I. Introduction
The ever-increasing amount of data generated nowadays from High-performance Computing, Internet-of-Things and 4G/5G applications is currently generating vast amounts of traffic within the Data Centers (DCs) [1] that need to embrace fundamental changes in their infrastructure to keep up with the ever-increasing computational needs at a cost-effective and energy-efficient manner. The energy restrictions in system processing performance advances can be outlined in the example of the HPC performance evolution during the past 10 years: Fig. 1 depicts the HPC performance predictions established in 2011 [2], which were building upon the respective progress and performance metrics of the top HPC machines witnessed until 2011 in order to forecast the Exaflop performance target by 2020, assuming that advances will continue to rely on the 10x performance improvement factor per 3.5–4 years. However, reality has been proven to be very different and the performance of HPC machines from 2011 until today has slowed down dramatically, improving by 10x every 6 years and reaching now the 200 Pflop/s target in world's No. 1 IBM Summit HPC machine at an energy envelope of already 10 MW. Inverting this performance slow-down can only be accomplished by increasing processing performance at a lower energy consumption framework, which in turn has to rely on the energy-efficient synergy of the most powerful high-performance multi-core engines.
HPC performance developments until June 2019 with the dashed lines depicting the performance evolution predictions made in 2011 [2].