Workload- and process-variation aware voltage/frequency tuning for energy efficient performance sustainability of NTC manycores

Abstract

The power-wall raised by the stagnation of supply voltage in deep-submicron technology nodes, is now the major scaling barrier for moving towards the manycore era. At the same time, the adoption of manycore architectures is considered to be crucial for satisfying the increasing computational power demands and throughput requirements imposed by the explosion in software complexity and volume. The rise of the so-called Dark Silicon, caused by the power budget violations that allow only a small portion of the available computational resources to be simultaneously exploited, points to the direction of energy efficient platforms. Near-Threshold voltage Computing (NTC) has emerged as a promising approach to overcome the manycore power-wall, at the expense of higher sensitivity to process variation and reduced performance which can be compensated with massive parallelization. Given that several application domains operate over specific performance constraints, the performance sustainability is considered a major issue for the wide adoption of NTC. In this work, assuming a feasible, low overhead Power Delivery Network (PDN) for NTC, we investigate how performance guarantees can be ensured when moving towards NTC manycores through a variability-aware voltage and frequency allocation methodology, showing that performance can be efficiently sustained at the NT region while reducing energy dramatically. Additionally, we propose an algorithm for balancing throughput under process (and workload) variation that sustains performance while providing significant energy savings.