading) and Turbo Boost (Core performance boost) on AMD-based servers running a specific software stack. The results showed that enabling both features resulted in an 86% improvement in requests per second, while latencies were generally lowered by either or both technologies. In the production environment, with both features enabled, requests per second improved by 136%. However, the baseline was skewed due to imbalance in CPU utilization and only partially reproduced lab results. The experiment also added power data, concluding that AMD's implementation of simultaneous multithreading is power-efficient as it achieves additional requests per second while consuming less power compared to the skewed baseline.